Abstract:With the ongoing transformation and upgrading of the global agricultural industry, fruits and vegetables agricultural products origin handling technologies are facing dual challenges: improving operational efficiency and ensuring stringent quality control. Intelligent solutions powered by artificial intelligence, the Internet of Things, and big data are reshaping the entire technological framework of origin handling, enabling automation, precision, and data-driven decision-making. It systematically reviewed the key stages of fruits and vegetables origin handling, providing an in-depth analysis of how intelligent technologies that were applied across different stages. It highlighted recent advancements in intelligent grading and sorting, pre-cooling and preservation, packaging, and origin traceability, emphasizing their role in enhancing supply chain efficiency and improving quality management. Furthermore, it explored how breakthroughs in these technologies contribute to optimizing logistics, reducing post-harvest losses, and ensuring food safety. Looking ahead, it discussed emerging trends in fruits and vegetables origin handling, stressing that the deep integration of intelligence, digitalization, and sustainability would be a key driving force for industry innovation. The adoption of smart technologies not only enhanced operational efficiency but also promoted environmentally friendly and resource-efficient practices. The research result can provide a quantifiable framework for technology selection and an all-encompassing optimization path for the evolution of the fruits and vegetables smart supply chain. Ultimately, it supported the paradigm shift from traditional experience-based post-harvest handling to a more efficient, data-driven approach, offering valuable insights for researchers and industry practitioners alike.

Abstract:The agricultural industry has witnessed transformative advances with the integration of digital technologies into quality assessment and processing. Online perception technology, which encompasses an array of sensor modalities, imaging techniques, and intelligent data processing algorithms, has emerged as a pivotal tool in ensuring the quality, safety, and marketability of agricultural products. A comprehensive survey of recent research progress in the field was provided, focusing on the application of online perception systems for agricultural product quality assessment. The technological components underpinning these systems were critically analyzed, including machine vision, hyperspectral imaging, near-infrared spectroscopy, and advanced data analytics. The discussion covered the evolution of sensor hardware, algorithmic developments in machine learning and deep learning, and the integration of multi-modal data for real-time quality control. Specific application examples were presented across different categories, such as fruit and vegetable sorting, grain quality monitoring, and the evaluation of meat and dairy products. The future trends and the potential of emerging technologies were also outlined to further revolutionize the perception of agricultural quality perception technologies. In summary, it highlighted how online perception technologies not only enhanced the precision and efficiency of quality assessment but also contributed to sustainable practices by reducing wastage and ensuring compliance with safety standards.

Abstract:The widespread application of food safety traceability system can not only reduce the cost of information acquisition for regulatory authorities and improve regulatory efficiency, but also force enterprises to raise safety awareness and actively control risks. However, due to the technical and management reasons of different food regulatory departments, regulatory data in the food supply chain was scattered, there was a lack of trust between regulatory domains, and data sharing was difficult. In response to the above issues, a cross trust domain supervision data sharing method for the food supply chain was proposed by utilizing the decentralized and tamper proof advantages of blockchain technology in identity verification and permission management. Firstly, cross domain identity authentication was achieved by introducing the Schnorr signature algorithm. Secondly, an attribute mapping mechanism was added to attribute-based access control (ABAC) to achieve dynamic fine-grained access control of food supply chain traceability and supervision data. Finally, a prototype system for cross domain secure transmission of food supply chain was developed based on Hyperledger Fabric, and performance testing was conducted. The results showed that when the transaction load was 300, the minimum delay for policy writing in the cross domain supervision data sharing method was 0.56s, and the maximum throughput was 113 transactions per second. When the transaction load was 600, the minimum delay for policy decision-making was 0.01s, and the maximum throughput was 414 transactions per second. The cross domain regulatory data sharing method proposed had good performance, providing methods and ideas for achieving cross domain sharing of food supply chain regulatory data.

Abstract:Blockchain traceability plays a crucial role in ensuring the safety of agricultural products and coordinating production in each link. In response to the privacy data sharing and security protection requirements in the blockchain traceability of fruits and vegetables, a privacy data trusted sharing method based on content extraction signature was proposed. Firstly, the data of each link in the fruit and vegetable supply chain was analyzed to design the classification method of traceability data. Secondly, the content extraction signature (CES) technology was combined with the ciphertext-policy attribute-based encryption (CP-ABE) technology. The CES technology was used to hide the sensitive data in the traceability information, achieving flexible and secure sharing of traceability data. The CP-ABE technology was used to implement different access control policies for different roles to achieve data encryption and privacy protection. A blockchain based on Ethereum was built for simulation experiments. The system test results showed that in the diffusion test, the average change rate of ciphertext was 90.2%, and in the correlation test, the average change rate of ciphertext reached 75.7%, demonstrating excellent security performance. When the number of plaintext sub-messages was 10 and 20 respectively, the average verification time was 0.102ms and 0.159ms, which was 80.1% and 84.3% faster than that of the traditional scheme. The verification time did not increase linearly with the increase in the number of sub-messages. In terms of storage efficiency, the required storage space was reduced by 70.6% and 85.3% when the number of sub-messages was 5 and 10 respectively. This method not only ensured the privacy of data in the fruit and vegetable supply chain but also realized an efficient and trusted data sharing mechanism, providing a beneficial technical reference and practical basis for improving the level of agricultural product safety supervision.

Abstract:The security of agricultural product supply chains plays a critical role in national development and social stability. However, the inherently complex structure of these supply chains—characterized by multiple stages, diverse stakeholders, and heterogeneous data sources—poses significant challenges for risk information sharing, especially in balancing data privacy protection with accurate risk detection. In response, a novel cross-domain risk information detection and trustworthy sharing model was proposed by integrating blockchain and federated learning technologies. Specifically, a distributed federated learning-based interaction framework was established to enable secure and decentralized circulation of risk information across different supply chain entities. To enhance anomaly detection, a multi-level evaluation mechanism based on the isolation forest algorithm was introduced to identify abnormal data patterns at various stages of the supply chain. Additionally, a dynamic risk contribution and credit evaluation model was developed to incentivize stakeholders to continuously share high-value risk data, while assessing their trustworthiness and participation levels in real time. Extensive experiments validated the effectiveness of the proposed approach in improving the efficiency, accuracy, and reliability of cross-domain risk information sharing. This work can provide a scalable and privacy-preserving solution tailored for the agricultural supply chain, offering practical implications for intelligent risk governance and data-driven decision-making in agri-food systems.

Abstract:In order to strengthen the implied quality and safety risk management of aquatic products cold chain enterprises on internal control data and realize cost reduction and efficiency, taking the HACCP plan of raw oysters as an example, the quality and safety risk warning indicator system was constructed from the construction of the risk indicator system and combined with the multimodal characteristics of the risk warning monitoring data to construct the quality and safety fusion of group experts-domain empirical knowledge and deep learning algorithms. On the basis of designing quality and safety early warning data collection points and obtaining monitoring data based on the HACCP plan, the AHP method was used to obtain the utility value scheme of the experts-risk indicators, and the entropy weight method was used to optimize the utility value scheme of multi-expert group decision-making, and then the rating data was determined, which constituted the original dataset from the monitoring data as well as the optimized expert rating data. In order to ensure the sensitivity of the early warning, the qualified dataset and the complete dataset were adopted as the dataset, and the LSTM model with the ability to capture the complex relationships in the multidimensional data and the RBF model with the ability to deal with the complex classification boundaries were fused to construct the early warning model to carry out the simulation experiments, and to do the comparative analyses between the LSTM model, the RBF model, and the fused LSTM-RBF model on the different datasets. The experimental results showed that the fused LSTM-RBF model had 96% and 90% accuracies on both qualified and complete datasets, and the test results on qualified datasets were significantly better.

Abstract:In order to effectively break through the problem of single temperature monitoring interval on traditional time temperature indicator (TTI), realize the visualization monitoring of the pork quality under the environment of room temperature (10℃

Abstract:Aiming at the problem of unclear interactions between pre-cooling environmental factors and fruit quality in the post-harvest cold chain of perishable berries, the pre-cooling contribution rate evaluation indexes were proposed, and the impacts of delayed pre-cooling time (0h, 3h, 6h), ambient wind speed (0.1m/s, 0.5m/s, 1.0m/s, 1.5m/s), and ambient temperatures (0℃, 5℃, 10℃, 15℃) on the quality of the cold chain of blueberries were investigated, and a pre-cooled quality evaluation based on differential pressure pre-cooling was carried out. The quality of blueberries in the cold chain cycle was evaluated based on differential pressure pre-cooling. The results showed that immediate post-harvest pre-cooling could increase soluble solids content by 2.8 percentage points (pre-cooling contribution of 20.49%) and commercialisation rate by 6 percentage points (pre-cooling contribution of 6.74%), and the more timely pre-cooling was implemented, the more conducive it was to maintaining post-harvest quality; the improvement of ambient wind speed could enhance the effectiveness of pre-cooling, and the increase of pre-cooling wind speed up to 1.0m/s, the soluble solids content and commercialisation rate were increased by 1.75 percentage points and 4 percentage points, respectively (pre-cooling contribution of 15.15% and 4.34%); the pre-cooling treatment had a doublesided effect on the texture characteristics of fruits; in the pre-cooling stage, low ambient temperature and high wind speed would reduce the elasticity of fruits and increase the risk of transport damage, and in the storage and transport stage, the low ambient temperature and high wind speed were conducive to the maintenance of the elasticity and tightness of blueberries and improve the ability of the fruits to resist bumps. The optimal pre-cooling parameters were ambient temperature of 5℃ and ambient wind speed of 1.0m/s.

Abstract:In recent years, core agricultural enterprises within China’s agri-food supply chains have invested heavily in IT applications, yet there has been a problem of IT investment not being proportional to output. Therefore, from the perspective of IT capabilities,the theoretical and empirical research were combined to deeply investigate how IT capabilities influenced agri-food supply chain performance and the mediating roles of partnerships and information sharing. Totally 350 valid questionnaires were collected through online and offline surveys and structural equation modeling was used to find that: firstly, IT capabilities were positively correlated with both partnerships and information sharing; secondly, both partnerships and information sharing positively influenced the overall agri-food supply chain performance; thirdly, IT capabilities indirectly affected agri-food supply chain performance, with both partnerships and information sharing acting as complete mediators. It innovatively built and empirically tested the influence mechanism among IT capabilities, partnerships, information sharing, and agri-food supply chain performance. The findings can provide agricultural enterprise managers with effective strategies (such as enhancing IT capabilities, partnerships, and information sharing) to improve agri-food supply chain performance. Furthermore, it offerred valuable reference and insights for the continuous improvement of China’s agri-food industry chain development policies.

Abstract:Aiming to handle the problem of multi-unmanned vehicle task allocation in agricultural product transportation scenarios between production sites (farms) and distribution sites (markets), a novel combinatorial optimization model that incorporated delivery time requirements, vehicle constraints, and task complexity was established at first. Subsequently, an improved fractional order particle swarm optimization (IFOPSO) algorithm was proposed. By introducing a fractional-order Lévy random step size into the particle swarm optimization (PSO) algorithm, the global search capability was significantly enhanced. Additionally, a mechanism for adaptively adjusting the Lévy order was designed to improve the convergence accuracy, robustness, and overall optimization performance of IFOPSO. Experimental results based on ten benchmark functions demonstrated that the proposed IFOPSO algorithm exhibited significant advantages in terms of convergence speed, accuracy, and global search ability compared with existing algorithms. Furthermore, an optimization model for unmanned vehicle pickup and delivery task scheduling was developed, where the total cost accounted for travel cost, time violation cost, load violation cost, and start-up cost. The IFOPSO algorithm was applied to solve this task allocation problem, and comparative experiments with traditional PSO, improved PSO, and fractional order PSO algorithms showed that the proposed algorithm effectively reduced scheduling costs, improved solution efficiency, and rapidly identified a feasible and optimal pickup and delivery solution.

Abstract:Aiming at the common problems in the collection and transportation of urban food waste, such as low loading rate or overloading, high vehicle exhaust emissions, strong subjectivity in route planning, high comprehensive costs, and low merchant satisfaction, according to the distribution, collection and transportation characteristics of urban food waste, a model of the dynamic vehicle routing problem with time windows based on traffic flow was established, and an improved genetic algorithm was used to solve it. The static optimization results showed that the strategy of “minimum collection and transportation cost+time window” was identified as the best static optimization strategy. Compared with the scenario without a hard time window, when one more vehicle was used, unit average collection and transportation cost, unit average carbon emission, and unit average fuel consumption were reduced by 8.16%, 12.12%, and 10.48%, respectively. The dynamic optimization results showed that the strategy of “minimum collection and transportation cost+time window+time-discrete” was the best dynamic optimization strategy. In this strategy, compared with the best static optimization strategy, the total cost was reduced by 15.23%, the fuel consumption and carbon emissions were reduced by 24.97%, and unit average collection and transportation cost, unit average carbon emission, and unit average fuel consumption were decreased by 25.85%, 39.39% and 36.36%, respectively. In addition, after simulating the installation of intelligent garbage bins to obtain the real-time amount of food waste, it was verified that the addition of this equipment had a further optimization effect on the proposed model. Finally, an environmental impact assessment was carried out for the actual operation and the six optimization strategies.

Abstract:In recent years, there have been frequent issues with the safety of agricultural products. However, enterprises can ensure the quality and safety of agricultural products by adopting traceability technology. Based on the technology-organization-environment (TOE) research framework, a theoretical model of the influencing factors of enterprises’ adoption of agricultural product traceability technology was constructed. Technological factors included relative advantages (RA) and complexity (CX). Organizational factors included perceived benefits (PB), cost (Cost) and upper management support (UMS). Environmental factors included competitive pressure (CP) and supplier support (SR). The model explored the impact of these three types of factors on the adoption of traceability technology. The data was sourced from a survey questionnaire of 204 enterprises in China. The research results showed that organizational factors such as supplier support, upper management support, and cost had significant direct effects on the adoption of traceability technology by enterprises. However, the direct impacts of the relative advantages and complexity of technological factors, as well as the competitive pressure and perceived benefits of environmental factors, on the adoption of traceability technology by enterprises were not significant, and only indirect effects can be observed. The research proposed three suggestions for enterprises to adopt and governments to promote agricultural product traceability technology: strengthening the promotion to senior management personnel, reasonably controlling complexity and costs, and evaluating the conditions for adopting traceability technology.

Abstract:In view of the insufficient theoretical basis for the design of special rotary tiller machinery for hilly and mountainous areas, the lack of basic data on sloping land conditions, and the common technical bottlenecks such as significant fluctuations in tillage depth, intense vibration, and excessive energy consumption when operating in complex terrain, a special rotary tiller test platform suitable for sloping land conditions in hilly and mountainous areas was designed, which can realize functions such as slope angle simulation and slope profiling operation. The test platform was mainly composed of the rotary tillage driving guidance device, rotary tillage driving device, rotary tillage lifting adjustment device, rotary tillage angle simulation device, rotary tillage operation device and slope angle simulation device, etc. It integrated functions such as slope angle simulation, rotary tillage angle coordination, rotary tillage operation and tillage depth adjustment. The performance test of the test platform was carried out. The results showed that the rotary tillage inclination angle simulation device can achieve precise coordination with the target slope (soil trough inclination angle) within the range of 0°to 20°. The electric sports car of the rotary tillage driving device can achieve a forward driving speed of 0~3.64km/h. The rotary tillage knife shaft can achieve a rotary tillage speed of 0~335r/min. The rotary tillage lifting adjustment device can achieve a stable and stepless adjustment of the tillage depth from 0~30cm. This test platform met the multi-factor and multi-level test requirements of rotary tillage on hilly and mountainous slopes. The design goals were achieved, including random simulation of the slope angle, automatic profiling of the rotary tillage knife shaft, precise and controllable tillage depth, and stepless adjustment of the forward speed. This research can provide platform support for the improvement of the theory of rotary tillage on sloping land and the innovative design of special rotary tillage tools, and offer methodological references for the development of test platforms for other operation equipment in hilly and mountainous areas.

Abstract:A maize sowing position prediction system based on spatial temporal coupling was designed for maize sowing process. The system integrated an opposed infrared photoelectric sensor, a GNSS-RTK high-precision positioning module and a data transmission unit to predict the spatial position of seed landing by real-time monitoring of seed falling signals and combining the sowing machine heading, speed and spatial-temporal lag compensation model, and data would eventually be uploaded to the cloud. The system adopted STM32F103 microcontroller as the central controller, and a segmented spatial position conversion model was constructed to solve the offset problem between the main antenna of the planter and the infrared sensors; the spatial temporal hysteresis compensation model was introduced, and the seed falling delay, the localization information transmission delay, and the program execution delay were measured to be 107.7ms, 50ms, and 39.5ms, respectively, and finally the position deviation of the planter’s forward direction was corrected. The final shape of the coupled prediction model was clarified by formulating the directional response rules for the positive and negative values of latitude and longitude deviations in different intervals. The results of the field test showed that the average deviation of the seed landing position predicted by the system from the actual position was 36.86mm, with a standard deviation of 3.57mm and a coefficient of variation of 9.69%, which verified the effectiveness of the model. The system was capable of real-time recording and cloud storage of seeding position data, providing a reference for the subsequent precise and collaborative management of mid-tillage and fertilization.

Abstract:Aiming to address the autonomous operation requirements of robotic platforms under controlled traffic farming, a universally applicable full production cycle path planning method for polygonal agricultural fields was proposed. The method constructed a dual-layer structure comprising a primary permanent road network and a suboperation path layer. Paths in the steering reserved area were planned through equidistant scaling and vertex smoothing, while the row direction in the central operation area was determined by minimizing the projection length perpendicular to the travel direction. An interval-shuttle traversal sequence was designed for the primary road network, and an adjacent-shuttle sequence was adopted for the sub-operation path layer. Dubins curves were employed to design transitional paths, while potential elastic entry/exit points were incorporated to address path connectivity challenges caused by operational elasticity interruptions. The sequential quadratic programming algorithm was employed to generate paths satisfying kinematic constraints, eliminating curvature discontinuity defects inherent in traditional linear-arc path planning. Field experiments on robotic platforms demonstrated that, for convex/concave polygonal fields, the planned paths achieved 77.21% operational path length proportion in primary road path length proportion, 56.87% operational path proportion in sub-operation layer, 90.65% operational coverage rate for both layers, maximum curvature change rate was not greater than 0.04m-2, jerk limit was not greater than 0.05m-3, and total soil compaction area ratio was 8.83%, successfully confining all production cycle paths to permanent fixed roads and fulfilling CTF requirements for robotic operation platforms.

Abstract:Aiming at the problems that most studies required the individual measurement of single soil parameters and there were few studies on the response of multiple parameters to the penetration resistance (PR), taking the clay loam as the research object and a study on the influence of soil moisture content and compaction pressure on the PR was conducted. Six moisture contents and three compaction pressures were set, and the resistance change curves during the process of the probe penetrating the soil under various conditions were recorded. A full-factor experiment was carried out with the final peak (FP) of the probe penetration curve, the relative fluctuation bandwidth (RFBW), and the Fourier energy of the probe penetration resistance curve (PRFE) as the indicators. The results showed that for the ascending effect, when the soil moisture content was increased from 0 to 10%, with the increase of the moisture content, FP was continuously increased, and reached the maximum value of 2.492N at a moisture content of 10% (an increase of 230.94% compared with 0); the compaction pressure had a greater impact on FP at low moisture content, and the impact weakened at high moisture content. For the fluctuation characteristics, when the moisture content was between 0 and 15%, both RFBW and PRFE were decreased with the increase of the moisture content. At a moisture content of 15%, the decrease rate of RFBW was 68.05% of that at a moisture content of 0, and PRFE dropped to 85.80% of that at a moisture content of 0; with the increase of the compaction pressure, both RFBW and PRFE were decreased. When the compaction pressure was 75kPa, the RFBW dropped to 74.62% of that when the compaction pressure was 15kPa, and PRFE dropped to 87.37% of that when the compaction pressure was 15kPa. The research results revealed the influence law of soil moisture content and compaction pressure on the probe PR, providing a basis for the development of soil property testing devices.

Abstract:In view of the problem that the grafting machine uses plastic clips to fix vulnerable seedlings and the plastic clips are difficult to recycle and pollute the planting environment, a grafting device based on UV glue coating and fixation was designed, which can automatically complete the operations of seedling clamping, synchronous cutting, butt glue spraying, seedling solidification and flexible seedling lowering. Firstly, the structural composition and working principle of the grafting device were expounded, and the key mechanisms and operating parameters such as clamping, cutting, handling, glue spraying, curing and seedling setting were designed; Secondly, in order to determine the optimal glue grafting parameters and improve the grafting success rate, taking white-seeded pumpkin (rootstock) and watermelon (scion) seedlings as the object, the atomization pressure, glue supply pressure and glue spraying distance were selected as the influencing factors, and the grafting success rate as the test index, the three-factor and three-level response surface test was carried out, and the significant ranking of the grafting success rate was as follows: atomization pressure, glue supply pressure, and dispensing distance; Finally, the Design-Expert 13 software was used to analyze the variance, respond surface and parameter optimization of the test results, and verify the test results to obtain the best parameter combination conditions. The test results showed that under the conditions of atomization pressure of 0.22MPa, glue supply pressure of 0.35MPa and spraying distance of 4.33mm, the average grafting success rate was 97.34%, the error from the theoretical value was 1.15 percentage points, and the average production efficiency was 576 plants/h, which met the quality requirements of mechanical grafting operation. The research results can provide a theoretical basis and technical support for the design and development of glue grafting robots.

Abstract:In addressing the challenge of coordinating multiple machines during autonomous maize grain harvesting and transportation under a one-sided unloading configuration, a collaborative scheduling strategy was introduced based on a dynamic unloading-threshold-optimized Markov decision process (MDOP). By continuously adjusting the harvester’s unloading threshold in real time, the proposed approach enabled seamless interaction between harvesters and grain transport vehicles, thereby ensuring that the harvester maintained uninterrupted operational efficiency. This dynamic adjustment mechanism significantly reduced the harvester’s nonproductive waiting time and curtailed both the transportation cost incurred by the vehicles and the grain loss that occurred during transfer. To evaluate its performance, the MDOP-based collaborative strategy was benchmarked against two alternative models: the conventional full-bin-triggered unloading protocol and a genetic-algorithm-optimized unloading strategy. Under identical field conditions, the MDOP approach achieved an 18.1%, 4.9% reduction in total operational time compared with the conventional approach, while transportation costs were lowered by 8.9%, 19.3%. Moreover, the grain transfer loss rate under the MDOP regime was measured at approximately 4.3%, underscoring its ability to mitigate kernel spillage more effectively than competing methods. These results confirmed the superior efficacy and robustness of the MDOP-based scheduling strategy in multimachine cooperative tasks. By optimizing unloading thresholds dynamically, it not only preserved the continuous harvesting pace of the combines but also minimized idle intervals and logistical overhead. Consequently, this research laid a theoretical and practical foundation for realizing fully autonomous, multi-machine cooperative operations in maize harvesting, thereby furnishing critical technological support for the development of unmanned maize farming systems.

Abstract:Aiming to address the issue of low targeting accuracy in controlling the unloading arm position during rice transfer from unmanned rice harvesters to grain transport vehicles, which relies on Beidou positioning information of the harvester and transport vehicle, a GTSM network for visual segmentation of grain compartment images was proposed to provide positional reference information for the unloading arm. Based on the DeepLabv3+ architecture, the lightweight ShuffleNetv2 backbone replaced Xception, and the atrous convolutions in the ASPP module were replaced with depthwise separable convolutions, followed by low-rank decomposition into micro-factorized convolutions to reduce model complexity and improve inference speed. Additionally, an SE channel attention mechanism was introduced in the shallow feature branch to enhance the model’s ability to utilize low-level features such as grain compartment edges and textures. Experimental results showed that GTSM achieved a mean intersection over union (mIoU) of 96.06% and a mean pixel accuracy (mPA) of 98.69%, representing improvements of 0.78 and 0.67 percentage points, respectively, over the baseline DeepLabv3+. Meanwhile, model complexity was significantly reduced, with parameter count and memory usage reduced to 1/9 of the original, and inference speed was increased by 166%. These results demonstrated that the proposed GTSM balanced segmentation accuracy and inference speed, providing a reference for automated grain compartment segmentation in field grain transport vehicles.

Abstract:Aiming to addressing challenges of low accuracy and poor adaptability of navigation line extraction algorithms in complex agricultural environments for maize-soybean intercropping scenarios, an improved YOLO v8-based method for extracting crop row navigation lines was proposed to enhance autonomous mobile platform navigation precision during field operations. For the specialized task of segmenting maize and soybean crop rows, a StarNet-YOLO backbone network was constructed by integrating the StarNet network with YOLO v8 and optimizing the detection head. The network was enhanced through strategies, including a custom-designed ASPPFE module, depth-separable convolution, and CSE structure optimization, while also implementing lightweight design by using the LAMP pruning algorithm. Additionally, the Douglas-Peucker algorithm was introduced to approximate crop row contours, and a scoring mechanism was developed to determine the midpoints of contour start and end segments, enabling precise fitting of crop row navigation lines. Ablation experiments showed that ASPPFE achieved an mean average precision for instance segmentation at 0.5 IoU (mAP50seg) of 99.5%, with its mAP across IoU thresholds 0.5~0.95 (mAP50-95seg) improved by 1.0, 1.0, and 0.4 percentage points compared with that of SPPELAN, SPPF, and ASPPF, respectively. After 25% pruning optimization, the StarNet-YOLO network’s mAP50-95seg was decreased by only 0.02 percentage points, while inference speed was increased from 390f/s to 563f/s, and floating-point operations were reduced from 7.2×109 to 4.7×109. Comparative testing on the same dataset showed that StarNet-YOLO’s mAP50-95seg outperformed YOLO v5, YOLO v7, and baseline YOLO v8 by 5.5, 4.8, and 2.8 percentage points, respectively. Validation of crop row navigation line fitting revealed average angular and distance errors of 2.01° and 23.17 pixels. This navigation line extraction algorithm demonstrated excellent performance in complex agricultural environments, balancing detection speed and accuracy, and provided a technical approach for visual navigation of autonomous robots operating in maize, soybean, and other crop fields.

Abstract:Aiming to achieve rapid and non-destructive detection of Fusarium head blight (FHB) severity levels, the hyperspectral imaging technology combined with machine learning models for analysis was employed. A total of 1660 wheat ear samples with varying degrees of infection were obtained by inoculating Fusarium fungi into the middle grains of wheat ears. Hyperspectral information of the samples was collected by using a hyperspectral imaging system, with the entire wheat ear designated as the region of interest (ROI) to extract average spectral data. By comparing the classification accuracy of four preprocessing methods—normalization, standard normal variate (SNV), multiplicative scatter correction (MSC), and Savitzky-Golay (SG) smoothing derivatives—the SNV algorithm was selected as the optimal preprocessing method. Subsequent analyses were conducted on the SNV-processed spectral data. Feature wavelength selection was performed by using the successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS), while dimensionality reduction was implemented via uniform manifold approximation and projection (UMAP) and linear discriminant analysis (LDA). After comparing these algorithms, LDA was ultimately chosen for its ability to reduce data to three dimensions while maintaining classification accuracy (96.05% for the test set and 94.71% for the training set) and low computational complexity (0.09s processing time). It was revealed that the critical spectral range for LDA-based FHB severity discrimination lay between 540nm (chlorophyll reflection peak) and 650nm (red light absorption valley), attributed to the synergistic effects of rapid chlorophyll degradation and structural tissue damage as infection progresses. A lightweight support vector machine (SVM) model integrating SNV and LDA was developed as the optimal framework for FHB severity classification. The results demonstrated that the proposed algorithm achieved high accuracy with excellent generalization capability, enabling efficient FHB severity assessment. The research result can lay a foundation for future large-scale, real-time field detection of FHB.

Abstract:Rice pests critically threaten rice cultivation by inflicting direct physiological damage, spreading diseases, and potentially causing catastrophic field extinction, leading to significant agricultural losses. To address challenges such as dense pest clusters, subtle morphological variations, and frequent small-target missed detections in pest detection lamp images, an intelligent recognition method was proposed by using an enhanced YOLO v8-STSF model. Key innovations included integrating a Swin Transformer module to boost backbone network multiscale feature extraction, optimizing neck network feature fusion via distribution shift convolution (DSConv), and adopting the Focal EIoU loss function to enhance small-target localization. Validated on a 7000 image multi-species pest dataset, the improved model achieved 95.45% of precision, 90.45% of recall, and 90.03% of F1-score, surpassing the original YOLO v8 by 2.13, 0.33, and 3.09 percentage points, respectively, while operating at 32f/s for real-time PC-based monitoring. A dual-platform system (Web and Android mobile) demonstrated field performance with 1.38s average response time, 96.34% of accuracy, and 3.86% of missed detection rate. This system can provide an efficient solution for precision pest control and advance intelligent agricultural monitoring.

Abstract:When dairy cows suffer from mastitis, the temperature difference between their udder surface and eye surface is relatively large. Therefore, the temperature difference between the udder and eye can be used as an indicator for mastitis detection. To address the issues of low accuracy, false detections, and missed detections in existing thermal infrared image-based mastitis detection methods, an improved YOLO v11n-seg method for mastitis detection in dairy cows was proposed, which utilized both thermal infrared and visible light registered images. To achieve more precise segmentation of the cow’s udder and eyes under limited computational resources, targeted improvements were made. Firstly, the ADown convolution module was used to replace some of the ordinary convolution layers in the baseline model (YOLO v11n-seg) for efficient feature extraction, which was beneficial for model deployment and usage in resource-constrained environments. Secondly, the MLCA attention mechanism was introduced at the end of the backbone network, significantly enhancing the feature extraction capability for smallscale objects. Finally, the RepGFPN structure was adopted in the neck network to optimize feature fusion and information transmission, further improving segmentation accuracy. The improved YOLO v11n-seg model achieved an average segmentation accuracy of 90.3% for cow eyes and 97.9% for udders. Compared with the baseline model, the improved YOLO v11n-seg model increased the average segmentation accuracy by 7.1 percentage points and 0.7 percentage points, respectively, while reducing the number of model parameters by 14.3% and the computational cost by 12.5%. The temperature difference between the udder and eye, extracted from the segmentation mask and temperature matrix, was compared with the set temperature difference threshold and verified by the somatic cell count method. The results showed that the accuracy of mastitis detection in dairy cows reached 88.46%. This proved that the proposed method can achieve udder and eye segmentation in dairy cows and can be applied to mastitis detection.

Abstract:In modern pig farming, the detection and counting of piglets presents a significant challenge due to their small size, high mobility, tendency to be occluded, and habit of clustering. Manual counting is not only time-consuming and labor-intensive but also prone to errors. To address these issues, a novel piglet detection algorithm, GSD-YOLO, was proposed based on YOLO v8n, which was designed to provide fast and accurate detection. The algorithm tackled the inherent difficulty of detecting piglets by introducing flexible nonmaximum suppression (FNMS) for bounding box intersection handling and employing the Inner-MPDIoU loss function to optimize the bounding box regression mechanism. Additionally, a coordinate attention (CA) mechanism was integrated to enhance the feature representation of target areas, effectively resolving issues related to long-range dependencies. For efficient deployment on embedded devices, the model was further optimized for lightweight performance. Specifically, the backbone and neck components of the original model were replaced with the GhostNet module, which reduced both model parameters and feature redundancy in the channels. Furthermore, a lightweight detection head, Detect_DG, was introduced, which reduced the overall model size by 18.48% while simultaneously improving detection accuracy. Compared with YOLO v8n, GSD-YOLO achieved improvements of 1.0 and 0.6 percentage points in F1-score and average precision (AP), respectively, while reducing the model size by 61.28% and improving the frame rate by 12.5%. In comprehensive detection performance, GSD-YOLO outperformed four widely used models, including YOLO v5. Experimental results demonstrated that the proposed model can rapidly and accurately detect piglets under challenging conditions, such as occlusion, overlap, and varying lighting environments. Moreover, with a compact memory footprint of only 2.6 MB. The deployment of GSD-YOLO on edge computing devices such as the Jetson Orin NX and Android mobile devices provided effective technological support for piglet counting and detection in practical applications.

Abstract:Aiming to address the challenges of low recognition accuracy, high labor intensity in manual identification, and minimal behavioral differences between selective feeding and normal feeding in dairy cows under complex environmental conditions, a method for identifying selective feeding behavior was proposed based on inspection robots and an improved RT-DETR model. An inspection robot was designed according to dairy cows-feeding characteristics to collect feeding process data. Data collection was conducted in three barns during three time periods (noon, afternoon, and night), ultimately establishing a dataset containing 10280 feeding behavior images across these periods. The RT-DETR model was enhanced by integrating a DBRA structure, which combined the DAttention (DAT) module and Bi-Level Routing Attention (BRA) module into the shallow layers, creating a novel image feature extraction architecture to improve the deep fusion capability of local and global features. Additionally, the Efficient Multi-Scale Attention (EMA) module was incorporated into the model encoder to strengthen high-level semantic information extraction and contextual correlation. Experimental results demonstrated that the improved model achieved a mean average precision (mAP@0.5) of 99.1% on the dairy cow feeding video dataset, with a model memory occupancy of 39.6MB and floating-point operations (FLOPs) of 4.67×1010. Compared with the original model, the mAP@0.5 was increased by 7.4 percentage points, memory occupancy was reduced by 0.9MB, and FLOPs was decreased by 2%. When compared with Sparse R-CNN, YOLO v7-L, YOLO v8n, DINO, Swin Transformer, and DETR models, the proposed model exhibited mAP@50 improvements of 8.5, 9.8, 7.8, 6.6, 11.4 and 9.5 percentage points, respectively. The findings enabled accurate differentiation between normal feeding and selective feeding behaviors, providing technical support for intelligent livestock farming.

Abstract:Portable near-infrared (NIR) spectroscopic detectors for soil total nitrogen content offer the advantages of rapid analysis, non-destructive measurement, and high efficiency. However, most existing instruments adopt filterbased designs with a limited number of spectral channels, which can lead to the loss of critical information and prevent the implementation of deep learning-based prediction models. With the commercialization of miniature spectrometers, a high-precision soil total nitrogen content detector was developed based on continuous spectral data. The detector primarily consisted of an NIR-R210 miniature spectrometer, a Raspberry Pi, a touchscreen display, and a portable power supply. The spectrometer was used to acquire soil spectral reflectance data, which were processed by a deep learning model embedded in the Raspberry Pi to predict soil total nitrogen content. The prediction results were then displayed in real time on the touchscreen. A total of 600 soil samples were collected from the Shangzhuang Experimental Station of China Agricultural University. The predictive performances of three models (partial least squares regression (PLSR), gated recurrent unit (GRU), and Transformer) were compared. Among them, the Transformer model based on full-spectrum data achieved the best performance, with a coefficient of determination (R2) of 0.89, a root mean square error (RMSE) of 0.19g/kg, and a residual predictive deviation (RPD) of 2.96. Further real-time in-situ field tests showed that the Transformer model maintained high accuracy under field conditions, with an R2 of up to 0.83. This portable device provided an effective solution for real-time soil nutrient detection and precision management in smart agriculture.

Abstract:Soil total nitrogen is an important nutrient index of soil. The soil of pear orchard of Jiangsu Academy of Agricultural Sciences was taken as the research object, the soil spectral reflectance data were obtained by hyperspectral imaging technology, the shuffled frog leaping algorithm and competitive adaptive reweighted sampling in total nitrogen content in orchards was studied and constructed based on hyperspectral data, which provided a method for accurately detecting soil total nitrogen content. The competitive adaptive reweighted sampling were introduced for spectral feature extraction, and the partial least squares regression, support vector regression, random forest and convolutional neural network models were used to estimate the total nitrogen content of the soil by using the full band and the characteristic band, respectively. The results showed that after the original spectrum was processed by a variety of preprocessing methods, it was found that the total nitrogen prediction model constructed by SG convolution smoothing combined with standard normal transform pretreatment had the best performance. Based on the shuffled frog leaping algorithm, totally ten key bands were extracted, accounting for 4.08% of the total number of bands, which effectively reduced the data dimension. The convolutional neural network model constructed based on the shuffled frog leaping algorithm to extract feature bands performed well, and the coefficient of determination of the model test set was 0.95, the root mean square error was 0.21g/kg, and the relative analysis error was 3.97. The results showed that the shuffled frog leaping-algorithm can efficiently extract the feature bands, reduce the data dimension, and improve the estimation accuracy of soil total nitrogen content, which provided a reference for the accurate estimation of soil total nitrogen content in orchards.

Abstract:Aiming at the problems of inaccurate traction load recognition in electric tractor plowing and cultivating operation and the dependence of the training process on massive labeled data, a training framework based on semi-supervised learning algorithm for fusion of multiple working condition parameters in electric tractor plowing and cultivating operation was proposed, and a model for electric tractor traction load grade recognition based on wide residual network and extended long and short-term memory network (WideResNet-xLSTM, WRNx) was constructed. Among them, the semi-supervised learning framework used two kinds of data, labeled and unlabeled, for the iterative training of the discriminative model, and applied C-means fuzzy clustering to analyze the linear output of the model;based on the WRNx combinatorial model, the effective features of the load data were deeply extracted through the feature expression capability of WideResNet, and the temporal relationship was processed through the xLSTM network, and finally, the load sequence was realized by the classifier for classification prediction. A multi-sensor-load parameter testing system for electric tractor plowing and tilling units was constructed and field tests for plowing and tilling operations were carried out. The results indicated that the semi-supervised learning framework proposed can reduce the training sample requirement of labeled data by 25.4%, which was better than the traditional supervised learning training framework, and the accuracy and F1-score of the model constructed for recognizing the hauling class of electric tractor plowing and cultivating operation were 94.35% and 94.27%, respectively. The research result can provide a solution for semi-supervised learning to recognize the load of electric tractor plowing operation.

Abstract:The gas-liquid twin-fluid nozzle capitalizes on the substantial velocity difference between the liquid and gas phases to generate high shear stress at the interface, effectively breaking and tearing the continuous liquid phase to produce finely atomized droplets. This process offers numerous advantages, including high atomization efficiency, low energy consumption, a broad range of dosage adjustments, precise control of droplet size distribution, resistance to clogging, strong adaptability to various agents, and high operational efficiency. Consequently, the gas-liquid twin-fluid nozzle has become a crucial component in precision spraying equipment. Firstly, the working principles, classification, and comparative performance of the gas-liquid two-phase nozzle against other types were detailed. Then a comprehensive summary of the fundamental atomization theories, nozzle structural design, simulation methods, experimental research techniques, and performance evaluation metrics was provided. Additionally, the research progress in agricultural pest control applications was discussed. It highlighted open research questions, such as refining the physical model of the atomization process, enhancing atomization performance under low pressure and energy conditions, and determining the benefit boundaries of droplet size. In conclusion, it summarized the performance advantages, challenges, and future issues to be addressed in the application of gas-liquid two-phase nozzles for precision spraying.

Abstract:Manual sowing of Sinonovacula constricta seedling in tidal flat is a slow and labour-intensive operation, and damage of mechanical sowing limits the production of Sinonovacula constricta. The mechanical properties of 2-month-old Sinonovacula constricta seedling were tested and a pneumatic seeding device was proposed which mainly consisted of a seedling box, a grooved pulley, seedling-blowing pipe, and an air flow distribution pipe. Sinonovacula constricta seedling in the seedling box entered the seedlingblowing pipe via the grooved pulley, and then with the effect of air flow, they were blown and scattered onto the tidal flat. The kinematic model of seedling in the seedling-blowing pipe and air were established respectively. The contact forces of the collision process were obtained and the main factors influencing the movement trajectories of Sinonovacula constricta seedling were analyzed. Based on CFD-DEM method, single-factor and three-factor and three-level quadratic orthogonal central composite experiments were conducted respectively, with the inlet velocity of air flow, the distance from the ground, and the speed of the device selected as experimental factors, and the relative error of seeding density and the damage rate as performance indicators. The results showed that the factors affecting the relative error of seeding density, in order, were the speed of the device, the inlet velocity of air flow, and distance from the ground. The factors affecting the damage rate were the inlet velocity of air flow, the distance from the ground, and the speed of the device in order. The optimal combination of parameters was determined to be 30m/s of the inlet velocity, 490mm of the distance from the ground, and 0.17m/s of the speed. Based on the parameters obtained in simulation experiments, a prototype pneumatic seeder was fabricated and tested for its performance. The seeder worked satisfactorily at the relative error of seeding density of 7.02% and the damage rate of 3.92%. The research result can provide a technical support for the seeding devices of thin-shelled shellfish seedling on tidal flats.

Abstract:Aiming at the problems such as the uneven ground and the shaking of the fuselage in the current ridging and film-covering machinery operation, the positioning of the machine is inaccurate, the straightness deviation of the ridge body is large, and the stability is poor. An autonomous navigation control method based on Express long range system (ExpressLRS) was proposed. A light and simplified electric crawler-type autonomous navigation ridging and film-covering machine was designed to realize the functions of ridging, film-covering, soilcovering, and improve the straightness of the operation. The key parameters of the ridging mechanism were designed and calculated, and the Kalman filter algorithm was used to build an autonomous navigation control system based on ExpressLRS. Using WitMotion’s uniaxial angle measuring sensor HWT101CT for prototype production, the test results showed that the yaw angle fluctuation range of the prototype was between -2.5°and 3.5°, the mean deviation of the lateral position was within 3cm, which met the design requirements and proves that the system was feasible and effective. The best navigation effect was at speed of 0.1m/s to 0.2m/s. The field test of prototype showed that the ridge height of 170mm had the best effect. Among them, the qualified rate of the ridge height was 86.67%, the qualified rate of the ridge top width was 93.33%, and the qualified rate of the ridge bottom width was 86.67%. All the coefficients of variation were less than 6%, indicating that the uniformity of the ridge body was high. Ridge straightness of 2.21cm, ridge top flatness of 1.22cm indicated that the autonomous navigation straight line driving function qualified, the evaluation indicators met the design requirements of ridge laminating machine.

Abstract:In view of the problem that the existing vegetable transplanting mechanisms cannot precisely achieve the required transplanting trajectory and posture for the integrated operation of transplanting and harvesting, a hybrid six-bar single-degree-of-freedom vegetable pot seedling transplanting mechanism was proposed based on non-circular gear constraints. According to the requirements of integrated transplanting and harvesting, an ideal transplanting trajectory integrating the static trajectory of taking seedlings and the dynamic trajectory of planting seedlings was determined as an “eagle-beak-shaped” static trajectory and an “approximate straight-line-shaped” dynamic trajectory. The kinematic model of the transplanting mechanism was established. The non-circular gear transmission function was constructed by using quintic B-spline interpolation method. Considering the agronomic requirements of vegetable transplanting, the spectral clustering balanced differential evolution algorithm (SCEDE) was used to optimize the design of the transplanting mechanism with the minimum motion error and the optimal gear curve of the non-circular gear as the optimization objectives. The optimal mechanism parameters for the integrated transplanting trajectory and posture were obtained. The structure design, simulation analysis and bench test of the transplanting mechanism were carried out. The results showed that the trajectory and posture of the physical prototype test and the virtual prototype simulation were basically consistent with the theoretical trajectory. When the operating speed of the transplanting mechanism was 25~45r/min, the success rate of taking seedlings was 96.1%, the success rate of planting was 91.4%, and the coefficient of variation of transplanting plant spacing was 2.31%. It can meet the requirements of vegetable transplanting operation, which verified the correctness of the proposed theoretical method and the feasibility of the mechanism.

Abstract:In order to solve the problem of low efficiency caused by stoppage unloading during field operation of bagging potato combine, a non-stoppage unloading device was designed. The device can switch between caching and bagging status, and achieve non-stop unloading during the machine operation process. The expression of the maximum contact stress during the collision between potato and potato box was derived by using Hertz contact theory, and the key factors affecting the contact stress were obtained. An analysis was conducted on the unloading process of the unloading device, and the size parameters of the unloading device were determined. Three possible motion directions of the potato after collision with the wall of the box were analyzed by the velocity vector diagram. Based on the principle of kinematics, the trajectory and velocity expression of potato in the falling process were defined, and the key factors affecting the potato speed were obtained. In order to determine the best structural parameters of the box, Box-Benhnken test method was used to carry out three-factor and threelevel tests on the device with wall angle, conveyor belt speed and box wall length as test factors and potato peel breakage rate and potato damage rate as test indexes. Analysis of variance was performed on the test results through Design-Expert 11.1.0 software to identify the factors that had a significant impact on the experimental indicators, and the rule of influence of experimental interaction factors on test indicators was analyzed through response surface tests, and the optimal structural parameters were obtained. On this basis, experimental verification was carried out, and field comparison experiments were conducted. The bench test showed that when the wall angle of the potato box was 44°, the sorting and conveying chain speed was 0.44m/s, and the wall length of the potato box was 603mm, the breaking rate was 0.96% and the damaged rate was 0.63%. The results of field comparative experiment showed that the obtained data met the operational requirements of the potato combine harvester and the harvesting efficiency was improved by 41.63% compared with that of shutdown unloading combine harvester.

Abstract:Aiming to address the common issues of low conveying success rate, high miss holding rate, high stem breakage rate, and poor reliability in the process of clamping and conveying garlic plants by garlic combine harvesters, a kind of variable stiffness flexible clamping conveying device for garlic was designed. The overall structure and working principle of the clamping conveying device were elaborated and analyzed. Taking different harvest periods (early stage, middle stage, late stage) of Lanling 46 cloves garlic, Jinxiang hybrid garlic, and Jinqi early maturing garlic varieties as research objects, physical property tests of garlic stems under different moisture contents were carried out. Through theoretical calculation and mechanical analysis, key components were optimized;the working process of clamping conveying was analyzed to determine the key factors affecting its performance. An EDEM-MFBD coupling simulation model of the clamping conveying device was established and simulated to explore the influence of different operating parameters on the stress of garlic stems. Taking the speed ratio of clamping conveying speed to forward speed, feeding angle, and elastic coefficient of floating wheel as test factors, and the conveying success rate, miss holding rate, and stem breakage rate as test indexes, a quadratic regression orthogonal rotation combination test was conducted. Regression models of each test index with test factors were established, and the optimal operating parameters of the clamping conveying device were determined. The test results showed that when the speed ratio of clamping conveying to forward speed was 2.31, the feeding angle was 6.2°, and the elastic coefficient of the floating wheel was 3.95N/mm, the working performance was the best. At this time, the conveying success rate was 97.21%, the miss holding rate was 1.06%, and the stem breakage rate was 1.73%. In order to verify the working performance of the optimized clamping conveying device, a bench test was carried out, and the test results were basically consistent with the predicted results of the regression model.

Abstract:Aiming at the contour-planting characteristics of sugarcane in hilly areas and the frequent occurrence of rollover accidents when tracked sugarcane harvesters operate on slopes of 2°~10°, a 1∶4 scale test platform was constructed according to similarity principles. Based on microcontroller and sensor control detection technologies, a lateral leveling system suitable for tracked sugarcane harvesters was designed. Considering the harvester’s narrow width and high center of gravity, singleside (and double-side) leveling control strategies were proposed. The system utilized hydraulic cylinders to drive synchronous extension and retraction on one side, enabling adaptive body leveling and improving the anti-rollover capacity of the sugarcane harvester. Experimental results showed that the system can efficiently level the vehicle under different load conditions, and the body leveling function was reliable. In static tests, the platform achieved lateral leveling in 1.22s with a tilt angle error within ±1°, meeting the requirements for hilly regions. Lateral critical rollover tests indicated that the critical rollover angle was increased from 24.32° (before leveling) to 29.20° (after leveling), with a maximum lateral leveling angle of 10°, effectively enhancing the machine’s resistance to rolling over. Finally, through real-time spectral analysis, changes in frequency distribution and vibration amplitude were detected, and spectral waveform diagrams under different conditions were analyzed. This laid a research foundation for modeling the dynamic behavior of harvesters in complex environments and provided a basis for rollover warning signals.

Abstract:Aiming at the problems of serious slippage, poor uniformity and poor crop adaptability of the existing flatbed straight blade chopping tool, based on the functional and structural characteristics of beaver’s lower incisors, the skull model of beaver was extracted and the characteristic curves of the lower incisors was fitted by adopting the three-dimensional laser scanning and computer-assisted processing of the inverse engineering technology, and then a kind of biomimetic silage forage chopping blade was designed according to the model. On the basis of constructing the bionic coupling structural characterization model of beaver lower incisor, the movement and force relationship between the clamp angle, wedge angle and slip angle of the lower incisor and the biological cutting behavior were determined through theoretical analysis, and the models of slanting cut and stable slip cut of the blade were established to derive the structural characteristics of the beaver’s lower incisor in the process of stabilizing the cutting process, and the mechanism of response to the resistance and slip reduction, and the model of the cutting force of the invasive material was established for the theoretical analysis of the cutting head force characteristics, and the model of the cutter head was used for the theoretical analysis of the cutter head force characteristics, and the model of the cutting force was used for the theoretical analysis of the cutter head. The theoretical analysis of the cutter head force characteristics was carried out, and the influence law of wedge angle and cutter head force and strength was obtained. The functional structure of beaver lower incisors was characterized in segments, and the blade shape was designed accordingly. In order to determine the optimal structural parameters of the blade shape, a four-factor, five-level orthogonal test was conducted to investigate the effects of the key parameters of the blade shape on the chopping effect of silage forage, using the standard grass length rate, slant stubble rate and broken pitch rate as the test indexes. The results of the bench test showed that when the clamp angle was 46°, the inner slip angle was 31°, the outer slip angle was 67° and the wedge angle was 28°, the standard grass length rate was 9584%, the slant stubble rate was 2.84%, and the broken knot rate was 95.46%. Under the optimal structural parameters determined in the bench test, the field harvesting validation test was carried out on ryegrass, whole silage corn and sweet sorghum respectively with the standard grass length rate, slant stubble rate and broken node rate as the test indexes, and the test results showed that when the operating parameters were 6km/h and the feeding rate was 8kg/s, the average standard grass length rate of ryegrass was 91.95%, the average slant stubble rate was 3.31%, and the average breaking rate was 96.11%;whole plant silage corn average standard grass length rate was 95.62%, the average slash stubble rate was 3.81%, the average breaking rate was 96.32%;sweet sorghum average standard grass length rate was 92.60%, the average slash stubble rate was 4.06%, the average breaking rate was 96.03%, the performance indexes were consistent with the national industry standards. The bionic chopping curved blade can cut smoothly and cut sections evenly, which had strong crop adaptability.

Abstract:In view of the problems that the existing wind selection residual film impurity separation device separates the residual film with soil, while the water-washing residual film impurity separation device consumes lots of energy and water, a countercurrent hook-type residual film mixture cleaning and separation device was designed. Through the design and dynamic analysis of the main working components, the operating conditions of the hook film tooth chain, the spiral conveyor speed range, the impurity removal motor speed range, and the water pump model required for the water circulation system were determined. A response surface test was conducted with the hook residual film tooth chain speed, the average flow rate and the water outlet height as test factors, and the residual film separation rate and the residual film cleaning rate as test indicators. A variance analysis was conducted on the test results, and it was concluded that the influence of each test factor on the residual film separation rate from large to small was the water outlet height,the average flow rate,the hook residual film tooth chain speed;the influence on the residual film cleanliness rate from large to small was the average flow rate,the hook residual film tooth chain speed,the water outlet height. The parameters of the test indicators were optimized, and the optimal parameter combination was obtained: the hook residual film tooth chain speed was 0.196m/s, the average flow rate was 1.51m/s, and the water outlet height was 457.0mm. At this time, the residual film separation rate and the residual film cleaning rate were 89.95% and 93.46%. The optimization results were experimentally verified, and the residual film separation rate and residual film cleanliness rate were obtained as 88.72% and 92.35%, respectively.

Abstract:In response to the demand for the resource utilization of residual films in Xinjiang and the challenges of shredding and separating existing film residue mixture, a combined shredding and separation device featuring a rolling knife and axial flow drum was designed. This device can facilitate the feeding and shredding of mixed residual films, as well as the separation of residual films and the crushing of cotton stalks. Utilizing dynamics and kinematics, the processes of feeding, shredding, and separation were analyzed, revealing the main factors and parameter ranges that affected the efficiency of shredding and separation. The experimental factors included the feeding roller speed, shredder speed, and separation drum speed, while the evaluation indicators were the residual film content in the residual films, the residual film content in impurities, and the qualified rate of cotton stalk crushing length. Single-factor and Box-Behnken response surface experiments were conducted by using Design-Expert software, followed by variance and response surface analyses to clarify the impact patterns of experimental factors and their interactions on the experimental indicators. Through multi-objective optimization of the established second-order polynomial response surface model, the optimal operational parameters were determined: feeding roller speed of 20.94r/min, shredder speed of 335.78r/min, and separation drum speed of 282.38r/min. Experimental validation with these optimized parameters yielded residual film content of 91.26% in the residual films, 7.22% in impurities, and qualified rate of 93.78% for cotton stalk crushing length, thereby meeting the operational requirements for shredding and separating film-mixed residues.

Abstract:In order to study the influence of flow instability under off-design operating conditions, based on the SST k-ω turbulence model, the generation and evolution mechanism of unsteady flow and the pressure pulsation characteristics of the vaneless space (VS) were studied under four offdesign operating conditions with three guide vane openings (GVO). The results showed that under the exceptionally small GVO (G11), the disturbance from the backflow of the tailwater pipe in the outlet area of the impeller was significant. With the increase of the GVO, the influence of this disturbance was gradually decreased. The phenomenon of blade vortex and flow separation also showed a decreasing trend with the increase of the GVO. The high turbulent kinetic energy (TKE) zone was mainly distributed in the upper reaches of the runner blade channel and the VS, the turbulent kinetic energy distribution zone was large under the turbine under low flow and runaway conditions, and the TKE zone was mainly distributed in the VS and the runner blade outlet under braking conditions. At small blade opening angles, vortices were generated under all four operating conditions, and the vortex core distribution was dense. As the blade opening angle was increased, the distribution of vortex cores was significantly weakened. Under the high flow condition, vortices mainly appeared in the form of central vortex bands, and as the random group flow rate was decreased, the distribution of vortex cores was gradually strengthened on the pipe wall side. Under abnormally small GVO, the low-frequency component (LFC) and high frequency-low amplitude component (HF-LAC) of the pressure pulsation in the VS were extremely minimal. Under the offdesign condition with large opening, the BPF component and LFC jointly became the dominant frequency of the pressure pulsation in the VS. Under abnormally small GVO, the flow structures in each flow domain had significant specificity compared with that under the large GVO, significantly impacting the pressure pulsation in the VS.

Abstract:Soil salinization is one of the major constraints to agricultural production, and accurate monitoring of soil salinization is particularly important. The ground-based measured salinity data and unmanned aerial vehicle (UAV) data collected during April 8-12, 2023, in four experimental areas of Hetao Irrigation District were used to construct partial least squares regression (PLSR), random forest (RF), backpropagation neural network (BPNN) and support vector machine regression (SVR) inversion models for soil salt content (SSC). The experimental results obtained from the inversion of the optimal model were analyzed. The soil salt distribution maps of the experimental area obtained from the inversion of the optimal model were resampled to be 1m, 5m, and 10m by using the Nearest, Bilinear, and Cubic methods, respectively, and the average values of the corresponding image elements of the Sentinel-2A satellites in the same period were calculated as the salt content of the inversion model constructed by the satellites, and then compared with the optimal model at all scales. The optimal model at each scale was analyzed and the soil salinity distribution map of Hetao Irrigation Area was drawn. The results showed that the correlation of the Bilinear method at three scales was slightly better than the other two resampling methods. The accuracy ranking of the models constructed at five scales, from the highest to the lowest, was 0.07m, 1m, 5m, 10m, and the original SSC (OSSC), the best model determination coefficient R2 of the training set and validation set at the optimal scale of 0.07m was 0.24 and 0.30 higher than that of OSSC, respectively, and the root mean square error (RMSE) was 0.06 percentage points and 0.19 percentage points lower. The research explored the promotion effect of multi-scale soil salinity on the accuracy of soil salinity model inversion by satellite multispectral remote sensing platform, which provided an effective theoretical basis for multi-source remote sensing large-scale accurate soil salinity inversion.

Abstract:Accurate information on tea plantation distribution provides scientific support for land use planning and optimization of planting layouts, contributing to the sustainable development of the tea industry. Multimodal remote sensing features of tea plantation were constructed based on RGB bands from GF-2 PMS imagery, NDVI calculated from Sentinel-2 optical imagery, phenological characteristics derived from Sentinel-1 time-series SAR data, including growth amplitude, GA, and growth length, GL), and slope aspect, slope gradient, and curvature calculated from GF-7 stereo imagery. The optimal feature combination was selected through a random forest feature selection algorithm. A dual-branch network model, multi-modal information parallel branch network (MIPBNet), was built by using a multinetwork joint learning strategy, with attentional multiscale lightweight encoder-decoder network (AMLNet) as the first branch and Vanilla AMLNet as the second branch. A feature fusion module (dual-branch feature fusion block, DBFF) was utilized for feature-level fusion at the end of the decoder, and a composite loss function was employed for optimization training. The research findings were as follows: the combination of NDVI, GA, slope aspect, and slope gradient best improved classification accuracy and was identified as the optimal multi-modal feature set. When RGB data was sequentially augmented with NDVI, GA, slope aspect, and slope gradient, experiments showed a significant reduction in both omitted and falsely extracted tea plantation areas, with an improvement in overall accuracy (OA) of 3.11%. Compared with typical semantic segmentation models such as UNet, UNeXt, and Segformer, the single-branch AMLNet within MIPBNet achieved superior tea plantation extraction results.

Abstract:The soil tillage layer is the foundation of crop growth and development. Accurately monitoring the moisture content of the soil tillage layer and providing precise irrigation to crops can improve crop yield and water resource utilization efficiency. To achieve efficient monitoring of soil tillage layer moisture content, a soil tillage layer moisture content inversion method based on collaborative Kriging interpolation was proposed. Firstly, Sentinel-1 satellite data, which can obtain soil surface information, was used as the data source. Combined with the reliable XGBoost (extreme gradient boosting) model, it can efficiently invert soil surface moisture content. Using the large-scale soil moisture inversion results as covariates and 115 measured soil tillage layer moisture contents as main variables, the synergistic relationship between soil surface and tillage layer variables was utilized to interpolate the soil tillage layer moisture content using the collaborative Kriging method. The collaborative Kriging method can effectively utilize the synergistic relationship between soil surface and tillage layer variables to improve interpolation accuracy, and to some extent solve the problem of insufficient measured data on soil tillage layer moisture content. Comparing the Kriging interpolation of soil tillage layer moisture content with the linear fitting of surface and tillage layer moisture content, the results showed that using collaborative Kriging interpolation to invert soil tillage layer moisture content can significantly improve prediction accuracy. The coefficient of determination R2 was increased by 0.25 and 0.20, the root mean square error (RMSE) was decreased by 0.029cm3/cm3 and 0.014cm3/cm3, and the average absolute error (MAE) was decreased by 0.028cm3/cm3 and 0.015cm3/cm3, respectively. The accuracy was significantly improved.

Abstract:Wheat stripe rust is one of the major diseases affecting wheat yield, and improving the remote sensing monitoring accuracy of wheat stripe rust is of great significance for disease prevention and control. Based on the inversion of full-band SIF spectra by using the F-SFM algorithm, shape characteristics were extracted, and the response characteristics of full-band SIF spectra and their shape characteristics under stripe rust stress were analyzed. A remote sensing monitoring model for wheat stripe rust was constructed by using the random forest algorithm, and compared with a single-band SIF model. The results showed that under stripe rust stress, both leaf and canopy-scale SIF spectral curves and their shape characteristics exhibited different response characteristics. At the leaf level, as the severity of wheat stripe rust increased, the peak value of far-red SIF (CFR), skewness of far-red SIF (SFR), and area of emission peak of far-red SIF (AFR) was decreased, while the peak wavelengths of red (λR) and far-red (λFR) SIF, the kurtosis of far-red SIF (KFR) was increased. At the canopy level, CFR, λR, λFR, and AFR were decreased with the severity of wheat stripe rust. Additionally, the remote sensing monitoring model for wheat stripe rust constructed with shape features such as AFR, λFR, full-band SIF kurtosis (K), skewness of redband SIF spectra (SR), and λR as independent variables showed higher accuracy compared with the model with red-band SIF peak value (CR) and CFR as independent variables, with an increase of 27.59% in R2 and a decrease of 19.83% in RMSE in the training set, and an increase of 96.43% in R2 and a decrease of 17.01% in RMSE in the testing set. The shape characteristics extracted using full-band SIF can more comprehensively and accurately reflect disease stress information.

Abstract:Vegetable heavy metal pollution monitoring is an essential component of precision agriculture. In order to explore the hyperspectral response of vegetable under different concentrations of heavy metal stress, hyperspectral imaging (HSI) data of cabbage leaves under ten concentrations of Cu2+ stress was collected, and feature wavelength selection and classification modeling was conducted. A convolutional long shortterm memory neural network (CNN-LSTM) model for cabbage leaf Cu2+ stress classification based on hyperspectral data was proposed. Ten cabbage samples with different concentration gradients of copper stress were set, with four pots of samples at each concentration. Hyperspectral data collection was carried out when the cabbage grew to 15~20 leaves. Ten leaf samples were collected for each cabbage. A total of 400 hyperspectral data were collected. Firstly, spectral data preprocessing was performed by using S-G smoothing and first-order differentiation. Then, competitive adaptive reweighted sampling (CARS) and uninformative variable elimination (UVE) were used to extract ten common feature bands. The experimental results indicated that using the common wavelengths extracted by the UVE and CARS methods as input for the CNN-LSTM model achieved a test set accuracy of 94.8%, precision of 93.1%, and recall of 93.5%. These values were higher than those achieved by the SVM, CNN, and LSTM models by 8.7, 5.7, and 6.4 percentage points in accuracy, 6.6, 4.7, and 5.9 percentage points in precision, and 10.1, 5.2, and 3.9 percentage points in recall, respectively. The results were verified by accurate measurement of heavy metal content in cabbage leaves using an ICP-700T inductively coupled plasma emission spectrometer. For non-destructive classification monitoring of cabbage leaves under copper stress, the CNN-LSTM classification model with UVE-CARS feature bands selection performed the best, providing a method for non-destructive detection of vegetable heavy metals.

Abstract:Crop diseases can seriously restrict crop yield and quality. Traditional disease monitoring methods are inefficient and susceptible to subjective factors. Hyperspectral remote sensing technology has shown great potential in crop disease monitoring due to its high spectral resolution and objective authenticity. Ground hyperspectral and field disease index (DI) data of winter wheat with multiple growth stages were used, the spectral data were preprocessed using correlation analysis (CA) and successful projection algorithm (SPA) respectively, and the sensitive bands of wheat stripe rust through competitive adaptive reweighted sampling (CARS) algorithm that can construct optimal parameters were optimized. Finally, partial least squares regression (PLSR), back propagation neural network (BPNN) and extreme learning machine (ELM) were used to establish the disease index model based on the characteristic spectrum, and the modeling effects of different modeling methods were compared to realize the monitoring of wheat stripe rust. The research results indicated that the sensitive characteristic bands of wheat stripe rust in different growth stages were mainly concentrated in the near infrared and shortwave infrared bands, with 842nm, 850nm, and 858nm in the flag leaf stage and 947nm, 953nm, 1275nm, 1277nm, 1590nm, 1663nm, and 1665nm in the filling stage. In the comparison of different modeling methods, PLSR model performed best, and the model met the needs of early monitoring of wheat diseases and pests, and showed more obvious characteristics in the middle of the disease. During the flag leaf stage and filling stages, the PLSR models constructed with SPA-CARS-MCX and CA-CARS-MSC respectively had the best prediction performance. The R2 of the validation sets were 0.782 and 0.861, the RMSE were 0.022 and 0.094, and the RPD were 2.140 and 2.687, respectively. The algorithm constructed can provide ideas for monitoring wheat stripe rust at different growth stages.

Abstract:In greenhouse environments, fast, high-precision, and low-cost acquisition of scene depth information is crucial for agricultural machine vision systems in tasks such as tomato phenotype analysis, autonomous harvesting and multimodal joint segmentation. An attention-embedded RGB-to-depth conversion network (RGB to depth conversion network，RDCN) for monocular depth estimation was proposed, addressing issues in traditional algorithms such as insufficient feature extraction capability of encoders, low depth estimation accuracy, and blurred boundaries. Firstly, ResNext101 was employed to replace the original ResNet101 backbone network, extracting feature maps from different levels and integrating them into the Laplacian pyramid branches. This approach emphasized the scale differences of features and enhances the depth and breadth of feature fusion. To enhance the models capacity for capturing global information and contextual interactions, a shuffle attention module (SAM) was introduced. This module also helped minimize the loss of local detail information caused by the down-sampling process. This module also mitigated the loss of local detail information caused by the downsampling process. Secondly, to address the issue of blurred boundaries in the predicted depth maps, a depth refinement module (DRM) was embedded to capture depth variations near object edges in the predicted feature maps. For the study, an RGBD image acquisition platform for tomatoes was constructed in a daylight greenhouse environment using an Azure Kinect DK depth camera. To ensure diversity in the dataset, images were collected at different times of the day based on varying light intensities in the greenhouse environment. The training set was then augmented by using three methods: horizontal mirroring, random rotation, and color jittering, resulting in a total of 8515 aligned RGBD image sets of tomatoes. Experimental results indicated that by introducing the shuffle attention module and the depth refinement module, the model achieved accurate depth information prediction in greenhouse scenes. Compared with the baseline model, the visualized depth maps generated by the network demonstrated global completeness and clarity, with more texture details, especially in regions with complex geometries and significant depth variations, exhibiting superior visual effects. Experimental results showed that, compared with the baseline model, RDCN reduced the mean relative error, root mean square error, log root mean square error, and log mean error on the test set by 20.5%, 10.3%, 8.3%, and 21.8%, respectively. Additionally, accuracy under the 1.25, 1.252, and 1.253 thresholds was improved by 3.2%, 1.2%, and 1.0%, respectively. Moreover, the depth images generated by the network were visually complete and clear, with abundant texture details. Studies showed that RDCN can obtain highquality depth information from RGB data in greenhouse environments, providing technical support for agricultural machine navigation in greenhouse scenarios using monocular sensors, as well as for the application of depth images in multi-modal tasks.

Abstract:Aiming to address the challenges of low accuracy and inefficiency in phenotypic data acquisition, a high-precision, low-cost method for lettuce phenotypic parameter extraction in controlled environments was developed. An unmanned ground vehicle (UGV) equipped with autonomous navigation, multi-modal sensing, and multi-view imaging was deployed for automated in-situ data collection. A phenotype analysis pipeline incorporating a random under-sampling algorithm was designed to enhance point cloud processing efficiency. Image segmentation and clustering algorithms were implemented to extract multi-dimensional features, including plant height, maximum width, vegetation indices, and texture indices. Pearson correlation analysis between these imaging features and fresh biomass measurements identified four key variables highly correlated with biomass prediction. Single-feature and multi-feature biomass estimation models were constructed by using a back propagation algorithm. Results showed that the phenotyping pipeline designed took an average of 0.41s to process a single frame of data from a 5000 downsampled point cloud. The estimation models for lettuce height and maximum width achieved R2 values of 0.79 and 0.77, with mean absolute percentage errors (MAPE) of 4.94% and 5.02%, respectively. Compared with other biomass estimation models, the hybrid model incorporating four feature variables (HWVD) showed optimal performance, achieving an R2 of 0.82, RMSE of 4.03g, and MAPE of 6.04%. This method can provide a rapid, accurate, and non-destructive solution for field-based phenotyping and serve as a robust framework for investigating additional phenotypic traits.

Abstract:Aiming to address the challenge of accurately identifying similar features of peanut leaf diseases in complex environments, an improved detection algorithm, YOLO-ADM, was proposed based on the YOLO v8n model. Firstly, the ADown module replaced part of the CBS module, reducing information loss during down sampling and decreasing the model’s parameter count. Secondly, a deformable attention mechanism was integrated into the C2f module to form the C2f-DA structure, which replaced the C2f module in the upper layer of the SPPF. This modification enabled the model to focus on critical regions of peanut leaf diseases and effectively capture their distinguishing features. Finally, a novel multi-scale feature fusion network, termed MFI Neck, was designed to replace the original YOLO v8n neck network, enhancing the model’s capacity for multi-scale feature fusion. Experimental results showed that the improved YOLO-ADM algorithm achieved precision of 92.3%, recall rate of 91.0%, mean average precision (mAP@0.5) of 95.6%, and mean average precision (mAP@0.5:0.95) of 85.2%. Compared with the original YOLO v8n model, these metrics were increased by 4.5, 0.2, 1.6, and 3.0 percentage points, respectively. This approach enhanced detection performance while maintaining model efficiency, effectively meeting the identification requirements of peanut leaf diseases in complex environments, and provided a reliable reference for the detection and monitoring of leaf diseases.

Abstract:Aiming to solve the problem of low efficiency and strong subjectivity in human visual identification of diseased tomato plants in natural environments, a tomato yellow leaf curl virus disease grading detection model based on improved YOLO v7 was proposed to detect mild, moderate, and severe diseased plants. The model perception of complex lesion areas was enhanced by introducing DCN modules into the backbone network. Furthermore, some ordinary convolutions were replaced by Pconv modules in the main network to more efficiently extract spatial features and reduce redundant calculations and memory access. A SimSPPF module was introduced into the detection head to reduce the amount of computation, increase the receptive field, and enhance feature extraction capabilities. After testing, the average precision of the improved YOLO v7 model for mildly infected, moderately infected and severely infected tomato plants was 97.5%, 92.1% and 93.6%, respectively. The improved YOLO v7 model showed a mean average precision of 95.0%, which was 0.8 percentage points higher than that of the original model. The number of parameters was reduced by 8.2×105, the number of floatingpoint operations was reduced by 2.7×1010, and the model memory usage was reduced by 15.7MB. The size of the model was reduced while ensuring detection accuracy. Compared with Faster R-CNN, YOLOX, YOLO v5l, and YOLO v8m models, the mean average precision was increased by 11.2, 5.7, 1.4, and 8.7 percentage points, respectively. The experimental results demonstrated that the model can achieve graded detection and identification of tomato yellow leaf curl virus disease, providing support for the intelligentization of tomato cultivation.

Abstract:Completing a series of commercialization processing steps such as cleaning, waxing, and grading of citrus on the same production line is conducive to reducing fruit damage and improving fruit quality. However, the presence of diseased and damaged citrus can easily cause cross-contamination and pollution to the subsequent production line. Therefore, it is necessary to remove diseased and damaged citrus at the feeding section of the production line. For this reason, a disease-damaged citrus preliminary screening system was developed based on deep learning and Delta robots. Firstly, through comparative experiments of different detection models, the YOLO v7 model with the highest accuracy was selected, and combined with the DeepSORT tracking algorithm to achieve rapid and precise tracking and detection of citrus on the production line. Secondly, an optimized Delta robot door-shaped trajectory was proposed, and the precise control strategy of the stepping motor was calculated based on the interpolation method. Finally, a prototype of a screening device with fast positioning and grasping capabilities was built and integrated into the production line. The experimental results showed that the F1-score of the YOLO v7 model was 90%, which was 2 and 4 percentage points higher than that of the YOLO v5 and SSD networks, respectively. The Delta robot designed had high positioning accuracy, with an average positioning error of 1.5mm for the same point, which met the precision requirements for grasping. The average success rate of screening out diseased and damaged oranges could reach 83.25%. Therefore, the equipment proposed had excellent automatic screening capabilities on the citrus sorting production line, which can effectively reduce the cross-contamination of diseased and damaged citrus and pollution to the production line, thus ensuring the normal operation of the citrus production line.

Abstract:Tomatoes are a globally important economic crop with a wide planting area. In order to ensure tomato food safety and improve the economic benefits of tomatoes, accurate surface defect detection and quality grading of tomatoes are necessary. However, traditional tomato defect detection mainly relies on manual sorting during harvesting, which results in low efficiency and high missed detection rate. What’s more, new defects generated during procurement and transportation (such as dents, cracks, etc.) would be ignored. Therefore, an improved YOLO v11 method (Tomato defect detection YOLO, TDD-YOLO) was proposed for surface defect detection of tomatoes automatically, including white spot, hyperplasia, depression, crack, and deterioration. Firstly, a HE-Head layer of the YOLO 11 was constructed by fusing the wavelet depth separable convolution module to detect small targets, such as white spot, while maintaining its lightweight design. Secondly, the WC3k2 module was used to replace the original C3k2 module of YOLOv 11 to expand the receptive field of the model in the feature extraction stage, and a lightweight dynamic upsampling method was used to replace the original upsampling. These two improvements of YOLO v11 were to reduce the number of parameters and improve the realtime performance. Finally, an adaptive threshold focus loss function was used to improve the model’s attention to various classification label samples in response to the diversity and complexity of tomato surface defect types and distributions. Several experiments were carried out to evaluate the performance of the proposed method. The experimental results showed that the proposed TDD-YOLO method effectively improved the detection accuracy while keeping the model parameters basically unchanged. The overall recognition precision was 89.0% and the recall rate was 84.9%, which was increased by 2.9 percentage points and 5.6 percentage points comparing with that of YOLO v11, respectively. In comparative experiments, the proposed model had better detection performance than all published YOLO series models, Faster R-CNN and EfficientDet on detecting tomato surface defects. The proposed method achieved a detection speed of 142.89f/s, meeting the real-time detection speed requirements of industrial production applications. This work can provide important technical support for standardized and industrialized tomato detection and inspection.

Abstract:Fecal defecation is a primary source of organic waste in intensive aquaculture systems. The increase in the amount of defecation and the extension of time will accelerate the accumulation of pollutants such as ammonia nitrogen and nitrite in the aquaculture water. Therefore, monitoring fish defecation behavior is essential for maintaining optimal water conditions and ensuring sustainable fish production. In order to solve the problem that traditional defecation behavior analysis is time-consuming and labor-intensive, a high-performance, lightweight fish defecation behavior recognition model CDW-YOLO v7 was proposed based on the innovative enhancement of the YOLO v7-tiny. In the proposed model, a bidirectional feature pyramid network (C2f-BiFPN) was applied to optimize feature extraction within the neck network, a DyHead target detection head with an attention mechanism was utilized to accurately detect fish defecation behavior and strengthen relevant features, and the WIoU loss function was incorporated to improve the accuracy of the model’s outputs. Experimental results indicated that the performance of the CDW-YOLO v7 model was much better than that of the baseline YOLO v7-tiny model because reducing the number of parameters loading models by 2.56×106 and giga floating-point operations per second (GFLOPs) by 5.90×109, while increasing mean average precision (mAP) by 2.04 percentage points. Additionally, the proposed model surpassed three classic object detection algorithms (YOLO v3-tiny, YOLO v4-tiny, and YOLO v5s) when evaluating criteria such as model size, accuracy, and detection speed. The research result can provide a theoretical foundation for subsequent detection of fish health and establishing a quantitative relationship between fish behavior and water quality.

Abstract:Perching behavior is an important habitual activity for laying hens. To achieve rapid and accurate identification of individual behaviors of hens in a free-range environment with perches, a three-dimensional free-range hen perching behavior recognition method that integrated machine learning with ultra-high frequency radio frequency identification technology was proposed. The data on signal strength indicator values from the collected leg band tags of hens were processed through Gaussian filtering and maximum value weighted filtering. Using the support vector machine algorithm, these signal strength indicator values were classified to transform individual location issues into a multi-region classification problem, thereby enabling the analysis of perching behavior. Experimental results showed that the accuracy of individual hen location reached 88.8%, with an average location error of 12.53cm. The duration of nocturnal perching for hens ranged from 9.80 h to 10.67h, with each hen requiring about 15cm of perch length. Behavior is one of the key parameters in animal welfare assessment. Hens exhibited a preference for perching at higher levels. Studies on individual perching behavior patterns of hens revealed a preference for the higher sides of the perch compared with the center of the upper level, and more so than the sides and center of the lower level, in descending order of preference as high sides, upper center, low sides and lower center.

Abstract:Accurately estimating the weight of Chinese mitten crab (hairy crab) plays an important role in monitoring growth status, controlling breeding density, determining feeding amount, and predicting production. Existing crab weight estimation methods usually use monocular cameras and rely on reference objects for real body size correction. And the uncertain angle of the hairy crab carapace in the image can easily lead to low target detection accuracy, which limits its application in actual breeding environments. In order to solve the above problems, a crab weight estimation method based on oriented object detection and binocular vision was proposed. Firstly, hairy crab images were collected through binocular cameras. Secondly, a hairy crab carapace oriented object detection model based on SSP-YOLO v7 (SK-SimCSPSPPF-ProbIoU-YOLO v7) was built. The model introduced the selective kernel (SK) attention mechanism in the backbone, used simplified cross stage partial spatial pyramid pooling fast (SimCSPSPPF) to optimize the spatial pyramid pooling, and used probabilistic intersection over union (ProbIoU) loss to calculate the rotation box regression loss, enhance feature extraction capability while reducing the amount of calculation, and effectively improve the accuracy of oriented object detection. Then three-dimensional reconstruction of the hairy crab binocular image was performed, and the body size of the hairy crab carapace was calculated through the Euclidean distance formula. Finally, a particle swarm optimization-extreme gradient boosting (PSO-XGBoost) model based on particle swarm optimization was constructed to fit the body size and true weight of hairy crabs of different genders to achieve weight estimation of hairy crabs of different genders. Tested on the self-built dataset, the mAP0.5 of the model proposed was 99.46%, the model parameters were 7.321×106, the floating point operations (FLOPs) was 1.6684×1011, and the FPS was 39f/s. The weight estimation model based on PSO-XGBoost had a root mean square error (RMSE) of 8.549g, a mean absolute error (MAE) of 6.172g, a coefficient of determination of 0.946 for male crabs, and a RMSE of 6.902g, a MAE of 5.175g, a coefficient of determination of 0.955 for female crabs. The results showed that this method can accurately estimate the weight of hairy crabs and provide technical support for growth monitoring and intelligent breeding of hairy crabs.

Abstract:Aiming to address the problems of insufficient robustness and high computational cost of the Cartographer algorithm during indoor relocation, a relocation method was proposed that enhanced Cartographer by incorporating semantic information, thereby improving robot relocation performance in complex indoor environments. Semantic object information from the robot’s surroundings was extracted by using an RGB-D camera and deep learning techniques, and was subsequently mapped into a structured point cloud. The extracted semantic point cloud was then fused with the grid map constructed by the Cartographer algorithm through projection transformation to generate a complete and informative 2D semantic grid map. A semantic object relationship linked list was also built based on the extracted semantic information to represent inter-object spatial context. During the relocation process, the semantic information provided by the 2D semantic grid map was used to offer a prior pose estimation for the robot, narrowing the search space of the Cartographer algorithm, reducing the number of iterations, and enabling rapid and efficient relocation. Experiments conducted in real indoor scenarios validated the effectiveness of the proposed method. The results showed that, in similar environments, the proposed method improved real-time performance by 49.78% and 78.27% compared with that of the original Cartographer and AMCL algorithms, respectively. In degraded scenarios, improvements reached 76.18% and 83.96%, respectively. Moreover, the relocation success rate was increased by over 75% on average. In addition, the constructed 2D semantic grid map supported semantic navigation and planning, demonstrating promising potential for application in service robots and related fields.

Abstract:In order to improve the autonomy and safety of the transport robot in navigation, it is necessary to plan the robot path reasonably to achieve the goal of optimal path and shortest time. For the problems of the classic A* algorithm in the process of path planning, such as long search time, redundant paths, many inflection points and unsmooth paths, and insufficient ability to avoid dynamic obstacles, an improved AFD (A* Fuzzy-DWA) fusion algorithm was proposed. Aiming at the problems of long search time and many traversal nodes in the planning path of the classical A* algorithm, the obstacle rate evaluation index was proposed to optimize the evaluation function. By calculating the proportion of obstacle grid in the global map grid, it was used as the weight of the evaluation function to reduce the number of nodes in the algorithm. Aiming at the problem of high redundancy of the path planned by the classical A* algorithm, the Smooth Floyd method was proposed according to the Floyd algorithm idea. Through the three times optimization of the initial path, the runnable path with less inflection points and small turning angles was obtained. Aiming at the problem of unsmooth path planning, the inner tangent smoothing strategy was used to optimize the path, and the turning angle in the path was optimized into an arc, which avoided the security threat caused by excessive turning and made the robot run more smoothly. In order to improve the efficiency of local path planning, the fuzzy reasoning method of evaluation function weight was proposed. By calculating the distance between the robot and the target point and the safety point, the evaluation function weight was dynamically adjusted according to the distance, so as to ensure that the robot could reach the predetermined point safely and timely. Experiments showed that compared with the comparison algorithm, the global and local path length of AFD algorithm in the static simulation environment was reduced by 6.6%, 6.9% and 3.3%, 2.7%, and the running time was shortened by 62.1%, 42.1% and 29.7%, 21.1%, respectively. In the dynamic simulation environment, the global and local path lengths were reduced by 11.4%, 7.8% and 8.3%, 4%, and the running time was shortened by 53.1%, 37.5% and 58.4%, 32.6%, respectively. The actual scenario verification further confirmed the effectiveness of the algorithm in improving the autonomous navigation ability and safety of transportation robots.

Abstract:The water air amphibious tilting multi rotor unmanned ships was subject to complex and variable surface fluid forces during cross domain operations in a multi fish pond environment, resulting in significant fluctuations in the ship’s attitude and flight altitude. In order to improve the attitude stability of unmanned ships during surface takeoff, a model compensation based linear self disturbance rejection surface takeoff control method was proposed. Firstly, a detailed dynamic modeling of the multimodal unmanned ship was conducted. Secondly, considering the attitude changes during the takeoff process on the water surface, a real-time attitude estimation model and buoyancy estimation model was proposed based on unmanned ships. Then, a linear self disturbance rejection attitude and altitude controller based on model compensation was designed. Finally, a linear active disturbance rejection controller based on model compensation was designed. In the simulation experiment, compared with the PID algorithm, the proposed method reduced the roll convergence time by 66.7% and the roll fluctuation by 98.3%, reduced convergence time on the x-axis by 34%, reduced convergence time on altitude by 41.2% and the fluctuation on altitude by 80.0%. The simulation results verified the effectiveness and stability of the proposed method. In practical experiments, the unmanned ship achieved takeoff from the water surface with a flying altitude of 1.2m, fluctuations of roll less than three degrees, and a fluctuation of pitch and yaw less than two degrees. The experimental results showed that the proposed control algorithm effectively improved the stability and antiinterference ability of the water air amphibious tilting multi rotor unmanned ships during the water air cross domain process.

Abstract:Aiming to study the effects of regulated deficit irrigation with different nitrogen application levels on winter wheat，a field experiment was conducted in Xinxiang County，Henan Province，located in the Huang-Huai-Hai Plain from 2021 to 2022. Three nitrogen application levels：N1 (120kg/hm2)，N2(240kg/hm2) and N3(360kg/hm2)，and three regulated deficit levels during the stage of jointing-heading：moderate deficit (irrigation amount 18mm，M)，light deficit (irrigation amount 24mm，L) and control (irrigation amount 30mm，CK) were set. The results showed that the net photosynthetic rate (Pn) of winter wheat flag leaves was significantly affected by regulated deficit and nitrogen application level. After re-watering，Pn of all treatments showed different degrees of compensation growth effect，and the compensation effect was enhanced with the increase of nitrogen application rate，N3M and N3L treatments showed a super compensation effect，which was increased by 42.5% and 32.4%, respectively, compared with that before re-watering. Both the aboveground dry matter accumulation at the flowering stage and dry matter accumulation of aboveground vegetative organs at the maturity stage were increased with the increase of nitrogen application rates, while the latter was also increased with the increase of irrigation amount. However, light regulated deficit could promote the above-ground dry matter accumulation at the flowering stage, the translocation of dry matter to grain and the accumulation of nitrogen in plant and ear，and then improved winter wheat yield，water use efficiency (WUE) and nitrogen partial factor productivity (NPFP). The quadratic regression equation of winter wheat yield, WUE, NPFP and nitrogen application rate, irrigation amount was established by using Mintab 15.1 software. It was calculated that when the nitrogen application rate and irrigation amount were 214.6kg/hm2, 166.6mm，the yield was 10973.1kg/hm2，WUE was 2.7kg/m3，NPFP was 52kg/kg，the composite desirability was the highest of 0.7，and the comprehensive benefit was in line with the expected target. Therefore，light water deficit during the jointing-heading stage combined with the nitrogen application of 214.6kg/hm2 could maintain high yield and high water and nitrogen use efficiency of winter wheat，which could be used as a suitable winter wheat deficit irrigation model in this area.

Abstract:The high content of nitrogen and phosphorus in the northern farmland drainage ditch is the main source of eutrophication in the downstream water body. In order to improve the removal effect of nitrogen and phosphorus in the drainage ditch, a pot experiment was carried out in Hetao Irrigation District of Inner Mongolia. According to the low, medium and high nitrogen and phosphorus concentrations in the drainage ditch, three plant combination modes of reed, cattail and reed + cattail were set up, and three planting densities of 15 plants/m2, 30 plants/m2 and 50 plants/m2 were set up for each mode of aquatic plants, including 30 treatments in the blank control. The effects of plant combination patterns and planting density on the purification of nitrogen and phosphorus at different drainage concentrations were studied, and the purification effects of different treatments were comprehensively evaluated based on the entropy weight TOPSIS model. The results showed that for different concentrations of nitrogen and phosphorus in water, single aquatic plants and combined aquatic plants had higher removal efficiency of TN and TP in water under different planting densities. The average removal rate of TN under the combined mode was 7.41 percentage points and 15.61 percentage points higher than that of single reed and cattail, and the average removal rate of TP was increased by 11.71 percentage points and 19.32 percentage points, respectively. In addition, the removal rate of TN and TP by aquatic plants was proportional to the planting density, compared with the planting density of 15 plants/m2 and 30 plants/m2. The average removal rates of TN were increased by 4.75 percentage points and 1.61 percentage points, respectively, and the average removal rates of TP were increased by 6.58 percentage points and 2.64 percentage points, respectively, under high planting density (50 plants/m2). At the same time, the removal rates of TN and TP in the three combined modes were positively correlated with the water concentration. The average removal rates of TN by aquatic plants at high nitrogen and phosphorus concentrations were 2.78 percentage points and 18.56 percentage points higher than those at medium and low nitrogen and phosphorus concentrations, respectively, and the average removal rates of TP were increased by 4.24 percentage points and 10.63 percentage points, respectively. The entropy weight TOPSIS evaluation showed that the comprehensive evaluation was the highest when the combination mode of reed + cattail was planted at low density (15 plants/m2) under low, medium and high nitrogen and phosphorus concentrations, and the comprehensive benefits of nitrogen and phosphorus removal rate and economic cost were the best. The research results had a guiding role in the phytoremediation of reeds and cattails in northern farmland drainage ditches.

Abstract:In recent years, microplastic contamination of agricultural land has become serious, and soil has become a sink for microplastics, posing great risks to soil health and crop safety. In order to investigate the influence of microplastics on the soilplant system, a rice cultivation experiment was conducted with different treatments of low density polyethylene (LDPE) with different mass fractions (0, 0.5%, 1.5%) and particle sizes (150μm, 500μm) to investigate the effects of LDPE on the soil nutrient content, rice growth and physiological characteristics. The results showed that microplastics could increase the total organic carbon (TOC) content of the surface (0~10cm) and middle (10~20cm) soil in the first and middle stages of rice growth (greening, tillering and nodulation) by 6.06%~43.24%, and decrease the TOC content in the late stage (tasseling and yellow ripening) by 6.10%~20.53%, and the trend of the deep soil (20~30cm) was the opposite. Meanwhile, it could significantly reduce the total nitrogen (TN) 5.23%~53.73% and total phosphorus (TP) 2.01%~24.66% in different soil layers. Microplastics were able to promote rice plant height by 3.42%~18.32% in the early stage, inhibit rice plant height by 1.90%~13.96% in the middle and late stage, and reduce rice yield by 7.80%~24.83%. Microplastics were able to significantly reduce rice net photosynthetic rate (Pn) by 6.36%~40.46%, stomatal conductance (Gs) by 3.40%~67.36%, and intercellular CO2 concentration (Ci) by 3.66%~21.86%, and significantly increase saturated vapor pressure difference (VPD) by 14.16%~109.60%, and the promoting effect of VPD on water vapor diffusion was much greater than that on the Gs. The inhibitory effect of VPD on water vapor diffusion was much greater than the inhibitory effect on Gs, which ultimately led to a significant increase in transpiration rate (Tr) by 10.79%~82.37%, and, at the same time, a decrease in chlorophyll a content by 0.27%~3.48%, chlorophyll b content by 0.36%~3.92%, and total chlorophyll content by 0.59%~3.47%. The research results can provide data support and scientific basis for the effect of microplastics on soil health and rice growth stress.

Abstract:Aiming to facilitate irrigation management of crops grown in greenhouse, a model was developed to estimate the reference evapotranspiration (ETo) inside greenhouse during ventilation by using outdoor meteorological data. An experimental study was carried out in May—June 2020, 2021 under grass cultivation conditions in a plastic greenhouse. The Penman-Monteith (PM) equation was used to calculate ETo inside and outside the plastic greenhouse, and variations of the ETo radiative component (ETo,rad) and the ETo advection component (ETo,adv) inside and outside were also analyzed. Thereafter, the radiative transmittance τ and wind speed attenuation rate ω were introduced to improve the radiative and advection sub-equations of the ETo equation, then a simple model for estimating indoor ETo was constructed based on outdoor meteorological data. The results showed that the average ETo, ETo,rad, and ETo, adv of indoors were decreased by 45.6%, 17.1%, and 94.0%, respectively, compared with that of outdoors. During the experiment, the attenuation factors of radiation components ζrad, advection components ζadv, and total flux ζo showed a tendency of ζrad >ζo > ζadv. The two years average ζradwas 23.1% to 33.3% and 90.1% to 93.3% higher than that of ζo and ζadv, respectively. Furthermore,a highly significant linear relationship (P<0.01) between ζrad and τ, as well as between ζadv and ω was found, with R2 of 0.74 and 0.90, MAE of 0.07 and 0.04, RMSE of 0.09 and 0.05, respectively. The simple model equation for estimating indoor ETo based on outdoor meteorological information included four meteorological parameters, including net radiation, air temperature, water vapour pressure deficit, and wind speed, moreover, two greenhouse characteristic factors,τand ω were also considered. The developed model was robust by comparing with the measurements, with R2 of 0.96, MAE and RMSE of 0.17mm/d and 0.22mm/d, respectively. This model developed had high precision and few parameters, and was easy to obtain, which can provide a reference method for the facilitation of greenhouse group irrigation management.

Abstract:Current agricultural greenhouse development still faces issues such as excessively long facility service life, aging equipment, unclear distribution and quantity, and irrational siting of some newly built greenhouses, resulting in low resource utilization efficiency and difficulty in forming highly efficient intensive production areas. To address these problems, the Google Earth Engine (GEE) platform with Landsat 5 TM and Sentinel-2 MSI imagery were utilized, employing random forest, support vector machine, and maximum likelihood classification algorithms to extract the spatial distribution of agricultural greenhouses in Shandong Province over the past 15 years and analyze their age. Furthermore, driving force analysis was conducted considering natural and socioeconomic factors—elevation, rivers, soil organic matter, roads, and population—to identify dominant influences. Finally, the development potential of agricultural greenhouses was quantitatively evaluated, and an optimized layout scheme was proposed. Results showed that the random forest algorithm achieved the highest classification accuracy, with overall accuracy consistently ranging from 83.45% to 92.83% and Kappa coefficients between 0.7531 and 0.8846, demonstrating stronger robustness and adaptability. The area of agricultural greenhouses in Shandong Province was increased from 100440hm2 in 2008 to 473306.67hm2 in 2023, a growth of approximately 471%. Old greenhouses used for over 10 years covered 70606.67hm2, while those over 15 years accounted for 29493.33hm2. Greenhouse development was significantly influenced by policies, modern road networks, and soil organic matter. Potential evaluation identified three key development zones: central Shandong (centered on Shouguang City and Zhangdian District), eastern Shandong (focused on Pingdu City and Laixi City), and southern Shandong (anchored by Lanling County and Xuecheng District). The province can be optimized into core development zones (e.g., Shouguang in Weifang for modern greenhouse expansion), recommended development zones (e.g., Pingdu and Lanling for moderate new construction to form growth poles), retrofit/relocation zones (e.g., Weifang and Liaocheng for upgrading or phased relocation of aging greenhouses), and general development zones (for ecological protection and specialty agriculture). The findings can provide data and decision-making support for optimizing agricultural greenhouse layouts, facilitating rural revitalization.

Abstract:Nitrogen and oxygen doping modification has a significant effect on improving the ammonia nitrogen adsorption and recovery performance of biochar. However, the enhanced adsorption behavior and mechanism of ammonia nitrogen at nitrogen-oxygen doped sites remain to be further clarified. Starting from the molecular atomic scale and relying on the density functional theory (DFT) method, the adsorption behavior and mechanism of ammonia nitrogen in different forms of nitrogen and oxygen were studied in depth by constructing biochar ammonia nitrogen adsorption systems with different nitrogen and oxygen doping structures. The calculation results showed that the adsorption energy between the undoped carbon skeleton structure and NH+4 was 4.65kJ/mol. After nitrogen and oxygen doping, the adsorption effect was increased by 2.59 to 14.81 times and the adsorption effect was significantly improved. At the same time, judging from the influence of doping position, there was little difference in the adsorption of the same group at different doping positions. The adsorption energy was changed between 1.09kJ/mol and 8.49kJ/mol. Therefore, the NH+4 adsorption mechanism of nitrogen oxygen groups in different forms was further analyzed and discovered. In nitrogen-oxygen single doping, the adsorption effect of nitrogen oxide and carbonyl groups was the strongest, which was the result of the combined effect of hydrogen bonding and van der Waals attraction. The adsorption effect of the co-doping of nitrogen and oxygen groups was increased by 4.7~9.8 times, and the adsorption capacity was between nitrogen and oxygen single doping. However, the co-doping of carbonyl and nitrogen oxide groups caused competitive adsorption, which weakened the adsorption effect of the co-doped structure on NH+4. Finally, the ammonia nitrogen adsorption efficiency experiment of different nitrogen and oxygen modified biochars effectively, which verified the rationality of the above theoretical calculation and analysis results. The research results can provide an important theoretical basis for the directional construction of nitrogen and oxygen heteroatom structures on the surface of biochar and its application in ammonia nitrogen adsorption and recovery.

Abstract:Aiming at the membrane contamination problem that is very likely to occur when recovering proteins from gluten processing wastewater by membrane separation method, a weight combination model based on bi-directional long shortterm memory (BiLSTM) was proposed for the prediction of membrane contamination status. Taking the 14 relevant variables collected from the pilot production line of gluten processing wastewater extraction and recycling as inputs, and the changes in membrane flux as outputs, four baseline models were established: support vector machine model (SVM), back propagation neural network model (BP), random forest model (RF), generalized regression neural network (GRNN), together with one given model: BiLSTM model. The weights of the baseline model and the given model were calculated by the inverse error method to construct the weight combination prediction model. Finally, the prediction performance of the single model and the weight combination model was analyzed by using the coefficient of determination R2 and the mean square error (MSE) as the evaluation indexes. The results showed that the weight combination model was able to synthesize the advantages of the singleitem model and significantly outperformed the single-item model in terms of performance. Among them, the BP+BiLSTM+RF model had a high R2 of 0.9906 with high fitting accuracy and MSE of 1.004L2/(h2·m4), which was the lowest among all models. Compared with BP, BiLSTM and RF single-item models, the reduction was 46.05%, 67.24% and 50.81%, respectively. The developed weight combination model can be used for accurate prediction of membrane contamination during protein recovery treatment of gluten processing wastewater.

Abstract:In order to address the issues of decreased stability and susceptibility to overturning in high ground clearance crop protection machinery caused by its high center of gravity and uneven terrain, focusing on the attitude control of the vehicle body of such machinery, a high ground clearance crop protection machinery attitude control system was developed. Firstly, based on kinematic analysis theory, a kinematic model was established to relate the displacement of hydraulic cylinders to the change in inclination angle when each supporting point reached the horizontal position of the vehicle body. Secondly, a fourpoint independent control hydraulic system was designed, and key parameters were calculated to select appropriate hydraulic components. Then, mechanical-hydraulic simulations of vehicle attitude adjustment were conducted by using RecurDyn and AMESim software. Simulation results indicated that under longitudinal and transverse slope conditions, the maximum pitch angle deviation was 0.19° and the lateral angle deviation was 0.1°. When traversing a 200mm high ridge, the maximum pitch angle deviation was 0.28° and the maximum roll angle deviation was 0.17°. Finally, prototype performance tests were carried out. The test results showed that under transverse and longitudinal slope conditions, the maximum lateral angle deviation was 0.21°, and the maximum pitch angle deviation was 0.36°, with absolute adjustment angles all less than 0.5°. During ridge crossing, the pitch angle remained within ±0.5° for over 95% of the time, with minimal terrain influence. The curves of prototype test results aligned closely with those obtained from simulation analysis, validating the effectiveness of the attitude control system and providing theoretical reference for the study of agricultural machinery chassis attitude adjustment systems.

Abstract:In order to solve the problems of high labor intensity, low production efficiency and low degree of mechanization in hilly and mountainous orchards, a power chassis for hilly and mountainous orchards with hydraulic control and mechanical transmission was designed. According to the working requirements of the power chassis of the hilly and mountainous orchards, the key components of the power chassis such as hydro static transmission (HST), suspension lifting device and hydraulic drive system were designed and matched, the hydraulic system model was established by using AMESim simulation software, the structural design and simulation analysis of the hydraulic system were carried out, and the motor speed of the system was determined to be 129r/min and the motor displacement was 35.2L/min, the displacement of the front and rear pumps of the duplex pump was 10mL/r and 6mL/r, respectively, and the working pressure was 25MPa. The performance test of the whole machine was carried out. The performance test results of the whole machine showed that the straight-line driving offset rate of the power chassis was 1.9%, which met the requirements of the corresponding national standard (≤6%). The walking speed of the whole machine was 5km/h, the minimum turning radius was 1.08m, the maximum climbing angle was 27°, the maximum lifting angle of the mounted tool was 30°, the response speed of the rear suspension device was 1s, and the steering performance of the power chassis was good, which can adapt to the narrow slope working environment of hilly and mountainous orchards. When the rotary tiller was mounted for rotary tillage operation, the stability coefficient of tillage width and tillage depth were 98.9% and 96.4%, respectively, which met the requirements of the national standard (≥85%). The hydraulic control system of the crawler chassis of hilly and mountainous orchards conformed to the design standards of crawler tractors for mountainous orchards, which can provide a reference for the development of power chassis for hilly and mountainous orchards.

Abstract:In response to global climate challenges and energy transition imperatives, the advancement of hydrogen fuel cell vehicles featuring high energy efficiency and zero-emission characteristics represents a strategic pathway. Focusing on the development of a two-stage aerodynamic gas bearing centrifugal compressor specifically engineered for a 130kW vehicular fuel cell system, achieving critical performance parameters, including pressure ratio of 2.8, mass flow rate of 130g/s, and operational speed of 88000r/min. A general flow model and a leakage flow model for the two-stage aerodynamic gas bearing centrifugal air compressor were constructed, numerical simulations and comparative analyses of these two models were carried out. The thickness and shape of the impeller back cavity and the geometric number and structural parameters of the annular seal teeth in the leakage flow model were investigated. Performance tests and validation of the air compressor were conducted on an upgraded centrifugal air compressor comprehensive performance test bench. The results indicated that the pressure ratio error of the flow model considering the leakage channel under the design condition was reduced from 10.63% to 5.11%. Appropriately reducing the thickness of the impeller back cavity and increasing the axial dimensions of the seal teeth can reduce gas leakage. The calculation formula of the leakage coefficient with the number of sealing teeth was fitted. The two-stage aerodynamic gas bearing centrifugal air compressor impeller had a high safety margin, its structural strength can meet the requirements of ultra-high speed operation. And the isentropic efficiency of the two-stage aerodynamic gas bearing centrifugal air compressor can reach 73.30% under design conditions, with a maximum stable rotational speed of up to 95000r/min.

Abstract:Redundant serial robots possess the advantages of a large workspace and good dynamic characteristics, and have been widely used in human-robot collaboration. Due to the characteristics of unstructured scenes and diversified tasks in human-robot collaboration, it is necessary to control the end stiffness of robots to ensure interaction safety. However, common stiffness planning algorithms suffer from low power transmission efficiency and inefficient computation. Aiming to address these issues, a Cartesian stiffness planner was proposed based on motion/force transmission performance. The algorithm employed a sequential least squares optimization technique to compute the robot’s workspace trajectory. Local transmission indices were introduced to enhance power transmission efficiency within the trajectory. The optimization objective was simplified through the use of a geometric shaping method based on stiffness ellipsoids. Furthermore, the optimization space dimensions were reduced by extending the arm angle description method, thereby improving the computational efficiency of the optimization algorithm and reducing computational resource consumption. The efficacy of the proposed trajectory planner, based on the geometric shaping method for robot Cartesian trajectories, was demonstrated through simulation verification conducted by using the Matlab-based Simulink platform. Experimental validation was performed on the Franka Panda redundant serial robot platform, confirming the feasibility of realizing the desired stiffness direction at the robot end by this planner.

Abstract:It is difficult for 3PRS parallel mechanism to control due to its strong coupling and susceptibility to disturbances. Thus the decoupling and control strategies of the mechanism were studied. Firstly, the inverse kinematics model of the mechanism was established, the kinematic equations were deduced between the position and pose of the moving platform and the heights of the input sliders. The kinetic and potential energies of the motion component within the mechanism were analyzed, the Lagrange dynamic equations for the 3-PRS parallel mechanism were obtained. The driving forces on the position and pose of the moving platform in both zero-gravity and normal gravity environments were further investigated. The theoretical model and numerical simulation were performed and the consistency and accuracy of the dynamic model were confirmed. The state-space equations based on the inverse dynamics model were formulated, and the Lie derivative expressions for the 3-PRS parallel mechanism were studied to enable feedback linearization decoupling of the state-space equations. The simulation analysis was conducted on the decoupled statespace model. The experiment platform of 3-PRS parallel was set up to verify the effectiveness of the proposed method. The controller for the 3-PRS parallel mechanism was designed on the basis of the complete decoupling and the integral sliding mode control. The effectiveness of the controller was validated through the simulation experiments. The results indicated that the designed controller can not only decouple its dynamical model but also can track the expected trajectory under the presence of disturbances and time-varying input driving forces, which made the system possess with strong robustness.