Abstract:Smart agriculture is the development direction of modern agriculture, and unmanned smart farms are an important way to achieve smart agriculture. Unmanned smart farm is an important direction for agricultural transformation and upgrading, and its precise and efficient operation quality depends on the accuracy and reliability of farmland boundary recognition technology. The technical system and workflow methods of farmland boundary recognition were systematically reviewed, with a focus on analyzing the characteristics and application scenarios of three types of data acquisition methods: satellite remote sensing, UAV remote sensing, and ground-based sensing. The advantage of satellite remote sensing lies in its wide-area, periodic monitoring capability that supports large-scale farmland change analysis, though its spatial resolution is limited. UAV high-resolution images, when deeply integrated with ground sensors ( LiDAR point clouds and RGB image registration), can achieve centimeter-level boundary segmentation, providing high-precision data support for complex farmland scenarios, but its field of view is limited. Traditional image processing algorithms (threshold segmentation, edge detection) offer real-time advantages in regular farmlands but struggle with scenarios involving objects of similar spectra and static element occlusions. Deep learning-based models such as U Net and DeepLab, through multi-scale feature fusion and attention mechanisms, significantly enhance the robustness of irregular boundary recognition. Current technologies support the construction of digital maps and agricultural machinery path planning. However, there are still three main bottlenecks: insufficient spatio- temporal alignment accuracy of multi-source data, resulting in low fusion efficiency; slow inference speeds of lightweight models on edge computing devices, which failed to meet real-time operation demands; and the lack of dynamic farmland boundary update mechanisms, restricting long-term monitoring effectiveness. Future research should focus on multi-modal spatio-temporal feature fusion, lightweight model technologies driven by edge inference, and a framework for autonomous updating of digital farmland maps supported by air space ground collaboration, to provide theoretical support for high-precision, high-response, and high-dynamic boundary recognition in farmlands.

Abstract:Path planning is one of the key factors determining the efficiency and quality of the first season harvest operation for regenerated rice. At present, there are many studies on the full coverage path planning technology of unmanned agricultural machinery in the operating area, but little consideration is given to the problem of the harvester crushing the regenerated rice during field harvesting. Therefore, research on the path planning of regenerated rice harvesting was conducted with reduced crushing. By analyzing farmland information, waiting areas, and unloading of grain, the problem of harvesting and unloading path planning for regenerated rice was transformed into a vehicle routing problem with capacity constraints (CVRP). A mathematical model for the harvesting path of regenerated rice was constructed with the goal of minimizing the rolling area and minimizing the total path of the harvester. A hybrid algorithm for path planning of regenerated rice with less compaction was proposed, using traditional ant colony algorithm (ACO) and 2-opt algorithm to obtain the optimal path. Taking the driverless harvester of ratooning rice as the object, the field experiments of straight path planning, headland turning, grain unloading planning and full link field operation were designed. The field experiments were conducted with the auto drive system to investigate the field compaction rate of the harvester. The results showed that the average absolute error of linear tracking was 3.51 cm, the maximum deviation was 8.24 cm, and the compaction rate of the linear segment was 17.55% . The compaction rate in the head area was decreased by 52.2% . The path planning designed had a total field compaction rate of 27.42% , which met the special operational requirements for regenerated rice. The research result can provide theoretical and technical support for reducing the crushing area of unmanned harvesting machines for regenerated rice.

Abstract:Automatic navigation tractors experience large path tracking errors, easy overshoot in correction, and long adjustment times under conditions such as lateral deviation and sideslip. Aiming to address the problem of fast correction for lateral deviation and sideslip in tractors, an improved pure pursuit path tracking control method was proposed, based on a lateral control compensation strategy for the drive wheels. By constructing a tractor sideslip model on a slope and combining it with a two-wheel vehicle kinematic model, a lateral control compensation strategy was introduced to improve the classic pure pursuit algorithm, achieving precise lateral compensation control for autonomous tractors. To validate the performance of the proposed lateral deviation compensation improved pure pursuit path tracking algorithm, CarSim/ Simulink co-simulation was designed. The results of the sloped test showed that the improved algorithm enhanced control accuracy by 73.6% on a 10°slope compared with that of the classic algorithm. In the continuous sideslip simulation, the classic algorithm failed to escape, whereas the improved algorithm completed escape within 3.9 s, with lateral deviation converging to within 0.01 m and an overshoot of 0.14 m. Field tests were conducted to verify the effectiveness of the improved algorithm, with multiple trials showing an average escape time of 7.03 s and a maximum overshoot of 0.054 m. The test results demonstrated that the proposed lateral control compensation strategy significantly improved the control accuracy and stability of autonomous navigation tractors in complex working conditions.

Abstract:In order to solve the problem that the sudden change of the speed of automatic driving agricultural machinery in paddy field leads to inaccurate angle estimation, a steering wheel angle estimation method of agricultural machinery in paddy field was proposed based on dual observation fusion Kalman filter, and a steering wheel angle estimation model of agricultural machinery in paddy field was established. Firstly, the improved two-wheeled agricultural machinery sideslip model was used to obtain the front wheel steering angle of paddy agricultural machinery based on kinematics model. Secondly, the collected GPS speed and inertial navigation speed were compensated by weighted observation fusion method. Finally, a method for estimating the front wheel steering angle of paddy agricultural machinery based on dual observation fusion Kalman filter was proposed, which took the front wheel steering angle based on kinematics model and the front wheel steering angle based on steering motor coding as dual observation values, so as to estimate the front wheel steering angle of paddy agricultural machinery. In order to verify the proposed method, speed correction, front wheel steering angle estimation test and linear tracking test were carried out in paddy field on the platform of rice direct seeding machine. The results of speed correction test showed that the unevenness of paddy field hard bottom layer was the direct reason for the poor fitting accuracy of front wheel angle. The proposed method stabilized the speed of direct seeding machine in a certain range, and solved the problem of poor fitting accuracy of front wheel angle caused by the fluctuation of paddy field hard bottom layer. The front wheel steering angle estimation experiment showed that the average tracking error of the virtual wheel angle relative to the angle change of the angle sensor was 0.12°, the maximum deviation was 1.67°and the standard deviation was 0.4°. The method can accurately measure the steering angle of the front wheel of agricultural machinery, and finally control the direct seeding machine to track the target angle stably, which met the accuracy requirements of estimation of front wheel angle of agricultural machinery in paddy field. The results of linear tracking test showed that the average error was 3.14 cm and the standard deviation of position deviation was 2.11 cm in paddy field environment. The method proposed was suitable for unmanned paddy field, which improved the accuracy of corner estimation and the quality of agricultural machinery navigation.

Abstract:Aiming at the problems of slow response speed, low control accuracy, poor stability, and insufficient robustness caused by the influence of complex environment of collaborative control in agricultural machinery transfer scenarios, a multi-machine collaborative control method is proposed for agricultural machinery formation transfer. The multi-machine collaborative model was constructed in which the host machine drived and navigated manually and the slave machines followed automatically. Based on the Frenet coordinate transformation, the collaborative control was decoupled into horizontal and vertical control, and the longitudinal controller was designed by using the model predictive control algorithm to achieve the maintenance of the relative distance and the following of speed and acceleration among the units, and the horizontal controller was designed by using the pure tracking algorithm to achieve the slave machines to follow the trajectory of the host machine and the introduction of fuzzy algorithm adjusted the key control coefficients to optimize the control effect in real time. Based on the CarSim/Simulink platform to design a variety of transfer typical working conditions on the designed method for simulation test analysis, comparisons showed that compared with the traditional control methods it had a more reliable and superior performance, and based on the intelligent tractor unit to carry out the real vehicle test verification, the results showed that the relative speed error was less than 0.570 m/s, the relative distance error was less than 0.169 m, the acceleration error was less than 0.252 m/s2 ,the lateral error was less than 0.090 m, all of them can gradually and steadily meet the actual needs of agricultural machinery formation transfer.

Abstract:Most of the unmanned tractors in the process of trajectory tracking focus on the tracking performance, ignoring the operational energy consumption resulting in poor economy. Aiming at the above problems, a layered collaborative control strategy that considered economy and trajectory tracking performance was proposed. Firstly, the trajectory tracking system was established by model predictive control algorithm with longitudinal deviation and transverse deviation as the target and acceleration and front wheel angular speed as the constraints, and secondly, the binary regulating economic control strategy based on external parameter optimization was established by taking the ratio of engine fuel consumption rate and hydro-mechanical CVT transmission efficiency as the optimization target. On this basis, the trajectory tracking system and economic control strategy were integrated to form a hierarchical cooperative control strategy with the predicted speed of the tractor at the next moment and the current plow resistance of the tractor as the transfer variables. Pure tracking and one-dimensional economic control strategy were used as the comparison strategy, and the cooperative control strategy was simulated based on the Matlab simulation platform, and the effectiveness of the cooperative control strategy was verified by the hardware-in-the-loop test platform. The results showed that compared with the comparison strategy, the layered cooperative control strategy effectively reduced the trajectory tracking deviation of the unmanned tractor and improved the tractor’s economy, the speed variance was reduced by 36.7% , the longitudinal tracking deviation was reduced by 89.8% , the lateral tracking deviation was reduced by 91.7% , and the tractor fuel consumption was reduced by 11.8% .

Abstract:Unmanned farm will be the ultimate form of rice and oil rape cultivation in the rice oil rape rotation area in the middle and lower reaches of the Yangtze River in China. By analyzing the production mode, implement type and operation path planning requirements of unmanned farming in rice and oil rape rotation, intelligent farm implement operation and maintenance software for unmanned farming was designed based on the Android application framework, including modules of plot management, implement attribute management and path planning for the path requirements of the field operation process of typical operation links of unmanned farming for rice and oil rape rotation cultivation. On the basis of the existing algorithms of the group, focusing on designing algorithms for operation path planning in two typical processes, namely integrated rice harvesting and rapeseed sowing and paddy ploughing, and the field operation path planning algorithms coded in Python beforehand was called through the mixed programming of Chaquopy plug-in. The results of simulation test and field test showed that the designed and developed Android software was stable and reliable, with good human computer interaction, and path planning algorithms were able to provide effective operation paths for different implements and common quadrilateral plots, and the operation time of the single field operation path planning algorithm ranged from 29 ms to 1 898 ms, and the computational efficiency and rationality of the paths met the needs of unmanned production of typical operation links in the real application. The computational efficiency and path reasonableness met the needs of unmanned production in typical operations, providing theoretical and technical support for the construction of unmanned farms in the middle and lower reaches of the Yangtze River for rice and oil rape rotations.

Abstract:This study addresses the issues of complexity in greenhouse operational environments and poor stability of existing mechanical walking systems by conducting research on the autonomous following electric platform walking speed control in greenhouses. Due to the system’s inherent nonlinearity and time-varying characteristics, traditional PID control algorithms fail to achieve effective control. Therefore, a dung beetle optimizer (DBO) optimized BP neural network PID control algorithm was proposed. This algorithm optimized the weights of the BP neural network by using the DBO algorithm, thereby accelerating the self-learning rate of the BP neural network. It achieved rapid and precise control of the greenhouse autonomous following electric platform walking speed, enhanced system response speed, and reduced overshoot. Experimental results demonstrated that at a walking speed of 1 m / s, the system exhibited an average response speed of 0.11 s, settling time of 0.27 s, and a maximum overshoot of 2.44% . When there were changes in track speed and direction, the system maintained advantages of fast response, minimal overshoot, and oscillation-free steady-state process. Compared with PID control algorithm, BP-PID control algorithm, GA-PID control algorithm, ACO-PID control algorithm, the DBO-BP-PID control algorithm showed superior performance in control stability and precision, effectively mitigating system hysteresis and nonlinear effects, thereby meeting the control requirements for greenhouse autonomous following electric platform walking speed.

Abstract:Digital twin technology digitizes the entire lifecycle of physical entities to achieve monitoring and control of their states, providing a solution for remote control and continuous operation of robots. The high-precision control of the operation chassis during the driving process is the key to ensuring the quality of robot operation. Aiming to address the issues of large error in the driving state prediction model due to changes in greenhouse environment and chassis wear, as well as the difficulty of online dynamic data collection, a digital twin based online monitoring method for the driving state of the greenhouse operation chassis was proposed. Firstly, a digital twin system of the greenhouse operation chassis geared towards the driving state was developed. It dynamically perceived the dynamic data in the driving process online and simulated the change process of the chassis driving state in real time. Then an online prediction model of the driving state of the greenhouse operation chassis was constructed by combining the temporal deviation quantification model of the chassis driving state and considering various uncertain factors during the driving process. Finally, an experimental environment for online monitoring of the chassis driving state was set up, and online monitoring experiments and driving effect verification tests were carried out. The results showed that the online prediction method proposed corresponded to lateral offset prediction accuracies of 96.32% , 95.96% , 95.69% and 96.11% for datasets M1 , M2 , M3 , M4 , respectively. The longitudinal offset prediction accuracies were 96.58% , 96.36% , 96.51% and 96.13% for datasets M1 , M2 , M3 , M4 , respectively. Compared with the BP + SVR method, the prediction accuracy of lateral displacement was increased by 3.61% , 3.26% , 3.92% , and 3.98% , respectively, and the prediction accuracy of longitudinal displacement was increased by 2.96% , 2.78% , 3.27% , and 3.06% , respectively. This proved that the online prediction method proposed can effectively correct for the bias effects caused by ground fluctuations and chassis wear and tear. The average values of the actual lateral and longitudinal deviations of the chassis during driving were reduced by 48.13% and 49.49% , respectively compared with the chassis driving method based on fixed driving parameters. This method can dynamically adjust based on the real-time driving status of the chassis. The online monitoring method for greenhouse operation chassis driving status based on digital twins had the characteristics of strong real-time and high accuracy, which can provide a basis and reference for the continuous operation method and technology of robots in facility agriculture.

Abstract:In order to solve the current stage of agricultural robots in the working channel of the demonstration greenhouse,dynamic obstacle processing is difficult,poor target accessibility,easy to fall into the local minimum and so on, the dual obstacle cost function,adaptive weights and virtual target_ dynamic window approach (DAV_DWA) was proposed to achieve greenhouse robot local path planning. Firstly,a dynamic static dual-strategy obstacle avoidance method was adopted,which divided the safety distance of dynamic and static obstacles into two evaluation functions to reduce the collision risk of dynamic obstacles and prevent excessive obstacle avoidance of static obstacles. Secondly, an adaptive strategy for evaluation function weights was proposed, adaptive adjustment of the weights of each evaluation function according to two obstacle distances to enhance the robot’s path-finding ability in different complex environments. Finally,the virtual goal method was proposed to enable it to continue navigation after detaching from the local minimum,so as to enhance its path planning ability for the local minimum. Comparative simulation experiments and greenhouse experiments were carried out, and the results showed that compared with other algorithms,DAV_DWA was able to reach the target point with a shorter path in a shorter time under the premise of guaranteeing the safety;in the greenhouse scenario,the robot can complete the autonomous navigation task,and the positioning error was no more than 0.12 m, and tracking error was no more than 0.10 m,which was in line with the practical requirements.

Abstract:Optimizing unmanned farm paths to improve farm management efficiency and resource utilization is a hot research topic in the field of mobile robot navigation. An improved sparrow search algorithm ( ISSA) incorporating improved Q-learning ( IQL) algorithm was designed to address the problems of low search efficiency and smooth paths that can easily fall into local optimization of traditional sparrow search algorithm (SSA) and reinforcement learning algorithm. ISSA incorporating the improved IQL algorithm was designed for global path planning of mobile robots in combination with Bessel curves. Firstly, a multi-strategy initialization of the population was used at the beginning of the algorithm, combining the IQL algorithm with Logistic chaos mapping and Latin hypercube sampling (LHS) methods to provide excellent and diverse initial solutions for the population;secondly, a linear dynamic inertia weight adjustment method was introduced into the finder position updating to balance the algorithm’s global search capability and local exploitation capability, and improve the convergence speed of the algorithm;then, the reverse learning strategy was introduced into the vigilant to further explore the unexplored area and prevent falling into the local optimal solution;finally, the path was smoothed by combining obstacle avoidance algorithms and Bessel curves to eliminate the problems of traveling paths too close to obstacles and unsmooth paths. The effectiveness and superiority of ISSA algorithm was verified through comparative simulation tests on Matlab platform. The experimental results showed that the ISSA algorithm effectively combined the self-learning characteristics of the IQL algorithm and the powerful search capability of the SSA algorithm, which significantly improved the efficiency of global path optimization and generated smoother paths in both the grid simulation environment and the field scenario. In the field scenario, the ISSA algorithm reduced the path planning time by 64.43% and 9.94% , and the average value of the shortest path length by 8.3% and 12% , respectively, compared with the SSA and ACO algorithms, which provided a high-quality path planning solution for the unmanned farm robots to work accurately and efficiently.

Abstract:Accurate row alignment harvesting of grapes can effectively reduce the collision between vibration mechanism of the harvester and the trellis, which is an important means to achieve large-scale mechanized harvesting. Based on the local driving scene model between grape rows in Frenet coordinate system, an automatic row alignment path planning algorithm for grape harvesters was proposed. Using the global operation path as a reference line, the algorithm utilized onboard LiDAR to identify grape rows ahead in real time, and applied the K-means algorithm to cluster the point cloud of grape rows. The Lattice algorithm was used to dynamically sample the driving area ahead according to the traveling speed, and then the local path clusters were generated based on fifth-order polynomials. The extreme steering positions of the front and rear wheels were taken as the feature points of the harvester, and then the collision detections were conducted between feature points and the lateral segmentation minimum bounding rectangle of grape rows, and the offset costs of each local path relative to grape rows and the global path were calculated. Based on the operating states and environment condition, the decision limits of the grape line deviating from the reference line were determined, and the weighted sum of the offset costs were optimized by dynamic programming algorithm, and then the path with the minimum cost in the path cluster can be obtained as the current local path. The algorithm was validated through simulation by using the robot simulation software Gazebo and Rviz, as well as real experimental tests. The results showed that the average lateral error of the planned local path relative to grape rows was 4.37 cm, and the maximum absolute curvature was 0.201 1 m-1. When the global path deviated significantly from the grape row, the local path can effectively correct the deviation and meet the driving requirements for grape harvesting operations. In the simulation test for planning a path of 6 m, the average processing time of this algorithm was 213 ms per iteration, with a maximum of 337 ms per iteration. In the experimental test for planning a path of 6 m, the average processing time was 577 ms per iteration, with a maximum of 816 ms per iteration. The relevant research methods can provide reference for local path planning of agricultural machinery in vineyard scenarios.

Abstract:To address the challenges of high-frequency oscillations and difficulty in parameter tuning typically encountered in traditional sliding mode control (SC) for path tracking in agricultural machinery, a novel control approach was proposed. The research focused on a tracked agricultural chassis, where a kinematic model and deviation dynamic equations were firstly developed. A super-twisting sliding mode control (TSC) law was then designed to ensure accurate path tracking. In addition, a parameter pre- tuning controller was introduced to facilitate the fine-tuning of control parameters. The finite memory Broyden algorithm was employed to optimize the parameters of the super-twisting sliding mode controller, enhancing its robustness. The stability of the path tracking system was verified by using Lyapunov stability analysis, ensuring that the proposed control method maintained stability under various operational conditions. Simulation results demonstrated that when the tracked agricultural chassis operated at a speed of 1.0 m / s, the maximum absolute tracking error was reduced to 0.063 m, and the average absolute error was minimized to 0.013 m. When compared with traditional linear PID control and conventional sliding mode control, the maximum deviation was reduced by 61.3% and 62.1% , respectively. Furthermore, the absolute average error was decreased by 89.2% and 75.4% , respectively. These results indicated a significant improvement in tracking accuracy with the proposed method. Field test results further validated the effectiveness of the control algorithm. When the operational speeds were 0.5 m / s and 1.0 m / s, the absolute average error of the parameter pre-tuned super-twisting sliding mode control algorithm was reduced by 69.2% and 50% , respectively, compared with that of the traditional sliding mode control. Additionally, the heading error was reduced by 61.1% and 40% , respectively, contributing to a noticeable reduction in high-frequency oscillations during operation. Overall, the proposed TSC strategy significantly enhanced the path tracking performance of agricultural machinery, providing a more stable and reliable control method compared with existing approaches. The method not only improved tracking accuracy but also effectively mitigated high-frequency oscillations, making it suitable for real-world applications in agricultural vehicle guidance systems.

Abstract:Working on irregular farmland along a curved path can effectively improve the applicability of unmanned agricultural machinery operations. Aiming at the problem of low coverage of unmanned curved farmland operations, a curved path tracking control method was proposed based on full-state feedback control. A third-order full-state feedback controller with lateral deviation, heading deviation and heading increment as state vectors was constructed. The system was linearized and converted into a spatial matrix form. The gain matrix was solved under the condition of satisfying the system stability constraint to obtain the front wheel steering angle. The front wheel steering angle was compensated according to the curvature of the nearest path point. The controller performance was verified based on the Matlab / Simulink simulation environment. The results showed that the average absolute error after compensation was reduced by 31% compared with that of the algorithm before compensation, and the maximum absolute error was reduced by 17.9% . Three sets of field experiments were designed to verify the proposed method. The test results of variable curvature path showed that when the operating speed of agricultural machinery was 0.7 m/ s, the maximum lateral deviation was 0.070 5 m, the absolute error was 0.021 8 m, the standard deviation was 0.023 4 m, the maximum heading deviation was 12.25°, the absolute error was1.37°, and the standard deviation was 1.68°;the test results of fixed curvature path showed that the maximum lateral deviation was 0.103 4 m, the absolute error was 0.042 4 m, the standard deviation was0.047 7 m, the maximum heading deviation was 8.9°, the absolute error was 1.86°, and the standard deviation was 2.45°;the results of tracking test along curved ridges showed that the maximum lateral deviation was 0.059 7 m, the absolute error was 0.012 3 m, the standard deviation was 0.015 8 m, the maximum heading deviation was 7.01°, the absolute error was 1.85°, and the standard deviation was 2.49°. The tracking accuracy can meet the operational requirements. The research results can provide theoretical and technical support for unmanned agricultural machinery to perform complex and changeable curved edge operations.

Abstract:The acquisition of field road boundary information is the basis for making high-precision farmland map. In order to solve the problems of inaccurate segmentation of field roads in high-resolution orthophoto maps, such as missed segmentation and false segmentation, a deep learning network model was proposed based on improved U Net. Firstly, the backbone network was replaced with ResNet50 to enhance the ability to extract the features of drivable roads in the field. Secondly, the DSConv module, which can improve the accuracy of tubular structure, improved the accuracy of the field drivable road, and inhibited the feature extraction of the background of field features similar to the field road. Finally, the complete context information was obtained by inserting the ECA Net attention mechanism, and the feature restoration process of the drivable road in the field was optimized, so as to improve the overall segmentation accuracy of the model. Then the traditional image processing method was used to further denoise and eliminate the hole of the segmentation results, and in view of the problem of losing geographic information in the recognition results, so as to obtain high-precision field road boundary information. The experimental results showed that the improved U Net model had the highest evaluation index values in the comparison with the semantic segmentation model in the test set of the constructed dataset with 95.46% MPA and 91.12% MIoU, after post-processing using traditional image processing methods, MIoU and MPA were 92.64% and 96.75% , respectively, the MIoU and MPA were increased by 1.29 percentage points and 1.52 percentage points;and 86.39% and 90.01% in the field drivable road recognition test of high-resolution orthophoto map, respectively, which can clearly identify the field road. After using the traditional image processing method to optimize the obtained high-resolution orthophoto results, the MIoU and MPA were 88.34% and 91.53% , respectively, and the MIoU and MPA were increased by 1.95 percentage points and 1.52 percentage points, respectively. The research result can provide accurate field road boundary information for the subsequent production of high- precision farmland map.

Abstract:Dryland agricultural cultivation modes include flat cropping and ridge cropping, and the terrain undulation of different cultivation modes varies greatly, so accurate crop row cultivation mode recognition is of great significance to the stability of robot travelling. A methodology for identifying the terrain of crop rows and ridges utilizing appropriate sensor signals was introduced. Initially, the inertial measurement unit ( IMU) signals were collected from a quadrupedal robot navigating through the crop rows of a corn field. The velocity data from the robot’s left front leg served as supplementary information to compile a comprehensive signal dataset, encompassing the robot’s movement in both flat cropping and row cropping modes, with two distinct row heights. Subsequently, spatial information features were extracted from the signals by using convolutional neural networks (CNN), while time series features were derived through bidirectional long short-term memory (BiLSTM) networks. Additionally, self-attention (SA) was employed to capture the attention scores of the output feature information from both CNN and BiLSTM. Ultimately, the efficacy of the proposed model in distinguishing between flat and ridge crop types was validated through model comparisons and field experiments. The results indicated that the F1 score of proposed CNN BiLSTM SA model reached 92% , marking an improvement of 10.17, 3.51, 2.57 and 1.27 percentage points over that of the CNN, CNN LSTM, CNN LSTM SA, and CNN BiLSTM models, respectively. When the recognition model was embedded in the field robot, it achieved a 90% accuracy rate in identifying the current crop row tillage type within 1.4 s, and met the classification criteria for flat and ridge categories within 4.8 s. This performance satisfied the robot’s requirements for rapid and accurate recognition across various tillage terrains. The algorithm can provide the robot with ability to recognize crop rows under typical tillage patterns in dry fields, and the results can provide technical support for improving the field stability of quadrupedal robots in autonomous operations.

Abstract:The complex operating environment of agricultural quadruped robots causes them to fall easily when walking in the field, which affects the operating efficiency of the robot, and accurate prediction of the body fall state is of great significance to the walking stability of the robot. A critical state prediction method for robot fall was proposed based on ontology sensor signal processing. Firstly, the inertial measurement sensor signals of the quadruped robot walking and falling in a corn field and the fall state of the robot during field walking simulated by Gazebo software were collected, and the signals of the robot’s normal walking, the two phases of the critical stable state of falling and the four working conditions of complete falling were classified to generate signal datasets of different body states. Secondly, a population optimisation algorithm was used to optimize the parameters of variational mode decomposition (VMD), and an improved population optimization variational mode decomposition ( IPO VMD) algorithm was proposed. And IPO algorithm was adopted to optimize the parameters of general regression neural network (GRNN), and improved population optimization general regression neural network (IPO GRNN) was proposed. Finally, based on the above signal processing method, a fall prediction method for field operation robots based on the IPO VMD GRNN model was established, and the signals of the traverse roll and pitch attitude angle of the robot’s actual field walking were used as the model test data to verify the performance of the fall prediction model for field operation robots. The test results showed that the IPO VMD GRNN model outputed a total error of 0.146 7, an average relative error of 0.006 5, and a mean square error of 0.000 3, and the extracted features were well represented;compared with the VMD BPNN, VMD GRNN, and PSO VMD GRNN models, the average prediction of a successful response time was faster than the average predicted response times of 127.75 ms, 91.5 ms, and 39.5 ms. The algorithm can provide the ability to predict the critical state of robot fall when the robot walked in the field, and the results can provide technical support to improve the field passability of quadruped robots for autonomous operation.

Abstract:Aiming to address the problems of difficulty in monitoring tractor-mounted agricultural implements in complex field environments and the excessive amount of model parameters, a lightweight RepViT-based agricultural implements recognition model, tractor-mounted agricultural implements net (TMAInet ), was proposed. Firstly, the self-developed agricultural machinery service platform ‘Agricultural Mechanisation Precision Operation Platform’ was used to collect the datasets of agricultural implements in six working states, and the training set was expanded to 6 627 frames by data enhancement methods such as copy-paste. Secondly, based on the RepVIT network model framework, a convolutional feed-forward module ( CFF) was designed to improve the ability of fine-grained feature extraction at different scales, and an attention mechanism, ECA, was introduced to optimize the model parameter structure and simplify the feature extraction module. Finally, the model was trained by pre-training + fine-tuning (PF) migration learning method and deployed on Jetson nano edge devices. The experimental results showed that the recognition accuracy, F1 score and recall of the TMAInet model reached 99.13% , 98.53 and 98.78% , respectively, by the PF migration learning method. Compared with the original RepVIT model, the recognition accuracy, F1 score and recall were improved by 1.86 percentage points, 3.04 percentage points and 1.95 percentage points, respectively, and the number of parameters was reduced to 7.3 × 10 6 while maintaining 73 f / s at the edge device side. TMAInet was able to accurately and efficiently monitor the common categories of agricultural implements in practical applications, and it can provide a technical reference for the development of unmanned smart farms.

Abstract:Weed control in paddy fields is a key agronomic measure to improve rice yield, and chemical weed control is widely used because of its high efficiency. Traditional chemical weed control relies on manual operation and often uses large area spraying, which increases the operation cost and causes negative problems such as environmental pollution. Based on this background, a precision spraying type paddy weeder for adaptive weeding operation was designed. The weeder spraying device and system were constructed, and the weed detection system with MS YOLO v7 as the core framework was designed based on the constructed diversified paddy field weed dataset. The MS YOLO v7 model combined the backbone network with MobileOne, and replaced the CIoU loss function with the SIoU loss function. The model performance was verified by ablation test and different model comparison test, and the results showed that the model recognition accuracy was 95.65% , the mean average precision ( mAP) was 92.67% , and the real-time performance reached 51.29 f / s. The IR model was reasoned by using OpenVINO on Raspberry Pi, and the detection of a single paddy field weed image took 0.806 s. The constructed spraying system can instantly capture and analyze the transmission signals from the weed detection system, and then realize the precise regulation of the weed spraying device. The results of the field test showed that the precision spraying type paddy field adaptive weeder had a seedling injury rate of 2.95% , a target application accuracy of 94.98% , and a coefficient of variation of 0.128% , which met the agronomic requirements for weed control in paddy fields. The weeder realized the unmanned operation of paddy field weeding and it can provide technical reference for the intelligent development of agriculture.

Abstract:Aiming to address the challenges of arduous threshold delineation, inadequate robustness, and insufficient adaptability of conventional clearing loss detection sensors that depend on temporal domain feature thresholds to distinguish kernel impact signals, a comprehensive clearing loss detection system for corn kernel direct collectors was developed, and a kernel impact classification algorithm predicated on a variable scale one-dimensional convolutional neural network (VS-1D CNN) was proposed. Initially, the hardware circuitry and software processing program were engineered for impact signal acquisition, processing, and transmission, alongside the development of the supporting host computer. Subsequently, a data acquisition testing platform was established to gather and archive the impact signals of weeds and maize kernels under varying impact heights and angles, thereby constructing a data set and training the VS 1D CNN seed impact classification algorithm, with the training outcomes indicating that the model’s accuracy was 94.2% on the testing set. Ultimately, the efficacy of the devised detection system under diverse operational conditions and the classification performance of distinct stray residues and seed mixtures were validated, with results demonstrating that the proposed VS-1D CNN algorithm performed commendably, achieving detection accuracy exceeding 95% across different installation sites and varying seed flow rates;the classification accuracy for identifying different proportions of stray residues and seed mixtures surpassed 93% , signifying that the proposed algorithm exhibited exceptional performance. This underscored that the algorithm delineated in this manuscript possessed remarkable efficacy and can accurately detect seed losses without establishing a fixed temporal domain feature threshold.

Abstract:In order to achieve efficient autonomous operation for soil parameter detection in the wide-row planting environment of fruits and vegetables, the structure and control circuit of the soil drilling module and the detection sensor movement module for the multi-parameter soil detection robot were designed according to the requirements of automated soil testing tasks, and it was equipped with a visual navigation module. The visual navigation control module used the lightweight segmentation model DS-U2Net for path recognition, extracted the region of interest from the segmented path, acquired the left and right boundary points to calculate the middle navigation point, and then fit the navigation line by using the least squares method. Combined with the real-time acquisition of the robot’s heading angle, the PID algorithm was applied for accurate walking navigation control. Experiments showed that the DS-U2Net model had only 6.5×10 5 parameters, with a recognition frame rate of 63.17 frames per second, an average accuracy of 94.68% , and an F1 score of 89.87% , demonstrating good real-time performance and accuracy. With no initial position deviation, the average error at different speeds was no more than 0.074 m, with a standard error of no more than 0.044 m. With initial position deviation, the average error was no more than 0.085 m, with a standard error of no more than 0.088 m. The soil drilling and detection sensor movement module operated stably, which was capable of drilling and loosening soil at different depths and detecting parameters. The research results can provide a technical solution for autonomous soil detection in the planting environment of fruits and vegetables.

Abstract:In recent years, drones have been explored as a potential tool for pollination support in hybrid rice breeding. However, the limited flight endurance of existing electrical agricultural drones necessitates frequent battery replacements, hindering the efficient utilization of the limited effective pollination time window and reducing pollination efficiency. To address the limited endurance of drones in pollination tasks and optimize operational parameters to enhance pollination efficiency, a supplementary pollination drone was designed with extendable flight duration, utilizing a time-sharing parallel power distribution scheme with multiple battery packs, achieving a maximum flight endurance of 50 min. To improve the pollination effectiveness of the prototype, a numerical simulation of the downwash airflow generated by the rotors was conducted by using the Lattice Boltzmann method (LBM). The optimal flight parameters of the prototype were found to be a speed of 4.5 m / s and an altitude of 2 m above the male parental canopy. Field experiments were conducted to validate the prototype’s pollination effectiveness and the optimal flight parameters by comparing three drones: the prototype, a quadrotor, and a hexacopter. Data were collected in four dimensions: average pollen grain count per single field of view, fruiting rate, yield, and endurance time for standardized deviation analysis. Results showed that the flight endurance (42 min), the pollination efficiency (10.5 hm 2 per flight), the averaged pollen grain count (6.98 grains per view, meeting the agronomic requirement of at least 3 grains ), the yield ( 1 996.5 kg / hm 2 ) and the comprehensive score of designed drone were better than two comparison drones. The research result may serve as a reference for enhancing the efficiency of drone supplementary pollination in hybrid rice breeding.

Abstract:Aiming to address the challenges of high spectral data redundancy and the limited accuracy of traditional machine learning models in identifying cotton Verticillium wilt severity levels, Nano Hyperspec hyperspectral cameras were mounted on drones to collect hyperspectral images of cotton fields. The spectral response characteristics of cotton canopies to different severity levels of Verticillium wilt were analyzed. An optimal vegetation index combination was identified and used to establish a monitoring model suitable for severity classification. This approach enabled precise classification of Verticillium wilt severity levels. The minimum redundancy maximum relevance algorithm was applied to rank the importance of features among 17 vegetation indices and 270 spectral bands. Features selected by this algorithm were incrementally grouped and input into an eXtreme gradient boosting model. This process determined the vegetation indices and spectral bands most strongly correlated with Verticillium wilt severity levels. A Transformer FNN ( feedforward neural network) classification model was then developed. Vegetation indices and spectral features were used as inputs to this model for classification. The classification accuracy of vegetation indices and spectral features in identifying Verticillium wilt severity levels was compared. Additionally, classification models based on back propagation neural network (BPNN), Transformer, and support vector machines ( SVM) were constructed. The accuracy of these models was validated and analyzed. The results showed that the optimal vegetation index combination for Verticillium wilt severity classification was MSR and TVI. The optimal spectral band combination included 430 nm, 439 nm, 488 nm, 566 nm, 697 nm, 722 nm, 742 nm, 764 nm, 769 nm, 782 nm, 822 nm, 831 nm, 858 nm, 873 nm, 878 nm, 893 nm, 909 nm, and 985 nm. Using the Transformer FNN model, the overall classification accuracy based on vegetation indices reached 95.6% . This represented a 6.2 percentage points improvement compared with the accuracy achieved by using spectral features, which was 89.4% . For vegetation indices, the Transformer FNN model achieved a classification accuracy of 95.6% . This was 11.2 percentage points higher than the accuracy of the BPNN model, 17.2 percentage points higher than that of the Transformer model, and 30.8 percentage points higher than that of the SVM model. The research proposed a high-accuracy monitoring method for cotton Verticillium wilt based on vegetation indices. It provided an effective approach for large-scale and precise monitoring of cotton Verticillium wilt.

Abstract:The leveling of paddy fields is an indispensable part of rice production. Aiming to improve the adaptability of existing leveling machines to the bumpy hard bottom layer of paddy fields and further enhance the operational performance, the supported paddy-field leveling implement was developed. Firstly, according to the working principle of the leveling shovel, the height adjustment mechanism of the leveling shovel was kinematically analyzed. On this basis, the leveling shovel and its height adjustment mechanism were designed. Secondly, the supported rod structure was designed, and the influence of the supported rods on the motion characteristics of the leveling shovel was studied. Finally, a comparison test with / without supported rods and a paddy-field leveling test were carried out. The results of the comparison test showed that the amplitude of height change of the leveling shovel with the supported rods was reduced by more than 15% throughout the test, and reduced by more than 30% in the convex position of the field surface. Meanwhile, the number of height changes of the leveling shovel was reduced. It showed that the leveling shovel with supported rods was more conducive to the height control of leveling shovels and was more suitable for the operation of bumpy hard bottom layer of paddy fields. The leveling test results for 0.21 hm2of paddy field showed that the standard deviation ( Sd ) of the topography height was 21.66 mm, and the proportion (ρ) of points ( | hi - h | ≤30 mm, where hi indicated the height of each measuring point, and h indicated the average height of all measuring points. )was 86.54% . The test results for two paddy fields with a total area of 1.89 hm2showed that the standard deviations Sd for the two fields were 26.02 mm and 27.43 mm, and the proportionsρ were 80.53% and81.03% , respectively. After leveling, the standard deviation Sd in all test fields was less than 30 mm, and the proportionρ was higher than 80% , which met the requirements of paddy-field leveling, and verified the structural advantages and design validity of the supported paddy-field leveling machine. This can provide equipment support for mechanized leveling of paddy fields with bumpy hard bottom layer.

Abstract:In view of the current 2BDQ maize delta-row seeder high-speed operation (12 ~ 16 km / h) due to the airflow wrapped seeds into the bilateral seed guide device speed is too fast to produce a violent collision, so that the bilateral seed staggered uniformity was decreased, resulting in the problem of low forming rate of the falling seed, a positive pressure air guide groove type seed guide device was designed. The structure of the seed guiding device and the principle of the upper and lower combined seed guiding groove were expounded. The mechanical model of the introduction and attachment process of corn seeds on the guide groove was established, and the relationship between the seed constraint state and the structural parameters of the guide groove during the seed guiding process was explored. The DEM-CFD coupling simulation method was used to analyze the speed trajectory curve of corn seed introduction to determine the width ratio of the upper guide groove and the lower guide groove, and the range of the upper guide groove slotting height, the lower guide groove width and the positive pressure value of the conveying airflow was determined by single factor test. Taking the range of single factor test results as the level value of test factors, the qualified index of glyph and the coefficient of variation of seed introduction consistency as the test indexes, the Box-Behnken design bench test of three factors and three levels was carried out by using JPS-16 high-speed seeding performance test bench, and the optimal parameter combination of seed introduction device was obtained and verified by test. Four kinds of seeds were selected to carry out the suitable sowing test of the device. The results showed that the optimal parameter combination was the upper guide groove slotting height of 354.491 mm, the lower guide groove width of 2.245 mm, and the positive pressure value of the conveying air flow of 0.739 kPa. Under the verification test, the qualified index of the parameter combination was 91.791% , and the variation coefficient of the seed guiding consistency was 6.111% , which was consistent with the optimization results. In the field test, the qualified index of the glyph was not less than 85.1% at working speed of 12 ~ 16 km / h.

Abstract:Aiming at the existing direct seeding equipment in the cotton area of the Yangtze River Basin, which has poor imitation effect, low seeding depth qualification index, and unsatisfactory consistency of seeding depth, an independent imitation cotton direct seeding machine was designed. The discrete element method and multi-body dynamics coupling method were used to simulate and analyze the operation process of imitation furrowing, and the key structure and working parameters affecting the depth of seeding furrow were determined. A one-factor quadratic orthogonal rotational combination test was conducted with the qualified index of furrowing depth and the coefficient of variation of furrowing depth consistency as evaluation indexes, and the optimal parameter combinations were determined to be the spring stiffness of 15 N/ mm and the operating speed of 0.8 m / s. Under the conditions of optimal parameter combinations, the qualified index of furrowing depth was 86.7% and the coefficient of variation of furrowing depth consistency was 11.6% . The results of field test under the optimal parameter combination showed that the qualified index of furrowing depth was 84% , the coefficient of variation of furrowing depth consistency was 12.80% , and the relative errors of field and simulation of qualified index of furrowing depth and coefficient of variation of furrowing depth consistency were 3.21% and 9.38% . Optimized parameter combinations and other combinations of field test results were as follows: open furrow depth qualified index was increased by not less than 6 percentage points, open furrow depth consistency coefficient of variation was decreased by not less than 1.69 percentage points, and the field emergence effect can meet the requirements of the Yangtze River Basin cotton area live operation.

Abstract:In the fragmented and small-scale fields of southern hilly regions, field-based mud collection is crucial for rice seedling cultivation. While mechanized mud collection can reduce costs, save time and labor, and improve seedling adaptability, challenges such as the hard underlying layer and uneven mud surface in paddy fields complicate dynamic adjustment of operational parameters. This results in unstable mud collection quantities and difficulties in achieving continuous field-based mud collection and seedling cultivation. A control system model for mud collection was developed by using Matlab / Simulink, targeting mud mass as the control objective, with mud collection depth and spiral conveyor speed as controlled variables, employing a fuzzy PID control approach. Simulation results showed that at a forward speed of 0.15 m / s, the mud mass in the storage tank was stabilized within 18 s, with an overshoot of 0.125% . To validate the system’s performance, a test rig equipped with displacement, pressure, and tilt sensors, and a PLC1200 controller was constructed. Experimental results indicated that the maximum standard deviation of mud mass was 0.46 kg, with a minimum of 0.21 kg;the maximum and minimum mud masses in the storage tank were 8.8 kg and 7.1 kg, respectively. The system effectively met field operational requirements, with no mud accumulation observed at the front end of the spiral conveyor. The successful implementation of this control system enabled more efficient and simplified rice cultivation. This system can provide a promising solution for improving agricultural productivity in fragmented and challenging terrains.

Abstract:The failure of nozzle performance is currently the most common and critical issue in plant protection operations. It directly affects the efficacy of pesticide spraying, consequently, has a significant impact on the quality and efficiency of plant protection tasks. Aiming to develop a nozzle performance failure detector for sprayers, enabling accurate evaluation of nozzle failure before field operations and ensuring efficient and precise plant protection. The system used an STM32 microcontroller as the core processing unit, integrating high-precision flow and pressure sensors to collect real-time data from the nozzle tip. The data was transmitted via Bluetooth to the LabVIEW platform, facilitating efficient, visual data analysis, early warnings, and storage to meet real-time nozzle performance monitoring requirements. The experimental results showed that for failed nozzle pressure detection, the maximum pressure measurement error was 2.174% , and the maximum flow measurement error was 1.936% . During the simultaneous performance testing of multiple nozzles, the maximum relative pressure measurement error was 1.515% , with an average error of 0.887% , while the maximum relative flow measurement error was 2.061% , with an average error of 1.151% . During field validation tests of multiple nozzle performances, the maximum relative pressure measurement error was 1.990% , with an average error of 1.629% . The maximum relative flow measurement error was 2.713% , with an average error of 2.014% , meeting the requirements for nozzle performance testing. This device provided a reliable and efficient intelligent method for detecting nozzle performance failures, meeting the technical requirements for pre-operational testing in standardized plant protection tasks.

Abstract:The threshing cylinder of the crawler-type grain combine harvester tends to become easily clogged under high feeding rate conditions, which leads to severe fragmentation of threshed materials, resulting in an accumulation of residuals on the cleaning screen, which affects grain sieving efficiency, increases cleaning losses, and significantly elevates power consumption. By optimizing the structure and arrangement of the rasp bars and threshing elements, a spiral interlaced arrangement threshing cylinder was designed. This design reduced fluctuations in cylinder load and decreased the impact intensity of the threshing elements on the material, thereby minimizing material breakage and fragmentation. Through discrete element simulation and bench tests, the distribution patterns of threshed grain were analyzed. A comparative analysis was conducted on the migration speed of stalks within the original cylinder and the designed cylinder, as well as the torque exerted on the cylinder. The simulation results indicated that at rotational speed of 750 r/ min, the migration speed of stalks within the designed cylinder was increased by 23.5% , facilitating the rearward conveyance of materials. Additionally, the average torque of the designed cylinder was decreased by 29.2% , with reduced torque fluctuations, thereby improving the stability of the threshing operation. Field comparison test results showed that the designed cylinder significantly reduced stalk breakage and fragmentation, thereby decreasing the load on the cleaning screen and increasing the probability of grain passing through the vibrating screen. The cleaning loss rate was reduced by an average of 22.8% . Moreover, under the same operating conditions, the combine harvester equipped with the designed cylinder demonstrated an average fuel consumption reduction of 18.1% . Combining the results of the simulation experiments and field comparison tests, it was evident that the designed threshing cylinder can reduce operational losses, enhance working efficiency, and lower threshing power consumption. It effectively met the requirements for low-loss, high-efficiency operation under high feeding rate conditions.

Abstract:In order to improve the detection accuracy and applicability of wheat moisture content detection device for combined harvester based on dielectric properties, the wheat moisture content prediction model was established based on GA-BP method. Focusing on three varieties of wheat, namely “Jingdong 22”, “Shumai 1958” and “Womai 33”. The measured range of wheat moisture content was 8.41% to 21.6% , with the detection temperature ranging from 5℃ to 40℃ and the bulk density ranging from 714.44 kg / m 3 to 777.58 kg / m 3 for wheat dielectric constant. The experiment results indicated that at constant temperature conditions, higher bulk density corresponded to a larger dielectric constant. Similarly, under consistent bulk density, the dielectric constant was increased with the increase of temperature and moisture content. To establish the relationship between dielectric constant, temperature, bulk density, and wheat moisture content, a genetic algorithm optimized back propagation neural network (GA-BP) with 150 samples in the calibration set and 42 samples in the prediction set was established. The model, with a 3-5-1 structure, a maximum iteration of 1 000 times, and a learning error threshold of 1×10 - 6 , demonstrated high detection accuracy and stability. The verification set R 2 , RMSE, and MAE values were 0.996, 0.241% , and 0.189% , respectively, while the prediction set returned values were 0.993, 0.295% , and 0.189% . These results underscored the model’s efficacy in providing a method for online moisture content detection in wheat of different varieties.

Abstract:Aiming at the problem that the end-effector picking force is poorly matched with fruits and vegetables,resulting in a high damage rate and low versatility,a contact force optimization method for non- destructive grasping of spherical fruits and vegetables based on dexterous hands was proposed. The separation scheme of spherical fruit and vegetable picking was analyzed, and a frictional non-destructive point contact model was established. Peaches in the picking period were taken as the research object,and the static friction coefficient and non-destructive contact parameters between the fingertip material and peaches were experimentally determined. The kinematic model and Jacobian matrix of the multi-finger dexterous hand were established based on the screw theory, the force balance constraint of fruit and vegetable picking was analyzed,and the force-position mapping model of joint contact point fruit and vegetable was constructed. A grasping planning method for dexterous hands based on minimum force was proposed,and the solution analysis was carried out using the picking of peaches as an example,and the optimal solution of the planning model was obtained based on the NSGA II algorithm. Field peach picking experiments showed that this method can achieve stable grasping and picking of peaches with a radius of 2.6 ~ 8 cm,effectively avoiding damage to fruits and vegetables during the picking process. The average absolute contact force error between the actual measured value and the theoretical calculated value was 0.39 N, the picking success rate was 92% , the non-destructive rate was 97.8% , and the average picking time was 10.3 s,which met the needs of dexterous hands for non-destructive and stable picking.

Abstract:Aiming at the problem of lacking dedicated harvesting equipment and having low clean harvest rates for small-seeded alfalfa in mechanized harvesting both domestically and internationally, a half-feed alfalfa seed harvester combining mechanical combing separation with negative pressure airflow suction and flow enhancement was designed. Based on the biological characteristics of alfalfa plants and planting agronomy, a mechanized harvesting scheme was proposed, including mechanical combing separation, airflow adsorption and transportation, and enhanced airflow through a fan for small-seeded alfalfa. According to the mechanism of loss suppression during the alfalfa seed harvesting process, key components of the combing platform were designed. For uncombed losses during the alfalfa seed harvesting process, a combing structure was designed to perform combing separation of highly dispersed seed pods. For seed fall and splash losses, a drum-wall combination airflow throat and negative pressure fan system were designed to adsorb and transport the seeds and pods. On this basis, a kinematic analysis of the combing detachment process was conducted. It was determined that the effective range for the comb detaching speed ratio was between 36 and 50, with a comb detaching drum radius of 290 mm. The critical parameters affecting incomplete combing losses were identified as the rotational speed of the comb detaching drum, the center-to-ground height, and the forward speed of the harvesting platform. Using CFD simulation combined with CCD orthogonal experimental methods, the interaction effects of the combing drum rotational speed, airflow throat dimensions, and negative pressure fan speed on the platform airflow field were explored. Using response surface methodology, the optimized values for the airflow throat dimensions, combing drum rotational speed, and negative pressure fan speed were determined to be 83 mm, 1 000 r/ min and 1 450 r/ min, respectively. To verify the effectiveness of the whole machine and the combing platform function, field trials of the alfalfa seed harvesting prototype were conducted. Results showed that when the height of the combing drum center above ground level was 430 mm and the machine’s travel speed was 0.68 m / s, there was virtually no uncombed seed pod after combing, the alfalfa seed clean harvest rate reached 83.4% , and the operational efficiency was 0.73 hm 2 / h, meeting the mechanized harvesting index requirements for alfalfa seeds, providing a reference for the design of small-seeded seed harvesters.

Abstract:Aerated drip irrigation can effectively improve crop production efficiency and crop yield. The formation of a large number of uniform smaller micro-bubbles in the aerated system and the improvement of dissolved oxygen concentration in irrigation water are the key to the long distance uniform water and air transport of aerated drip irrigation. The working conditions of a micro- and nano-bubble generator designed was optimized based on the principle of cyclonic shear fragmentation. Through one-way test analysis, high-speed photography and dissolved oxygen meter, the influence of system operating pressure and air inlet on bubble characteristics ( count, diameter and uniformity) and dissolved oxygen was studied. Finally, combined with the response surface RSM optimization test, the combination of operating pressure and air inlet under the optimal aerating effect of the device was determined. The results showed that the generation of bubbles and dissolved oxygen effect were better when the pressure was controlled within 300 ~ 400 kPa. In the range of 1% ~ 5% of air inlet, the best diameter distribution and the smallest average diameter were obtained with 1% of air inlet, while the rate of bubble generation was higher with 5% of air inlet. Meanwhile, the dissolved oxygen test showed that increasing the air intake significantly increased the dissolved oxygen efficiency as well as the final oxygen content. Through the response optimization solution, it was concluded that the plant could operate at a pressure of 350 kPa and an air intake of 2% with the lowest energy consumption. Under this condition, the final diameter of the small bubbles generated by the device was 68.6 μm, the rate of oxygen enhancement was about 0.81 mg / (L·min), the net increase in dissolved oxygen after stabilization was 1.08 mg / L, and the bubble dissipation time was up to 323 s.

Abstract:Accurate crop map is critical for agricultural monitoring and related decision-making. Although many algorithms were adopted for crop classification, the performance of individual classifier varied with study area and data used. Aiming to address this issue, a novel ensemble learning (EL) framework was developed, which adopted objective weighting methods to assign weights to five widely used classifiers, including K-nearest neighbor, support vector machine, random forest, back propagation neural network and convolutional neural network. Four study sites in the United States were selected to examine the performance of the proposed EL framework. Time series of normalized difference vegetation index derived from Sentinel 2 multispectral instrument images were used as the input features for crop classification. Two modified objective weighting methods, termed modified entropy method (mEN) and modified coefficient of variation method ( mCV), were proposed to determine the weights of base classifiers. The probability outputs of base classifiers were combined with weights to determine the final label. The results showed that weights assigned by modified weighting methods were more reasonable than those by original weighting methods in multi-classifier ensemble. The combination of mEN and mCV (mEN mCV) further amplified the weight difference of base classifiers, and achieved an improved performance than single weighting method. Compared with five base classifiers, overall accuracy achieved by mEN mCV with F1-score as input ( mEN mCV F) was increased by 1.12 ~ 6.45 percentage points, 0.75 ~ 3.98 percentage points, 0.45 ~ 2.70 percentage points and 1.15 ~ 2.50 percentage points at four sites, respectively. The advantage of the proposed EL framework over unweighted ( i. e. , majority vote, probability fusion) and accuracy-weighted methods was that both classification accuracy and stability of base classifiers were considered, thus resulting in a higher performance. These results indicated that the proposed EL framework had potential in improving the accuracy of crop classification.

Abstract:Aiming to address the challenges of real-time and accurate detection of the seeding quality of chili seedling trays, considering the computing power limitations of edge devices, a lightweight detection algorithm YOLO v8n SCS (YOLO v8n improved with StarNet, CAM, and SCConv) was proposed based on YOLO v8n. Meanwhile, the technical strategy of “ single-cell training + whole-tray detection冶 was adopted to reduce training costs and improve training efficiency. Firstly, the StarNet lightweight network and the CAM ( Context augmentation module) were used as the backbone network to achieve multi- receptive field information fusion of deep features while reducing the complexity of the model. Secondly, the spatial and channel reconstruction convolution ( SCConv) was employed to optimize the bottleneck structure of the intermediate layer cross stage partial network fusion (C2f) module to enhance the feature extraction ability of the module and improve the model inference speed. Finally, the P2 detection layer was fused and the detection heads were reduced to one to enhance the model’s detection performance for small targets. The results showed that the YOLO v8n SCS model had a parameter quantity of 1.2 × 10 6 , a memory occupation of 2.7 MB, and a computation amount of 7.6 伊 10 9 FLOPs. On the single-cell dataset of the seedling trays, its mAP50 was 98.3% , mAP50 - 95 was 83.8% , and the frame rate was 112 f / s. Compared with the baseline model YOLO v8n, the parameter quantity was reduced by 62.5% , mAP50 was increased by 2.5 percentage points, mAP50 - 95 was increased by 2.1 percentage points, the floating-point operations were reduced by 1.3 × 10 9 , and the frame rate was increased by 23.1% . In the whole-tray detection task, its detection frame rate was 21 f / s and the detection accuracy rate was 98.2% . Compared with the baseline model, the detection frame rate was increased by 8.2% and the accuracy rate was increased by 1.1 percentage points. For 72-cell seedling trays with a seeding speed within 800 trays/ h and 128-cell seedling trays with a seeding speed within 600 trays/ h, its average detection accuracy was above 96% , and the detection errors of single-seed rate, reseeding rate, and miss-seeding rate were less than 1.8% . This study achieved a good balance between performance and computational cost, reduced the computing power requirements for deploying edge devices, met the online detection needs for the seeding quality of chili seedling trays, and provided key technical support for the intelligent upgrade of the seedling production line.

Abstract:Laser flower thinning technology, as an emerging and promising technology in the field of smart orchard management, still faced critical challenges in optimizing laser parameters and achieving precise apple flower detection. Aiming at the optimization of key parameters in apple blossom laser flower thinning technology, a laser flower thinning test stand was designed, and the height of the test stand, the laser striking time and the PWM duty cycle were optimized by the orthogonal test method to obtain the optimal parameter combinations: a laser height of 20 cm,a striking time of 10 s and a laser power(PWM duty cycle) of 50% would achieve the best flower thinning effect. For apple flower identification and localization in laser flower thinning, the lightweight anl targeted-you only look once (LT-YOL0)apple flower detection model was proposed, the DPRVITBlock module based on ViTBlock and the DPRVBC2f module based on the C2f module were designed, and the ELA attention module of the DPRVBC2f module was added,which was applied in the feature extraction of the detection backbone and the detection head to enhance the apple blossom detection performance, validation focused on the accuracy,recall and average mean of the model were 83.16%,82.15% and 87.47%,respectively,compared with that of the YOLO v8 model it was improved by 5.04 percentage points,2.12 percentage points and 2.15 percentage poin1ts,respectively. The model size was 5.26 MB and the detection speed was 128/s,which met the accuracy and real-time requirements for use. The research result can provide a scientific basis for the further optimization and intelligent application of apple flower thinning technology.

Abstract:In real-world scenarios where pest data is scarce, existing pest image recognition methods are prone to overfitting, resulting in insufficient model expressiveness. To address this issue, a novel few-shot field pest image classification method that integrated metric learning with transfer learning was proposed. Firstly, the ECA Pyramid ResNet12 model was pretrained on the mini ImageNet dataset. Subsequently, PN was chosen as the classifier, and cosine similarity was selected as the distance metric. The ECA channel attention mechanism was then incorporated into the metric module to enhance pest image feature representation by capturing inter-channel dependencies, with a kernel size of 3. Additionally, a feature pyramid structure was employed to capture the local and multi-scale features of pest images. After evaluating different pooling combinations, the 2 × 2 + 4 × 4 pooling combination was selected. Finally, a meta-dataset comprising 20 self-built categories of Lepidoptera pest images was utilized for meta-training and meta-testing of the model. Experimental results demonstrated that under 3-way 5-shot and 5-way 5-shot conditions, the proposed method achieved accuracy rates of 91.16% and 87.26% , respectively, surpassing the most relevant works of the past two years, SSFormers and DeepBDC, by 4.58 percentage points and 1.35 percentage points. The proposed model effectively enhanced the feature representation of target images in few-shot learning, providing a methodological reference for the automatic identification of field pests in data-scarce scenarios.

Abstract:With the rapid development of modern pig breeding industry, the demand for precise recognition of pig behaviors is increasing. Aiming to address the issues of diversity of pig behaviors, similarity of features, mutual occlusion and stacking, a pig behavior recognition method based on the improved YOLO v8 model was proposed. Firstly, the ConvNeXt V2 was introduced as the backbone feature extraction network to enhance the ability to extract semantic information of the detection target. Secondly, the bi-directional feature pyramid network (BiFPN) was added to the feature fusion network to enhance the feature fusion ability of the model. Thirdly, combined with the CARAFE up-sampling operator, the feature extraction ability of the model in the process of behavior recognition was further improved. Finally, the WIoUv3 was used as the loss function to optimize the detection accuracy of the model. The experimental results showed that the precision rate, recall rate, mean average precision and F1 value of the improved model reached 89.6% , 88.0% , 91.9% and 88.8% , respectively. Compared with TOOD, YOLO v7 and YOLO v8 models, the mean average precision was increased by 10.9, 6.3 and 3.7 percentage points, respectively, which significantly improved the accuracy of pig behavior recognition. The ablation experiments showed that all the improvements improved the recognition performance of the model, and the ConvNeXt V2 backbone feature extraction network had the most obvious improvement effect on the model. In summary, the CBCW-YOLO v8 model demonstrated excellent overall performance in pig behavior recognition tasks and provided powerful technical support for pig health management and disease early warning.

Abstract:In response to the dense and randomly positioned growth of citrus branches, leaves, and fruits, to achieve safe interactive picking of interlaced and overlapping branches and fruits, a citrus obstacle avoidance picking method was proposed. To enhance the positioning accuracy and picking efficiency, the hand-eye calibration problem was transformed into solving the equation T1X = XT2 , completing the calibration from the camera coordinate system to the base coordinate system of the robotic arm. For the citrus open-center tree shape in the natural environment, tree skeleton extraction based on point cloud density was conducted, and noise reduction processing of the branch and trunk point clouds was performed through the point density threshold method to increase the operation speed. The octree map method was utilized to construct the obstacle map of branches and trunks, and the hierarchical bounding box method was employed to fit the robotic arm and carry out collision detection. With the objective of time optimization, an improved RRT connect obstacle avoidance planning algorithm that conforms to the agricultural requirements of picking was proposed. Target bias was introduced to the RRT connect algorithm for optimizing and guiding the sampling points. To verify the feasibility of this obstacle avoidance method, taking the citrus orchard with standard dwarf and dense planting cultivation as the research object, an obstacle avoidance system for the picking robot was established. Multiple sets of obstacle avoidance picking experiments were respectively conducted for citrus fruits growing inside the fruit tree and close to the trunk in the natural environment. The experimental results indicated that the obstacle avoidance movement time for fruits growing close to the trunk was 9.5 s, and the success rate of obstacle avoidance picking was 91% ;for fruits growing inside the fruit tree, the obstacle avoidance movement time was 10.5 s, and the success rate of obstacle avoidance picking was 88% .

Abstract:Global climate change will have a huge impact on future food production. Water and temperature are the most important environmental factors in the growth of winter wheat and summer maize, which significantly affect their yield. Based on the irrigation experimental data of winter wheat and summer maize in Baoding Irrigation Experimental Station in North China Plain from 2006 to 2015, the AquaCrop model was calibrated and validated, offering crop growth process simulations following local conditions. Being similar in structure of the four typical water production functions ( Blank model, Stewart model, Jensen model, Minhas model), the water-heat production functions were set up between accumulated temperature, water consumption, and yield at each growth stage of winter wheat and summer maize. Using the data from the sixth version of the model for interdisciplinary research on climate (MIROC6) of the commentary on the coupled model intercomparison project (CMIP6), the daily rainfall and temperature data were downscaled to consider future climate change, including low carbon emission forcing scenario SSP1 RCP2.6 and SSP4 RCP3.4, medium carbon emission forcing scenario SSP2 RCP4.5, medium to high forcing emission scenario SSP3 RCP7.0 and high forcing scenario SSP5 RCP8.5. On this basis, the yields and their changes for winter wheat and summer maize in 2024—2064 were obtained and analyzed by the presented water-heat production function. Results showed that the AquaCrop model made good performances to simulate the growth process of winter wheat summer maize in this region after its calibration and verification by using ten years of irrigation test data. Among the four kinds of water-heat production functions constructed by the verified AquaCrop model simulation data, the Jensen type function had the highest output simulation accuracy. According to the water-heat production function, winter wheat was most sensitive to water during heading filling stage, and accumulated temperature during greening jointing stage had the most obvious effect on yield. Summer maize was most sensitive to water in jointing and heading period, and the accumulated temperature in this period had the most obvious effect on yield. Under the emission scenarios of SSP1 2.6, SSP2 4.5, SSP3 7.0, SSP4 3.4, and SSP5 8.5 in the five future climates, the potential yield of winter wheat tended to fluctuate, but it was higher than the current average potential yield. By the 2050s, the average potential yield of winter wheat would be 6.07 t / hm 2 , 6.26 t / hm 2 , 6.93 t / hm 2 , 5.74 t / hm 2 , and 5.95 t / hm 2 , respectively. The overall potential yield of summer corn was on the rise, and by 2050s, the average annual potential yield of summer corn would reach 9.27 t / hm 2 , 9.20 t / hm 2 , 9.05 t / hm 2 , 9.10 t / hm 2 , and 9.24 t / hm 2 , respectively. Overall, winter wheat and summer corn were more suitable for growth and development under SSP3 7.0 and SSP1 2.6 scenarios, respectively. Considering the hydrothermal conditions, the potential yield of winter wheat fluctuated down under the five climate scenarios, while the potential yield of summer maize showed an overall upward trend. Supplementary irrigation can bring about 70% of contribution rate to the potential yield of winter wheat. The contribution rate of rainfall during the growing period to the potential yield of summer maize was about 94% . The results can be used to evaluate the change of crop grain yield in this region under future climate change, and provide theoretical basis and technical support for the national strategy of ensuring food security.

Abstract:When the irrigation area uses surface water as its water source and has sufficient supply, the optimization of the micro-irrigation pipe network requires determining the reasonable control area and layout form of individual micro-irrigation systems and conducting optimized design. The current pipe network optimization method is not applicable to the optimization of random water supply micro-irrigation pipe network systems under this situation. In order to achieve the synchronization optimization of layout and pipe diameter for random water supply micro-irrigation pipe network systems without limiting the area, and propose the optimal design criteria, a mathematical model for optimizing micro-irrigation main pipe network was established, and a model solution method based on a hybrid encoding genetic algorithm was provided. The correlation between the annual cost per unit area of the main pipe network and its layout was analyzed and the influence of lateral diameter, emitter design discharge, emitter spacing, and emitter emission exponent on the unit area annual cost of main pipe network and micro-irrigation pipe network system was analyzed. The results of example calculations showed that the unit area annual cost of the micro-irrigation pipe network system optimized based on the model with maximum control area for lateral double-sided layout was the lowest, while the unit area annual cost of the micro-irrigation pipe network system optimized based on the model with minimum unit area annual cost for lateral one-sided layout was the highest. Compared with the model with the minimum unit area annual cost for lateral one- sided layout, the percentage reduction in unit area annual cost of the micro-irrigation pipe network system optimized based on the model with maximum control area for lateral double-sided layout was 4.46% to 15.74% . In actual engineering projects, using smaller lateral diameters, emitter design discharges, emitter emission exponents, larger spacing between emitters, optimizing subunit based on the model with maximum control area for lateral double-sided layout, can effectively reduce costs. The results can provide a basis for the optimization design of random water supply micro-irrigation pipe network systems without limiting the area.

Abstract:The soil erosion is severe in the cultivation of black soil slope farmland along the slope, and the soil organic carbon content is decreasing year by year. Aiming to investigate the effects of different soil and water conservation tillage techniques on soil carbon balance in maize fields on black soil slopes, a runoff plot experiment was conducted. Conventional tillage ( CK) was used as a control, and six comprehensive management techniques for slope farmland were set up, including contour tillage (CT), ridge oriented field (RT), deep scarification tillage (ST), contour tillage + ridge oriented field (CR), contour tillage + deep scarification tillage ( CS), and ridge oriented field + deep scarification tillage (RS). The dry matter mass and carbon content of various organs of maize harvested under different soil and water conservation tillage techniques were observed, the incremental of soil carbon storage, CO2 emission carbon, and soil erosion loss carbon in maize fields were synchronously monitored, the net primary productivity (NPP) and net soil carbon income (NSCB) of maize fields were estimated. The results showed that the total carbon sequestration of maize plants was 10 201.93 kg / hm 2 ~12 357.34 kg / hm 2 , and the carbon sequestration of each organ in descending order was as follows: grain, stem sheath, leaf, ear axis, and root. The NPP of CT, RT, CR, CS, and RS treatments was significantly higher than that of CK treatment ( P < 0.05 ) , which was increased by 20.28% , 11.55% , 21.68% , 16.55% , and 7.95% , respectively. However, there was no significant difference between the NPP of ST treatment and CK treatment (P > 0.05) . Ridge oriented field, deep scarification tillage, and their combination mode would increase the total CO2 emissions of maize fields, while contour tillage and contour tillage + ridge oriented field can reduce the total CO2 emissions. The CT treatment had the highest NSCB, at 1 402.29 kg / hm 2 . The NSCB of CT, RT, CR, and CS treatments were significantly higher than those of CK treatment (P < 0.05) , while the NSCB of ST treatment were significantly lower than those of CK treatment (P < 0.05) . Overall, the contour tillage management technology model was the best, with the strongest carbon sequestration capacity. The research results can provide theoretical reference and technical support for the protection and management of sloping farmland in the black soil area of Northeast China.

Abstract:The process of lateral seepage constitutes a crucial component of the water cycle in paddy fields. Previous studies have focused on the effects of control regulation field water layer management on lateral seepage in paddy fields. However, less attention has been paid to the lateral seepage of paddy field water under controlled irrigation, and there is a paucity of research findings regarding seepage in paddy fields under coupled controlled drainage. In order to optimize the water management of paddy field, a two-year in-situ field test was conducted, and diverse irrigation and drainage treatments were implemented. The characteristics of lateral seepage in paddy fields subjected to controlled irrigation and drainage regulation were systematically analyzed, and the response mechanism of lateral seepage to such control regulation was revealed. The findings were as follows: compared with free drainage, controlled drainage inhibited the change of soil moisture content in the field bund. Compared with flooding irrigation, the fluctuation of soil moisture content in the field bund under controlled irrigation field bund was more intense. Controlled irrigation and drainage regulation had a direct impact on the characteristics of lateral seepage in the field bund canal transition zone, and significantly reduced the total amount of lateral seepage. Controlled irrigation and drainage, as opposed to flooding irrigation and free drainage, respectively reduced the peak and mean values of lateral seepage intensity, and significantly reduced the total amount of lateral seepage in the paddy field, during the two-year experiment period, the controlled irrigation and drainage treatment of the rice fields resulted in a reduction of 63.49% compared with the free irrigation approach. Both paddy field irrigation and drainage treatments exerted a considerable influence on lateral seepage intensity, with irrigation treatments demonstrating a stronger effect. Paddy field irrigation treatment and drainage treatment had a significant effect on the lateral seepage intensity, and the effect of irrigation treatment was stronger. Compared with flooding irrigation, controlled irrigation significantly increased the proportion of water seeping from the paddy field ridge channel area. To maximize irrigation effects, it was imperative to strengthen management of water lateral seepage within the fields during controlled irrigation implementation. Lateral seepage in paddy fields subjected to irrigation and drainage regulation primarily occurred in the field bund soil at a depth of 10 ~ 20 cm. Water percolation channels were presented at depths of 10 ~ 20 cm below the ground surface. By implementing effective waterproofing measures at this depth, it can significantly reduce the loss of field moisture. The research revealed the lateral seepage process and its responsiveness to soil moisture changes in the paddy field bund canal transition zone under irrigation and drainage regulation. The insights gained aim to serve as a reference for the efficient utilization and meticulous management of agricultural water resources.

Abstract:In order to explore the fertilization mode of emission reduction, superior yield and high efficiency, taking winter wheat (Xiaoyan 22) as the research object, the experiment set six treatments: no fertilization ( KB ), conventional fertilization ( NP ), and organic fertilizer substituting 25% (25% OF),50% (50% OF), 75% (75% OF), and 100% (100% OF) of inorganic fertilizer on equal nitrogen basis, was used to analyze the effects of different proportions of combined organic and inorganic fertilizer application on greenhouse gas emissions, yield, and nitrogen use efficiency in winter wheat farmland, put forward the combined organic-inorganic fertilization mode with the goal of emission reduction, superior yield and high efficiency. The results showed that in the two winter wheat growing seasons, the average cumulative CO2 emissions of 75% OF and 25% OF were the lowest, which was decreased by 7.62% and 15.31% compared with that of NP, respectively. The CO2 emission flux was increased in summer and decreased in winter. Among all fertilization treatments, NP had the highest CH4 absorption, while 100% OF had the lowest CH4 absorption. The absorption of CH4 in soil was decreased with the increase of the amount of organic fertilizer. The cumulative N2 O emissions of 75% OF and 100% OF were the lowest, which was decreased by 92.94% and 159.47% compared with that of NP, respectively. The global warming potential (GWP) and greenhouse gas intensity (GHGI) of 75% OF were the lowest, which was decreased by 27.19% and 41.38% compared with NP, respectively. Compared with NP, the winter wheat yield of 50% OF and 75% OF was increased by 15.78% ~ 17.73% and 18.64% ~ 23.07% , respectively, from 2018 to 2020. Compared with conventional fertilization, 75% OF significantly increased the nitrogen uptake and nitrogen utilization efficiency of winter wheat. Considering the economic and ecological factors, replacing 75% inorganic fertilizer with organic fertilizer can take into account the requirements of emission reduction, superior yield and high efficiency, and provide a reference for the better fertilization mode of winter wheat in Northwest China.

Abstract:Scene recognition could be used as an alternative for spatial positioning in greenhouse environments, and it was also one of the important functions of the visual system of intelligent agricultural machinery equipment. Addressing the issue that scene recognition paradigms based on feature clustering could not adapt to the recognition of greenhouse scenes with high dynamic changes and high similarity, a greenhouse scene recognition method based on deep feature aggregation was proposed. This method, grounded on a pre-trained visual transformer network, extracted local features from scene images. It applied the global receptive field characteristics of multi-layer perceptron, took into account the spatial relationships of local features, fused the local features of the images, and generated global descriptors for the scene images. With the goal of minimizing multi-similarity loss as the optimization objective, a greenhouse scene recognition model was constructed. The test results indicated that the R@ 1 ( top 1 recall rate), R @ 5, and R @ 10 of the model’s scene recognition reached 78.43% , 89.21% , and 92.47% , respectively, and it possessed high scene recognition accuracy. The proposed feature mixing method based on multi-layer perceptron was proven effective, with an improvement of 8.01 percentages in R@ 1 compared with that of feature aggregation using pooling operations. The model demonstrated a certain robustness to changes in lighting conditions, with the R@ 1 metric decreased by no more than 4.00 percentages under strong and weak lighting conditions compared with that under normal medium lighting conditions. Changes in camera angle and sampling distance also impacted the model’s recognition performance, with a decline of 6.61 percentages for angle changes within 20 degrees, and a drop of 17.87 percentages for distance changes within twice the original distance. Compared with the existing scene recognition benchmark methods, including NetVLAD, GeM, Patch-NetVLAD, MultiRes-NetVLAD, and MixVPR, the R@ 1 of proposed model was improved by 7.82, 6.59, 3.56, 4.14, and 1.88 percentages, respectively, demonstrating a significant performance enhancement on the greenhouse scene recognition task. The image global feature aggregation method based on multi-layer perceptron constructed was able to generate reliable global descriptors for greenhouse scene recognition, and exhibited robustness to changes in lighting, viewpoint, distance, and time. The research findings would provide technical references for the design of visual systems for intelligent agricultural machinery.

Abstract:As a tuber crop with the same origin as medicine and food, yam is becoming more and more favored by people. With the development of yam processing industry, rapid non-destructive testing and grading of yam quality is of great practical significance to the healthy development of the industrial chain. For the purpose of developing a multi-quality non-destructive rapid detection device for yam, based on the principle of visible / near-infrared local diffuse transmission, a special detection probe for yam was designed according to the special appearance characteristics of yam, and a hand-held multi-quality nondestructive testing device for yam was developed by designing the optical path through comparative experiments. The overall dimensions of the device were 150 mm × 80 mm × 150 mm and the weight was about 590 g. Based on the spectral information of 150 yams collected by the R&D device, the collected spectra were corrected by multiplicative scatter correction ( MSC ), and then the characteristic wavelengths were screened by the shuffled frog leaping algorithm (SFLA) to establish partial least squares regression (PLSR) prediction model of dry matter, starch and protein content of yams, the correlation coefficients of the validation set of dry matter, starch and protein were 0.965 3, 0.967 5 and 0.956 3, respectively, and the root mean square errors were 1.09% , 0.83% and 0.15% , respectively. Based on the Qt development tool, the real-time analysis and control software was written in C language, and the prediction model was implanted into the device for external verification. The dry matter, starch and protein contents of 50 yam samples not participating in the modeling were detected five times by using the R&D device, and the coefficients of variation were 1.0% ~ 1.2% , 1.5% ~ 1.7% and 1.4% ~ 1.6% , respectively. The absolute values of the maximum residuals of the dry matter, starch and protein of 50 yam samples were 1.83% , 1.64% and 0.26% , respectively. The results showed that the handheld yam multi-quality non-destructive testing device can meet the needs of real-time detection in the field.

Abstract:The stress behavior representation based on research on ammonia nitrogen stress behavior is the premise and basis for realizing the recognition of ammonia nitrogen stress of Epinephelus akaara. However, most of the existing methods rely on high-performance hardware, which is not conducive to the embedded deployment and application of behavior representation methods in aquaculture. Taking symptoms such as reduced activity and imbalanced body of Epinephelus akaara under stress environment into account, a behavior representation method was proposed to represent the ammonia nitrogen stress behavior of Epinephelus akaara based on lightweight detection and tracking algorithm. In the detection algorithm, GhostV2 convolution was firstly used to lighten the feature extraction network of YOLO v5s. Then asymptotic feature pyramid network was integrated into the neck of YOLO v5s to support direct interactive fusion of different dimensional features. The results of ablation and comparison experiments showed that the accuracy and recall rate achieved 94.3% and 89.5% and mAP@ 0.5 of the lightweight model was 96.2% , which was 1.6 percentage points higher than that of the original model while the model volume was about 60% of that of the original model. In the tracking algorithm, a lightweight ReID network was embeded into Ocsort and the appearance similarity matrix was introduced into the matching cost matrix in target association period. Comparison experiments showed that MOTA and IDF1 of improved tracking algorithm achieved 94.7% and 69.3% , which were 3.2 percentage points and 6.7 percentage points higher than that of the original Ocsort with YOLO v5s. Combined with the research on ammonia nitrogen stress behavior, average velocity and number of imbalanced Epinephelus akaara were selected to characterize the ammonia nitrogen stress behavior of Epinephelus akaara. The accuracy of identifying the behavior of Epinephelus akaara based on the characterization proposed method was 92.2% , which can accurately classify whether the Epinephelus akaara was under ammonia nitrogen stress environment. The lightweight characterization method can be deployed on Jetson Orin Nano embedded system, with an average speed of 6 f / s, providing technical support for efficient and real-time identification of ammonia nitrogen stress in aquaculture.

Abstract:The change of center of gravity position of mountain agricultural machinery seriously affects its stability, traction and obstacle crossing ability. Aiming at the problems of design theory of center of gravity control device of mountain agricultural machinery chassis and the lack of special experimental platform, an omnidirectional control experimental platform of center of gravity of mountain agricultural machinery chassis was designed. The theoretical analysis showed that the slope angle caused the shift of center of gravity of chassis of mountain farm machinery, which seriously affected the stability of contour line driving and longitudinal climbing performance. Therefore, the overall structure of the experimental platform was determined by considering the relationship between slope angle and center of gravity position. The experimental platform mainly included inclination simulation device and center of gravity adjustment device. Among them, the inclination simulation device adopted the synergistic effect of multiple electric push bars to realize the omni directional slope simulation of 0° ~ 15°, and the center of gravity adjustment device adopted the mode of “ H-shaped ” multi-slide combination to realize omnidirectional center of gravity adaptive adjustment, so as to realize the center of gravity adjustment function under different inclination conditions. The performance test results showed that the mean value of simulated slope and the relative errors between data median line and simulated angle were within 0.5° in horizontal, longitudinal and oblique slope simulation of 0° ~ 15°. In the test, the maximum error of center of gravity position was - 21.4 mm. The center of gravity adaptive control can be achieved on the horizontal and vertical slopes of 0° ~ 15° and the oblique slopes of 0° ~ 12°. The average error of the test was 2.6% , 3.4% and 5.9% , respectively, and the maximum error of the center of gravity adjustment was 6.7 mm, 7.3 mm and 10.8 mm, respectively. The adaptive control of center of gravity can also be achieved on horizontal and longitudinal simulated slopes of 0° ~ 15° and oblique simulated slopes of 0° ~ 12°. The average test errors were 5.4% , 6.5% and 9.7% , respectively, and the maximum errors of center of gravity adjustment were 9.7 mm, 10.3 mm and 15.8 mm, respectively, which met the basic design requirements. The inclination simulation and center of gravity adaptive control method proposed can provide reference for the research of center of gravity control theory of agricultural machinery chassis in hills and mountains.

Abstract:Estimating trawler fishing effort plays a critical role in characterizing marine fisheries activities, quantifying the ecological impact of trawling, and refining regulatory frameworks and policies. Understanding trawler fishing inputs offers crucial scientific data to support the sustainable management of offshore fishery resources in China. An XGBoost algorithm was introduced and optimized through Harris Hawks Optimization (HHO), to develop a model for identifying trawler fishing behaviour. The model demonstrated exceptional performance, achieving accuracy, sensitivity, specificity, and the Matthews correlation coefficient of 0.971 3, 0.980 6, 0.963 2, and 0.942 5, respectively. Using this model to detect fishing activities, the fishing effort of trawlers from Shandong Province in the sea area between 119°E to 124°E and 32°N to 40°N in 2021 was quantified. A heatmap depicting fishing effort, generated with a spatial resolution of 1 / 8°, revealed that fishing activities were predominantly concentrated in two regions: 121.1°E to 124°E, 35.7°N to 38.7°N, and 119.8°E to 122.8°E, 33.6°N to 35.4°N. This research can provide a foundation for quantitative evaluations of fishery resources, which can offer vital data to promote the sustainable development of marine capture fisheries.

Abstract:Aiming to address challenges in predicting the remaining useful life ( RUL) of harmonic drives-such as difficulties in selecting degradation nodes, poor physical interpretability of degradation indicators, and large prediction deviations, a novel approach was proposed. The method combined a one- dimensional stacked convolutional autoencoder ( SCAE) integrated with deep convolutional embedded clustering (DCEC) for degradation point extraction, along with an improved dung beetle optimization (DBO) algorithm to enhance the performance of a CNN-LSTM-based RUL prediction model. The vibration signals were processed by using the SCAE DCEC framework to identify degradation nodes, addressing issues related to the difficulty of node selection and the low compatibility between degradation indicators and the predictive network. Secondly, a modified dung beetle optimization (MDBO) algorithm was developed, incorporating SPM chaotic mapping, adaptive probability thresholds, and differential mutation perturbations, with its performance rigorously evaluated. Thirdly, the MDBO algorithm was applied to optimize the hyperparameters of the CNN-LSTM model, forming the MDBO-CNN-LSTM-RUL prediction model. An accelerated life test and validation experiment were conducted by using a harmonic drive test bench. The experimental results demonstrated that the MDBO CNN LSTM model significantly outperformed CNN, LSTM, CNN-LSTM, DBO-CNN-LSTM, fully convolutional networks, and Bayesian-optimized LSTM models in terms of goodness of fit. The proposed model achieved a prediction accuracy of 91.33% and exhibited superior recognition capability for capturing the degradation trends during the late stages of the lifecycle of harmonic drive.

Abstract:Aiming at the problem that the tracking accuracy of the three variable controller ( TVC), which requires high precision of the controlled model, is reduced due to the influence of system parameter changes and external interference on the loading test bench of electro-hydraulic damping damper. An adaptive three variable controller ( ATVC) was proposed, with a model reference adaptive control (MRAC) to compensate the loading system into an ideal stable reference model, and then an adaptive three variable control strategy was applied to the ideal reference model. According to the theoretical model of electro-hydraulic loading system of damping damper, the transfer function of the loading system considering the damping of the damping damper under test was derived. The model reference adaptive control system was designed based on Diophantine equation polynomial, and the loading system was compensated to realize the ideal reference model with non-minimum, which improved the system stability. The natural frequency and damping ratio of the loading system with a stable ideal reference model were compensated by three-state feedback, and pole assignment was realized by three-state feedforward to improve the dynamic characteristics of the system. Through the position step and sine response tests of the three-state control and the adaptive three-state control, under the action of the position square wave loading signal of 2.0 mm / 10 Hz, the overshoot was 13.5% and 9.6% , respectively, decreasing 28.9% , and the rise time was 7.4 ms and 5.8 ms respectively, decreasing 21.6% . The steady-state errors were 0.030 mm and 0.018 mm, respectively, which were decreased by 40.0% . Under the position sinusoidal loading signal of 1.5 mm / 30 Hz, the displacement errors were 0.29 mm and 0.18 mm, respectively, which were decreased by 37.9% . The adaptive three-state control can obviously improve the transient performance index and displacement tracking accuracy of the electro-hydraulic servo loading test bench.