Abstract:Selective harvesting robots are an important technical approach for intelligent harvesting of agricultural products. The harvesting end, which consisted of the end-effector, wrist joint, and corresponding sensing and control system, was the key unit determining harvesting success rate, damage level, and environmental adaptability. In view of the insufficient coupling among structure, sensing, and control in current studies, the research progress of harvesting ends for selective harvesting robots targeting different objects, such as fruits, tea leaves, and flowers was systematically reviewed. Firstly, the functional requirements of harvesting ends were analyzed from the aspects of biological characteristics of harvesting objects, plant attachment characteristics, canopy environment, and local spatial constraints. Secondly, focusing on end-effector and wrist-joint structures, the structural characteristics and applicability of end-effectors with different contact modes and serial/ parallel wrist-joint configurations were reviewed. Furthermore, the research status of visual perception, visuotactile fusion, key-part recognition, contact-state perception, approach and alignment, compliant grasping, separation, and transfer control was summarized. Finally, the existing problems were discussed in terms of structural fault tolerance, sensing robustness, and control adaptability. Future development trends were proposed, including modular and reconfigurable harvesting ends, multimodal perception, integration of agricultural machinery and agronomy, and embodied-intelligence-driven closed-loop harvesting with perception, decision-making, and execution.

Abstract:Against the backdrop of large-scale and industrialized development of the edible mushroom industry, harvesting remains highly dependent on manual labor and has become a key bottleneck restricting full-process automation. Focusing on major cultivated species in China, such as shiitake mushroom, enoki mushroom, button mushroom, and oyster mushroom, the characteristics of harvesting environments under different cultivation modes were systematically reviewed, including log cultivation, bed cultivation, bag cultivation, and bottle cultivation, and the operational constraints and technical requirements associated with typical production systems were analyzed. The research progress and commercialization of edible mushroom harvesting robots at home and abroad were comprehensively reviewed. The automated technical routes adopted in foreign countries under standardized cultivation systems were compared with the domestic technical features oriented toward multiple species and unstructured environments, and the differences and implications of domestic and international harvesting technologies were summarized. Key technologies of edible mushroom harvesting robots were discussed, including end-effectors, perception, mobile platforms, system integration, and embodied intelligence. Particular attention was given to the development of gripping, suction, cutting, and bio-inspired flexible end-effectors, as well as to the application trends of deep learning, stereo vision, and multimodal information fusion in maturity recognition and target localization. The adaptability and integration characteristics of fixed, mobile, and rail-guided platforms in narrow and humid cultivation environments were also examined. Current technical challenges were summarized, including insufficient perception robustness, poor compatibility of end-effectors, low operational efficiency, and weak economic feasibility. Finally, future prospects were proposed from three aspects: strengthening the standardization of cultivation modes and the integration of agronomy with agricultural machinery, developing embodied intelligent harvesting systems integrating perception, decision-making, and control, and constructing automated production lines that coordinated harvesting, root cutting, grading, and other postharvest processes, so as to provide theoretical support and technical references for the intelligent upgrading of China’s edible mushroom industry.

Abstract:YOLO series algorithms, characterized by high inference speed, high detection accuracy, and a highly modular architecture, have become the preferred methods for object detection in smart agriculture. The research progress and application status of YOLO algorithms in the field of crop planting agriculture were systematically reviewsed. Firstly, according to the implementation mode of visual tasks, visual tasks in planting agriculture were classified into three categories: fruit detection tasks, crop condition analysis tasks, and robotic harvesting vision tasks. Based on this classification, a model suitability evaluation framework was constructed to clarify the adaptability and application status of YOLO models in different tasks. Secondly, in response to common challenges in complex agricultural planting scenarios, such as small targets, occlusion, diverse illumination conditions, and limited computing resources, the mainstream improvement strategies for YOLO models were summarized, including model structure and module optimization, as well as the introduction of attention mechanisms. Finally, from the perspectives of key factors dominating visual task performance, the demand for high-quality data, model version selection, and optimization strategies, the application patterns of YOLO series algorithms in planting agriculture were comprehensively analyzed, and the future development trends were discussed from two directions: breakthroughs in perception capability and improvements in generalization ability.

Abstract:Picking path planning is the core decision-making link that determines the harvesting efficiency of automated tea bud picking robots. Aiming at the problems of difficult three-dimensional spatial modeling, high real-time response requirements and poor adaptability to dynamic node scales in automatic tea bud picking path planning, the tea bud picking task was modelled as a three-dimensional traveling salesman problem (3D-TSP), and a deep reinforcement learning-based 3D path planning algorithm was proposed for tea bud picking. A parametric 3D tea bud dataset with a zero-padding strategy was constructed, a dynamic masking mechanism was designed to ensure path validity, a long short-term memory (LSTM) module was introduced to dynamically fuse historical path information, a constrained multi-head self-attention mechanism was developed to improve decision-making accuracy, and the stability model training was done based on the Actor-Critic framework. Comparative experiments showed that the relative optimality gap between the proposed algorithm and the exact Concorde algorithm was controlled within 2% , and its computation time in the scenario with 80 tea buds was only 1.7% of that of Concorde. The algorithm outperformed classical heuristic algorithms and other deep reinforcement learning-based algorithms, maintained stable performance in real tea garden scenarios, and provided an efficient solution for tea picking robot path planning.

Abstract:Intelligent picking is a key development for improving the efficiency of the tea industry. Aiming to address the current issues of poor positioning accuracy and stability of mechanical arms used for picking high-quality tea, an adaptive model compensation control strategy was proposed. Firstly, a mechanical arm dynamics model was constructed by using the Lagrange method. Secondly, a radial basis function (RBF) neural network and a nonlinear disturbance observer (NDO) were designed to adaptively compensate for the dynamic model errors and external disturbances, respectively. Furthermore, the stability of the proposed control system was proven based on Lyapunov stability theory. Simulation test results showed that, following the introduction of the nonlinear disturbance observer, the position and velocity tracking performance of the three-axis robotic arm improved significantly, with position tracking errors reduced to 0. 18 rad, 0. 01 rad, and 0. 43 rad, and fluctuation amplitudes noticeably decreased. Field picking validation tests showed that, compared with traditional PID control, this strategy reduced the robotic arm??s acceleration fluctuations by 86. 8% , reduced vibration amplitude by 94. 8% , achieved an average picking time of approximately one second per cycle and a successful picking rate of 51. 82% , as well as an optimal picking speed of 0. 79 s per tea bud. The research successfully overcame the challenges of high-precision positioning and smooth motion coordination control during tea bud picking, offering a reliable technical solution for the intelligent picking of premium tea.

Abstract:Aiming to address the demand for high-precision, high-flexibility robotic arms in the picking of premium tea, a hybrid-configuration tea-picking robotic arm based on a Delta parallel mechanism and a 3-degree-of-freedom serial mechanism was designed. Firstly, considering the tea garden environment and the growth characteristics of tea bushes, kinematic models of the parallel and serial mechanisms were established by using the geometric method and the D-H parameter method, respectively. The overall kinematic solution of the hybrid robotic arm was achieved through Euler angle transformation. Subsequently, the NSGA-Ⅱmulti-objective optimization algorithm was employed to optimize the structural dimensions of the robotic arm, resulting in improvements of 54. 3% , 81. 3% , and 57. 92% in global orientation manipulability, regional orientation manipulability, and global dexterity, respectively. Furthermore, through ADAMS dynamic simulation and ANSYS topology optimization, the mass of the active arm was reduced by 27. 62% , while the maximum stress and deformation were controlled within 2. 118 MPa and 0. 015 mm, respectively, and the joint driving torque was decreased by 13. 06% . Prototype tests demonstrated that the robotic arm achieved a repeated positioning accuracy of 0. 036 mm, a single-bud plucking success rate of 79% , and an average cycle time of 2. 27 s, verifying the feasibility of the hybrid configuration for premium tea plucking.

Abstract:Aiming to address the motion path planning and obstacle avoidance issues of honey pear harvesting robot arms in unstructured orchard environments, a path planning method integrating collision margin constraints was proposed, aiming to improve the motion efficiency and operational safety of the robot arm. By establishing a branch elastic deflection model and using a three-point bending test to determine the critical force for destroying branches of different diameters, the force collision threshold was determined. Combined with the cantilever beam model, the collision margin was quantified by translating the maximum deflection angle of the branch and the maximum linear displacement perpendicular to the axis into a collision margin. A hybrid collision detection model based on octree and envelope methods was constructed, and an A-Informed RRT? algorithm with a target attraction mechanism was proposed, achieving robot arm motion path planning that allowed non-destructive elastic collisions. Path planning experiments showed that the fully obstacle-avoidance A-Informed RRT? algorithm had an average planning time of 1. 175 s and a path length of 88. 463 mm in a three-dimensional environment. After incorporating the collision margin, the collision-margin-based A-Informed RRT? algorithm reduced the average planning time to 0. 089 s and shortened the path length to 85. 036 mm. In orchard harvesting tests, the success rate of the robot arm in an unobstructed scenario reached 96% , while in scenarios with thin branch obstructions it was 76% , which was significantly higher than the 52% success rate in the full obstacle-avoidance scenario, validating the effectiveness of the proposed method in improving the motion efficiency and success rate of the robot arm. This method can significantly enhance the robot arm's motion efficiency and harvesting success rate while ensuring the safety of both the robot arm and the plants, providing an approach for robot arm harvesting operations in unstructured orchards.

Abstract:Aiming to address the current challenges of low efficiency, inconsistent quality, and high labor intensity in bulk tea harvesting, a motorized self-propelled profiling tea harvester was designed by incorporating the topographic and environmental characteristics of tea plantations, cultivation management requirements, and agronomic operational protocols. Firstly, the overall structural design of the electric- driven self-propelled cantilever profiling tea harvester was carried out, including four parts: the crawler chassis, posture adjustment device, tea harvesting device, and image acquisition device. Subsequently, a binocular camera and YOLO v8s were employed to detect and locate fresh leaves on the tea canopy in real time, after which the least squares method was used to solve the theoretical operating posture of the tea harvester. Through the coordinated action of a posture adaptive adjustment device and an intelligent harvesting actuator, profiling control and automated harvesting of bulk tea were achieved. Finally, a prototype of the self-propelled cantilever profiling tea harvester was developed. The stability of the entire machine was analyzed to determine the critical tipping angles under longitudinal and transverse slopes during both transfer and operation conditions, thereby verifying the safe operating boundaries. Field harvesting experiments for bulk tea were conducted in a tea plantation. The experimental results demonstrated that the positional error of the machine remained consistently within the preset threshold during the tea harvesting process, and the actual working posture dynamically adapted to changes in the height and tilt angle of the tea canopy. The average bud integrity rate reached 83. 9% , with a missed collection rate of 0. 74% and a missed picking rate of 0. 95% . The proportion of apical buds with three leaves or fewer accounted for 87. 8% . All performance indicators met the industry standards for tea harvester operation quality, providing technical and equipment support for efficient and precise harvesting of bulk tea.

Abstract:Accurate morphological detection of mechanically harvested fresh tea leaves is beneficial for improving the accuracy and efficiency of automated grading. YOLO v5n-DRSW, an advanced YOLO v5n based model specifically designed for the precise recognition of machine-harvested tea leaf morphologies was introduced. The model integrated several innovative features: a distribution shifting convolution (DSConv) module in the head network to reduce complexity and enhance efficiency; a reparameterized generalized feature pyramid network (RepGFPN) in the neck to improve generalization and robustness; and the squeeze-and-excitation (SE) attention mechanism embedded in the backbone to strengthen feature perception for small targets like tender buds. By leveraging a global field of view, this mechanism further enhanced the perception capability of the feature maps. Additionally, the wise intersection over union (WIoU) loss function was used to dynamically adjust gradient contributions during training. Compared with the baseline, YOLO v5n-DRSW exhibited significant advantages in detecting machine-harvested tea leaves with complex morphologies. Experimental results demonstrated that YOLO v5n-DRSW achieved 98. 1% accuracy, a 2. 1 percentage points improvement over the baseline, with an inference time of just 2. 11 ms per frame. This rapid processing speed represented a notable improvement over the baseline model. The model also reduced floating-point operations by 2. 44% , confirming its lightweight nature. In practical applications, it attained an average online recognition accuracy of 94. 34% with a miss rate below 1. 1% , highlighting its strong potential for enhancing automated tea leaf grading systems. Overall, the model demonstrated excellent detection performance, providing reliable assistance for improving the morphological detection outcomes of fresh tea leaves.

Abstract:In response to the problems in three-dimensional shelf cultivation of shiitake mushrooms, such as high labor intensity, high cost of manual picking, and insufficient research on selective picking robots, an integrated shiitake picking robot was proposed that combined autonomous movement, mushroom stick handling, selective picking and shiitake collection, taking into account the growth characteristics of shiitake mushrooms and the agronomic requirements of factory-based production. To adapt to the multi-layer three-dimensional shelf structure, a lifting-clamping mechanism was developed. Through the linkage between the lifting platform and the lead screw module, this mechanism enabled the non-destructive handling and clamping fixation of mushroom sticks in different layers, effectively solving the problem of handling mushroom sticks in multi-layer shelf environments that were both difficult and prone to damage. A multi-view vision system was adopted to collect multi-dimensional information, including the phenotypic characteristics of shiitake caps and the opening-closing state of gills. An improved YOLO v8 model was used to enhance the accuracy of target detection, and a cross-view target matching rule based on geometric constraints was designed to associate the detection results of different views. This combination enabled the accurate identification and positioning of shiitake mushrooms that were ready for picking, laying a foundation for subsequent selective picking operations. To avoid damage to shiitake caps and mushroom sticks during the picking process, a clamping end-effector was designed by simulating manual picking actions based on the morphological characteristics and force-bearing properties of shiitake mushrooms. By applying a lateral clamping force on the shiitake stipes, the end-effector achieved the separation and picking of shiitake mushrooms without causing damage to the caps or affecting the subsequent growth of mushroom sticks. The field experimental results indicated that the picking success rate of the robot reached 91. 6% , with an average picking time of 16. 5 s per shiitake mushroom and a damage rate of 15. 9% .

Abstract:Aiming at the problems of complex field environment and difficulty in taking into account the accuracy of weed identification and detection efficiency in the critical window period of 2 ~ 5 leaf weeding of maize, a lightweight detection model BACD-YOLO for maize seedlings and weeds based on improved YOLO v8n was proposed. Using Adamax optimizer to enhance the robustness of model in field environment; the weighted bidirectional feature pyramid network (BiFPN) was introduced to the feature fusion network as the connection layer to improve the detection effect of the model on weeds with different growth. Adopting lightweight down sampling module (Adown) replaced the conventional convolution in the network to reduce the amount of parameter calculation of model redundancy; the coordinate attention (CA) mechanism was embedded in the SPPF layer and the feature fusion network to improve the positioning ability of the model for small targets and densely distributed weeds; DualConv lightweight double convolution was used to replace the ordinary convolution structure in original model to further realize the lightweight of the model and the detection ability of the model for weeds with similar characteristics. The experimental results showed the accuracy, recall and average accuracy of improved model were 86. 6%, 86. 2% and 91. 2%, respectively, which were 2. 2, 1. 5 and 1. 6 percentage points higher than that of the original model, and floating-point calculation and parameter quantity were only 6. 2 × 109 and 2. 3 ×106, which were 23. 5% and 23. 3% lower than that of original model, respectively. According to the verification test results, the improved model was more suitable for edge device deployment application with high detection accuracy, strong generalization ability and excellent lightweight performance. The frame rate was 19. 4 f/ s, and the detection accuracy was 86. 6%, which can meet requirements of field real-time detection. The research result can provide an effective lightweight solution for accurate identification of corn seedlings and weeds and robot weeding.

Abstract:Hericium erinaceus is a traditional edible and medicinal fungus with high nutritional and economic value, and its commercial circulation requires rapid, objective, and stable quality evaluation. However, manual sensory grading is easily affected by evaluator experience, illumination conditions, and fatigue, which leads to low efficiency and inconsistent judgment results in large-scale production scenarios. Aiming to address these problems, an intelligent quality detection method for H. erinaceus based on machine vision and deep learning was proposed. Firstly, images of normal and defective samples were collected under practical acquisition conditions, and preprocessing operations were used to reduce background interference and improve the visibility of surface texture features. A multi-scale optimized watershed segmentation algorithm was then constructed to extract the target region accurately, suppress over-segmentation in complex backgrounds, and provide reliable input for subsequent feature learning. On this basis, image enhancement and data augmentation strategies were introduced to enrich sample diversity, alleviate the influence of limited datasets, and strengthen the generalization ability of the model. In the classification stage, a Swin Transformer Tiny network was adopted to combine local window attention with shifted window attention, so that both fine-grained local texture information and broader contextual relationships can be effectively captured. The model was used to discriminate between normal and defective H. erinaceus samples, and its performance was compared with several mainstream deep learning models under the same experimental conditions. Experimental results showed that the proposed method achieved a classification accuracy of 94. 3% on real-world collected images, outperforming ResNet-18, EfficientNet, DilateFormer, and MambaVision by 3. 5, 2. 2, 1. 5, and 1. 2 percentage points, respectively. The results indicated that the combination of accurate segmentation and attention-based feature extraction can improve the reliability of H. erinaceus quality detection. The proposed system demonstrated strong engineering practicability and deployment potential, providing a feasible technical route for intelligent quality inspection of structurally complex edible fungi and supporting the standardization and intelligent development of the edible and medicinal fungus industry.

Abstract:A white asparagus harvesting robot was designed to address the high labor intensity and low operational efficiency associated with manual harvesting. Based on the planting and harvesting characteristics of white asparagus, the overall structure and control system of the robot were investigated. The primary motion sequence of the harvesting process was analyzed to determine the main structure and core design parameters of the end-effector. The robot was configured for continuous ridge-straddling operation, in which visual detection, lateral positioning, blade insertion, in-soil cutting, lifting, and discharge were coordinated without stopping. An MBD-DEM (Multi-body dynamics and discrete element method) coupled simulation model was established to verify the rationality of the design parameters. Simulation results indicated that the thrust required for the end-effector??s insertion was 580. 2 N, and the thrust required for the horizontal cutting action in the soil was 44. 2 N. A three-factor, three-level orthogonal experiment was conducted using the cutting position multiplier, cutting distance multiplier, and the height of the end-effector relative to the ridge surface as indices. The results showed that the optimal harvesting parameter combination was a cutting position multiplier of 1. 2, a cutting distance multiplier of 1. 2, and a height of 5 cm. Field experiments demonstrated an actual spear recognition success rate of 90. 2% with an average detection time of 23 ms. For successfully recognized spears, the harvesting success rate was 92. 7% , with an average single positioning time of 1. 7 s and an average single harvesting cycle time of 3. 2 s. The asparagus damage rate was found to be 4. 3% . These results provide a basis for optimizing selective harvesting equipment for white asparagus grown on high ridges under field conditions.

Abstract:Aiming to address the challenges of missed detection, low recognition accuracy caused by complex field conditions such as varying lighting and occlusion by overlapping branches and filaments, as well as the difficulty in deploying large models on edge devices, lightweight network for safflower recognition ( LNSR), an improved lightweight network structure was proposed based on YOLO 12. Specifically, the heterogeneous edge-pooling dual-stream fusion module (HEP-DSF) was designed to enhance the network’s ability to extract edge and texture information in the early stage, thereby improving recognition accuracy. The triple cooperative adaptive fusion module (TriCAFusion) was constructed to strengthen the model??s representation of key target features and reduce the probability of missed detection. The adaptive neighborhood pooling for down sampling module (AdaPool) was introduced to improve the model’s robustness to lighting variations and occluded scenes. Furthermore, the lightweight shared-BN network (LSBNet) detection head was developed to reduce model complexity and improve deployment efficiency. Experimental results on the safflower dataset showed that LNSR achieved only 1. 72 × 106 parameters, a reduction of 31. 5% compared with that of YOLO 12, with a model size of 4. 4 MB (1. 1 MB smaller). It also reached an mAP50 of 98. 9% and a recall rate of 98. 3%, representing improvements of 1. 4 and 1. 0 percentage points, respectively. When generalized to the chrysanthemum dataset, LNSR achieved 1. 63 × 106 parameters, a model size of 4. 4 MB, and an mAP50 of 97. 0%, which was 5. 9 percentage points higher than that of YOLO 12. Heatmap validation confirmed its precise focus and characterization ability on petal edges and stamen textures. Deployed on the Jetson edge device, it achieved a real-time frame rate of 30 f/ s. It demonstrated that through the collaborative innovation of the four modules, LNSR achieved an optimal balance between accuracy and lightweight design, providing an efficient and reliable visual recognition solution for selective harvesting of safflower.

Abstract:Aiming at the technical challenges in the robotic harvesting process of asparagus, such as insufficient perception accuracy, high miss rate, adhesion of overlapping instances, and weak cross- domain generalization caused by dense plant distribution and severe occlusion, an in-depth analysis of the slender geometric morphology and spatial features of dense occlusion in asparagus was conducted. An instance segmentation model, dynamic gated attention-detection transformer (DGA-DETR), which integrated dynamic gated attention and edge awareness, was proposed. By designing a dynamic gated attention (DGA) module, an instance-semantic-driven dynamic gating mechanism was utilized to achieve precise feature focusing across space and scales, effectively suppressing background noise interference under dense occlusion at the mechanistic level. Additionally, an edge-aware up-and-down sampling mechanism was constructed to strengthen boundary decoupling capabilities, improving the model’s segmentation modeling accuracy for the slender morphology of asparagus and occlusion boundary regions. To verify the effectiveness of the segmentation model, an asparagus harvesting dataset covering different lighting conditions and occlusion levels was constructed for model training, encompassing multi-level domain shifts in pixels, scenes, and semantics, and a cross-domain dataset was established for testing. Experimental results demonstrated that the proposed DGA-DETR model achieved Mask precision, Mask recall, Mask mAP@ 0. 5, and Mask mAP@ 0. 5:0. 95 of 91. 60% , 67. 24% , 90. 98% , and 59. 13% , respectively, representing improvements of 5. 33, 8. 43, 6. 56, and 9. 51 percentage points over the baseline model, and significantly reducing missed instances and mask discontinuities under dense occlusion conditions. Cross-domain test results indicated that this method exhibited excellent robustness and generalization capabilities for the precise identification of asparagus under occlusion and complex backgrounds. The research results can provide a technical reference for the robotic harvesting of other tender-stem crops in complex environments.

Abstract:The entire process of selective harvesting of broccoli requires precise maturity recognition, harvesting posture analysis, and rapid crop row navigation. However, at present, there is a lack of a visual algorithm suitable for the full-process operation. To address these issues, an efficient harvesting visual system, EH-YOLO, was proposed for the selective harvesting robot of broccoli, aiming to achieve accurate maturity recognition and autonomous harvesting in complex agricultural environments. The efficient multi-scale attention module was incorporated into the YOLO v8n backbone network, and a soft spatial pyramid pooling-fast module was designed. Additionally, a lightweight neck structure based on grouped shuffle convolution was employed, and the loss function was improved to enhance feature extraction and fusion capabilities, reduce computational complexity, and improve model performance in complex environments. Furthermore, by analyzing the crop??s agronomic features, the plant location was determined based on the center point of the broccoli head. The harvesting process was adjusted according to the quadrant in which the broccoli was located, and clustering algorithms, combined with the least squares method, were used to extract the crop row navigation lines. The model size was 3. 3 MB, and the detection speed reached 61. 45 f/ s. The accuracy of maturity recognition for mature broccoli was 92. 84% . Additionally, compared with the baseline model, EH-YOLO reduced the computational parameters by 38. 68% , decreased the model size by 46. 77% , and increased the average accuracy by 4. 01 percentage points. In field trials, the system successfully completed the full-process selective harvesting task, with a motion and positioning success rate of 96. 77% and a selective harvesting success rate of 87. 37% . Therefore, EH-YOLO was a practical and efficient visual system for the selective harvesting robot of broccoli, with strong applicability for commercial broccoli fields.

Abstract:Aiming to address issues such as poor seed delivery stability and low seeding accuracy caused by seed collision with the inner wall of seed delivery tubes during high-speed operation of precision wheat strip seeders, a split air-assisted seed guiding system was designed. Key components of the guiding system were developed, and single-factor experiments were conducted by using a DEM-CFD coupled simulation method to determine the selection range of major structural parameters. Based on the results of single-factor tests, a Box-Behnken three-factor, three-level orthogonal combination simulation experiment was carried out to further determine the optimal structural parameter combination of the air- assisted seed guiding device. The experimental results showed that when the Venturi inlet pipe contraction section diameter was 16 mm, the contraction section length was 30 mm, and the Venturi inlet pipe jet angle was 25°, the seed delivery performance of the air-assisted guiding device was optimal, with a coefficient of variation for instantaneous flow consistency of 3. 82% and a coefficient of variation for seed mass flow stability of 4. 54% . The model was 3D printed, and bench tests were carried out to verify the optimal working range of the air-assisted seed guiding device. Bench test results indicated that with airflow assistance, the seed guiding uniformity of the device was significantly better than that without airflow assistance, with the optimal coefficients of variation for instantaneous flow consistency and seed mass flow stability reduced from 5. 82% and 5. 51% to 3. 51% and 3. 27% , respectively. Field test results showed that at an operating speed of 8 ~ 10 km/ h, the coefficients of variation for instantaneous flow consistency and the rate of seedling gaps did not exceed 4. 82% and 1. 95% , meeting the seed delivery performance requirements of the air-assisted guiding system. The research findings on this split air-assisted seed guiding device can provide guidance for subsequent design and performance improvement of split air-assisted seed guiding systems.

Abstract:High-density planting, high-precision operation and high-speed production have become the new paradigm of modern corn production, putting forward strict requirements for sowing equipment. Pneumatic maize planters have attracted much attention because they can maintain a high seeding qualification rate even at high speeds. However, sowing operations confronted with three major challenges: unstructured environment, high agronomic constraints, and strong farming seasons, which led to complex and changeable working conditions. Improper matching of operational parameters to conditions reduced seeding quality and underutilized equipment performance. The DEM-CFD gas-solid two-phase coupling method was employed to elucidate seed motion dynamics and mechanical behavior within the seed metering device at the microscale. Air pressure, seed feed rate, and seed cleaning gap were identified as critical factors affecting seeding quality. Using single grain rate as the evaluation metric, the operating parameters were optimized for two representative corn varieties in the Huang-Huai-Hai region (Huangjinliang MY73 and Zhengdan 958) through Box-Behnken response surface methodology. The optimal parameter combinations were determined as follows: for MY73 at 12 km/ h, maximum singulation rate was achieved at 4 kPa air pressure, with seed feed rate and cleaning gap set to Level 7; for Zhengdan 958 at 12 km/ h, optimal performance occurred at 4. 5 kPa air pressure, Level 8 feed rate, and Level 7 cleaning gap. Orthogonal test results demonstrated that air pressure, seed feed rate, and seed cleaning gap had a very significant effect on single grain rate, with obvious interactions between factors; the established regression model fit well and can accurately predict seed discharge performance. Results demonstrated significant variation in optimal parameters across varieties and operating speeds. The research result can provide a reference for parameter selection in practical operations, which can establish a theoretical foundation for automated parameter configuration in intelligent seeding systems.

Abstract:Aiming to address mechanical transplantation challenges of vegetative cutting plug seedlings caused by insufficient root anchorage strength and significant canopy expansion, leading to high substrate fragmentation rates and low seedling extraction success, a low-damage high-efficiency seedling extractor was proposed based on a top-clamping mechanism. By establishing a kinematic model of the clamping mechanism based on non-circular gear planetary transmission characteristics and a “ linear clamping- vertical extraction” trajectory principle, key structural parameters were optimized through parameter optimization to minimize transmission ratio fluctuations: planetary carrier length (100 mm), extraction arm length (130 mm), locking arc base radius (35 mm), and contact angle (44. 43°), achieving a 59. 31% reduction in transmission ratio fluctuation. The non-circular gear pitch curve was derived based on the angular mapping relationship of a two-bar linkage mechanism. Combining discrete element method (EDEM) simulations with experimental validation using Photinia × fraseri cutting plug seedlings, optimal operational parameters were determined: ejector diameter (6 mm), stroke (6 mm), speed (28 mm/ s). At a seedling extraction rate of 60 seedlings/ min, the extraction and delivery success rate reached 94. 33% , substrate fragmentation rate remained below 3% , and stem damage rate was under 1. 67% , with no observable seedling injury. The device employed a wedge-shaped actuator to reduce adhesion between the substrate and seedling trays, combined with a kidney-shaped clamping trajectory for low- damage and high-efficiency seedling extraction, meeting the requirements for automated transplantation of cutting seedlings.

Abstract:Rice ridge-furrow cultivation has significant advantages in carbon sequestration, emission reduction, energy saving, and yield increase. Currently, rice ridge-furrow cultivation faces problems such as numerous operation steps, high labor intensity, and low efficiency. To improve the operation efficiency of ridge-furrow cultivation and simplify the operation process, based on meeting agronomic requirements, a dedicated transmission system for rice transplanting was designed and matched with existing ditching and ridging equipment to meet the normal working needs of rice transplanting. Aiming at the stability of transplanting operation, a transplanting depth control profiling system was designed to ensure the quality of transplanting. An integrated machine for rice ditching, ridging, and transplanting was created, which can complete multiple operations such as rotary tillage, ditching, ridging, and transplanting at one time. Targeting the qualification rate of transplanting depth, a discrete element simulation test was conducted on the profiling system to obtain the optimal structural parameters: when the installation height of the profiling board was 199 mm, the width of the profiling board was 463 mm, and the front angle of the profiling board was 20°, the shortest recovery time of the profiling board posture was 0. 697 s, the minimum resistance of soil to the profiling board was 589. 52 N, the minimum soil subsidence was 16. 2 mm, the qualification rate of transplanting depth was 93. 22% , and the profiling performance was good. A prototype was trial-manufactured according to these parameters, and field performance tests and yield comparisons of ditching, ridging, and transplanting for early rice and late rice in two years were carried out for four different operation modes: ordinary rotary tillage + transplanter (W1, control), one-time ditching, ridging, and transplanting operation ( W2), one rotary tillage + ditching, ridging, and transplanting ( W3 ), and two rotary tillages + ditching, ridging, and transplanting (W4). The field operation performance test showed that when the forward speed of the machine was 0. 5 m/ s and the rotary tillage depth was 166 ~ 225 mm, the average soil subsidence was about 17 mm, and the qualification rate of transplanting depth reached 91. 20% ; the yield comparison test of different operation modes showed that the yield increases of early rice were 0. 42% , 4. 00% , and 4. 70% , respectively, and the yield increases of late rice were 1. 05% , 5. 30% , and 6. 10% , respectively. From a comprehensive perspective, the two rotary tillage operations increased the operation cost, but the yield increase was not significant, thus W3 had the best comprehensive benefit.

Abstract:Regarding the challenge of constructing a discrete element model for the thin-walled leaves of rice pot seedlings, which limits the simulation analysis of pneumatic seedling throwing due to the lack of a suitable leaf model to characterize the effect of seedling posture changes on its trajectory under the interaction between flexible leaves and airflow fields, taking rice pot seedlings at the three-leaf and one- heart stage as the research object, a reverse reconstruction method for the discrete element model of seedling leaves was proposed based on multi-view image fitting. This method discretized the continuous contour of the leaves using unit grids, mapped the spatial morphological relationships of the leaves into particle coordinate information, and constructed a discrete element model of the flexible leaves of rice pot seedlings based on the Hertz-Mindlin with Bonding V2 contact model. Using the relative deflection of the flexible leaves as the evaluation index, the parameters affecting leaf relative deflection were screened for significance through a Plackett-Burman design. A second-order regression model for leaf relative deflection was established based on a Box-Behnken design. It was clarified that normal stiffness per unit area, tangential stiffness per unit area, and bonding ratio significantly affected leaf relative deflection. Simulation tests performed with the calibrated parameters showed a relative error of 2. 51% between the simulation and physical experiment, verifying the validity of the constructed model. The research result can provide a reference for the subsequent development of discrete element models suitable for simulating pneumatic seedling throwing.

Abstract:Watermelon and cantaloupe seedling pots, characterized by their tender stems and expansive foliage, present challenges for existing fully automated transplanters, including low pickup success rates, high damage rates, and poor operational stability. This study aims to address these issues by designing pick-up device with ejecting pot-receiving seedling capable of low-damage, high-efficiency removal of watermelon and cantaloupe seedlings. This study used “ Xizhoumi 25 ” cantaloupe seedlings as test samples, and the device adopted sequential lifting-and-grasping with intermittent seedling placement. A corresponding pneumatic control system was designed. Through ADAMS motion trajectory simulation and FluidSIM-P3. 6 pneumatic system analysis, the structural design and pneumatic timing control strategy were validated, ensuring high operational success rates. A mechanical analysis model for the lifting process was established, identifying the primary factors influencing successful seedling retrieval as the diameter, length, and speed of the lifting pins within the device. EDEM simulation methods were employed to analyse the lifting process. Based on determining the research range for pin diameter, length, and speed, a simulation test plan for the lifting process was constructed by using Box-Behnken experimental design methodology to ensure low-damage seedling retrieval. Through variance analysis and response surface analysis, the optimal combination of ejector pin parameters was determined as: pin diameter of 1. 4 mm, pin length of 20 mm, and pin speed of 150 mm/ s. Field trials using a transplanting seedling extraction rig confirmed that this parameter combination achieved a 94. 50% seedling extraction success rate with a 5. 94% seedling loss rate. No stem damage was observed, demonstrating suitability for low-loss, high-efficiency seedling extraction by watermelon and melon transplanters.

Abstract:Precise estimation of winter wheat yield at the field plot scale was considered to be of significant practical importance for agricultural yield management under large-scale farming conditions. Given that single vegetation indices were insufficient to comprehensively characterize crop growth status, the development of multi-parameter and multi-variable UAV-based remote sensing yield estimation methods was regarded as an emerging trend. An air-ground integrated decision-making framework was adopted, and winter wheat from the 2023—2024 growing season was selected as the research object. UAV multispectral imagery acquired during the heading and grain filling stages was utilized and combined with a limited amount of ground-measured leaf area index (LAI) as a crop morphological parameter to construct yield estimation models, including single-variable models ( vegetation index only), dual- variable models (vegetation index + texture features), and triple-variable models (vegetation index + texture features + LAI). Yield estimation modelling was conducted by using random forest ( RF), extreme gradient boosting ( XGBoost ), and support vector machine ( SVM ) algorithms, and the estimation performance of different models and feature combinations was comparatively analyzed. The results indicated that the RF algorithm based on the three-variable combination of vegetation index, texture features, and LAI achieved the highest modelling accuracy at both growth stages (heading stage: R2 = 0. 729, RMSE = 524. 475 kg / hm2, NRMSE = 11. 181% , RE = 3. 481% ; grain filling stage: R2 = 0. 779, RMSE = 479. 265 kg / hm2, NRMSE = 9. 736% , RE = 3. 205% ), and significantly outperformed the XGBoost and SVM algorithms. Based on the optimal model-feature combination, spatial distribution maps of winter wheat yield for the 2023—2024 and 2024—2025 seasons during the heading and grain filling stages were generated. The results revealed that yield estimation accuracy during the grain filling stage was generally higher than that during the heading stage. The average differences between estimated and measured yields during the grain filling stage for the two seasons were 334. 035 kg / hm2 and 284. 235 kg / hm2, respectively. SHAP analysis further indicated that LAI contributed substantially to the improvement of yield estimation accuracy across all growth stages. Overall, the findings demonstrated that the proposed air-ground integrated multivariate stepwise fusion decision-making approach enabled rapid and accurate estimation of winter wheat yield at the field plot scale, thereby providing effective technical support for smart agricultural management.

Abstract:Aiming to address the damage issues in robotic Agaricus bisporus harvesting caused by dense clustering growth patterns, a hierarchical strategy-based picking task planning method was proposed. Based on a global-local hierarchical framework, this method employed a YOLO 11 deep learning model to classify and detect mushrooms according to their occlusion degree, and formulated the picking sequence optimization problem as a traveling salesman problem. At the global planning level, a strategic framework was constructed to prioritize the harvesting of occluded mushrooms and dynamically updated environmental information through iterative detection processes, thereby effectively reducing collision damage caused by occlusion. At the local execution level, an improved simulated annealing algorithm was designed to optimize picking paths for different mushroom categories separately to enhance operational efficiency, and an adaptive concentric-layer peripheral picking algorithm was developed to minimize adhesion damage caused by dense spatial distribution. Simulation experiments demonstrated that the improved simulated annealing algorithm achieved path length optimizations of 19. 2% , 24. 7% , and 35. 0% in 20-, 40-, and 60-node scenarios respectively, with corresponding convergence efficiency improvements of 61. 5% , 39. 6% , and 18. 2% . Field experiments conducted with 20 groups comprising 550 mushroom samples validated that, compared with conventional height-based picking strategies, the proposed method achieved a success rate of 89. 8% , substantially reduced the collision rate from 5. 09% to 2. 16% ( a 57. 6% reduction), thereby validating that the method effectively reduced harvesting damage while maintaining operational efficiency.

Abstract:Aiming to further improve the speed and accuracy of cucumber disease recognition in greenhouses, a model based on an improved YOLO v10s was proposed. Firstly, the ResNet50 network was integrated into the backbone network to enhance the network depth, through which the model??s expressive capability was significantly improved. Subsequently, a CSPPC convolutional neural network structure was added to the neck, where computational redundancy was reduced while the feature extraction ability for incomplete or occluded data was strengthened. Simultaneously, the NAM attention mechanism was incorporated to amplify attention to critical information, avoiding complex computations in traditional attention mechanisms and achieving efficient feature enhancement, ultimately forming the RCN model for cucumber disease detection. Experimental results demonstrated that the RCN model achieved precision, recall, mAP@ 0. 5, and mAP@ 0. 5:0. 95 rates of 95. 0% , 98. 1% , 98. 3% , and 70. 4% , respectively, representing improvements of 4. 3, 7. 4, 2. 9, and 5. 3 percentage points compared with the baseline YOLO v10s, with significant enhancements observed. Ablation studies revealed that the integration of the ResNet50 network contributed most significantly to accuracy improvement, with all proposed modifications collectively enhancing the recognition precision of the YOLO v10s model. Comparative evaluations revealed that the RCN model exhibited superior performance relative to mainstream models, meeting detection requirements and providing an optimized solution for cucumber disease recognition in greenhouse environments. This approach was validated as holding substantial significance for the prevention and control of cucumber diseases in agricultural systems.

Abstract:Aiming to address the challenges in nighttime automated harvesting of edible roses in Yunnan, including limited samples, low maturity recognition accuracy, missed small targets, and constrained edge computing capacity, the targeted technical optimizations were conducted. In the data enhancement stage, an improved AttentionGAN-based day-night image conversion model was constructed, incorporating the multi-scale attention mechanism, designing a progressive fusion strategy, and simultaneously implementing lightweight modification of the generator. For the detection model, the YOLO-NRP model was built based on YOLO v8n: the C2f backbone was replaced with the RVG-Ghost (RepVGG-Ghost) module, the ECA-Spatial attention mechanism was embedded at the end of the backbone network to enhance key feature expression, and the neck network was optimized with PAFPN-Lite to strengthen multi-scale feature fusion. Experimental results showed that the images generated by the improved AttentionGAN achieved a Frechet inception distance (FID) of 87. 97 and a structural similarity index (SSIM) of 0. 588 2, with significantly better quality than that of the original model, effectively expanding the nighttime dataset. The YOLO-NRP model achieved 91. 3% precision, 90. 2% recall, 94. 3% mAP50, and 80. 7% mAP50 95, with all indicators outperforming the YOLO v8n baseline model. Meanwhile, the model weighed only 5. 49 MB and had an inference speed of 70. 26 f/ s, meeting the requirements of edge device deployment and providing reliable technical support for the development of automated harvesting equipment.

Abstract:In response to the degradation of cross-sectional light stripe images caused by water fog scattering during the visual detection of feed pellet accumulation in high-humidity caged environments, a novel image block frequency-domain fusion enhancement method was proposed based on fog concentration level recognition and dual gamma correction. Existing multi-exposure image fusion enhancement methods exhibited limitations such as poor specificity, computational redundancy, and inadequate real-time performance. Firstly, a fog concentration level recognition model integrated image grayscale statistical features and texture features was constructed to achieve fog concentration level identification. Subsequently, homomorphic filtering was applied for image preprocessing, and a dual gamma correction strategy was implemented based on the identified fog concentration level to generate a pair of luminance- complementary images. Next, using a 32 pixel × 32 pixel sliding window with a 50% overlap rate, the image pair was decomposed into overlapping sub-blocks, followed by frequency domain decomposition of each sub-block. The low-frequency components were subjected to t-distribution weighted fusion to balance brightness, while the high-frequency components were processed by using detail saliency weighted fusion to preserve edge details. The weight distribution of both components adhered to the physical properties of mist-induced degradation, subsequently reconstructing the image. Finally, the fully enhanced image was reconstructed through windowed accumulation and normalization. Experiments conducted on 2 358 images across six different fog concentration levels demonstrated that, compared with two classical and extensively validated image enhancement methods, this approach achieved superior comprehensive performance: the mean square error between image pixel values was reduced to 0. 007 8, the signal-to-noise ratio was increased to 10. 18 dB, structural similarity reached 0. 89, image entropy was 4. 54, naturalness index NIQE was 6. 42, and the average processing time was less than 0. 06 s per frame. This image enhancement method effectively mitigated contrast degradation and detail blurring caused by water fog, thus providing a solid foundation for the accurate visual detection of feed pellet accumulation in caged environments. It demonstrated significant potential for widespread applications in automated feeding systems within poultry and livestock caged farming operations.

Abstract:A rapid separation algorithm was proposed to solve the issue of cherry overlapping on sorting lines caused by cherries??small size, which negatively impacted sorting efficiency. The algorithm aimed to improve the accuracy and speed of cherry separation, addressing the problem of different overlapping scenarios: the proposed algorithm efficiently combined centroid detection and concave point matching to separate cherries. Firstly, the foreground image was expanded by using edge tangent extension and the distance transform values of the cut fruit bodies were accurately calculated. The centroid of the cherries was then extracted by using the distance transform and neighborhood maximum methods. The thresholding and small window maximum methods were used to accelerate centroid extraction. Curvature analysis and point clustering methods were employed to accurately detect the concave points of the inner and outer contours for concave point extraction. These concave points were used to handle areas with varying overlapping situations and applied concave point matching to filter out preliminary separation lines. Finally, to enhance segmentation accuracy, the regions around the separation lines, which might affect the fruit’s shape, were sharpened by using Unsharp Mask ( USM), and refined segmentation was performed by incorporating gradient information, resulting in more accurate overlapping separation curves. Experimental results showed that the centroid detection method proposed achieved precision and recall rates of 98. 54% and 97. 93% , respectively, improving by 5. 64 and 11. 18 percentage points compared with that of the distance transform erosion method. The precision and recall rates for the concave point detection were 92. 95% and 94. 76% , respectively, showing improvements of 3. 52 and 20. 58 percentage points over that of the convex hull method. The proposed algorithm demonstrated greater precision in separating cherries under various overlapping conditions than watershed-based separation methods. On an AMD Ryzen 5 3600 6-Core Processor, the segmentation of 989 pixel × 200 pixel images took 6 ms. The proposed method achieved high detection accuracy and significantly reduced processing times, making it suitable for real-time processing in cherry sorting production lines. It effectively met the demands of both accuracy and efficiency for the industry.

Abstract:Cow rear udder traits are key indicators for evaluating dairy cows’ roduction performance and breeding value, and their accurate and automated evaluation is of great significance for improving dairy farm management efficiency and genetic breeding levels. The complex structural morphology, naturally blurred boundaries of cow rear udders, as well as interferences such as occlusion and variable lighting in milking sites, make high-precision and automated image segmentation and trait evaluation extremely challenging. A multi-stage adaptive enhancement segmentation network for cow rear udder traits (MAAE-SegNet) was proposed. By introducing an adaptive parameter activation mechanism, it enhanced the backbone network’s dynamic expression capability for udder features in complex scenarios, and constructed a dynamic gated attention module to effectively focus on the key regions of the rear udder, thereby improving the clarity and completeness of rear udder segmentation boundaries. Experimental results showed that compared with the Mask2Former model, the improved model achieved 0. 5 and 1. 5 percentage points improvements in detection box accuracy and recall rate respectively, and 1. 8 and 2. 0 percentage points improvements in segmentation accuracy and recall rate respectively. The model had a parameter count of 4. 705 5 × 107 and a floating-point operation ( FLOP) count of 1. 59 × 1011, demonstrating higher accuracy without a significant increase in parameter quantity.

Abstract:In complex natural environments, enhancing the detection efficiency of underwater biological resources is vital for China??s marine economic development. To address the issues of limited computational resources and poor detection due to underwater complexity, CBM-YOLO, an improved YOLO v8s-based underwater target detection algorithm was proposed. Firstly, CSP-DLN, a lightweight feature extraction module, was designed, combining the advantages of 3 × 3 and 1 × 1 convolutional kernels to efficiently capture detailed spatial features and reduce redundant calculations. Secondly, to address the loss of feature information in underwater biological targets, a new feature fusion network, BGL-FPN, was introduced, which effectively improved detection accuracy through cross-scale connections combined with global and local spatial attention mechanisms. Lastly, max pooling downsampling ( MPD) was proposed, leveraging parallel processing of max pooling and convolutional branches to better capture edges and details of small targets, thereby enhancing their detection capability. Experimental results indicated that the algorithm attained mAP @ 0. 5 of 78. 2% and 69. 1% on the URPC2020 and UDD datasets, respectively, with improvements of 1. 5 and 2. 6 percentage points over the baseline model, while reducing parameters and computations by 47. 3% and 25. 5% . It outperformed other mainstream target detection algorithms. Deployed on the Jetson TX2 embedded device and accelerated by TensorRT, the model achieved mAP@ 0. 5 of 77. 4% and a detection speed of 37. 6 f/ s, enabling real-time underwater detection with high accuracy.

Abstract:Loess collapse-pits, as key triggers of soil erosion in China??s Loess Plateau, require accurate identification for effective land degradation control. An integrated framework combining UAV remote sensing with object-oriented image analysis (OBIA) and machine learning (KNN and CART algorithms) was developed to map and analyze loess collapse-pits in the Zhoutungou Basin. Key findings included optimized multi-scale segmentation ( scale was 30 ) with KNN classifier achieved superior accuracy (Kappa coefficient was 0. 896, extraction quality was 82. 6% ). Compared with the CART algorithm, the Kappa coefficient increased by 0. 069, and the extraction quality improved by 1. 9% . Morphological parameter inversion reveals advantages in capturing linear features, with standard errors for the major axis (R2 = 0. 773) and perimeter (R2 = 0. 842) lower than 1. 1 m and 4. 7 m, respectively. Spatial analysis revealed 453 collapse-pits showing distinct topographic preferences: 84. 99% occurred on slopes greater than 15° with density increased 1. 8-fold per 10° slope increment ( R2 = 0. 91), semi-shaded slopes hosted 2. 76 times more pits (5. 8 pits/ km2 ) than sunny slopes, and 71. 1% concentrated at mid- elevations (1 125 ~ 1 210 m). Morphometric analysis indicated 75% of pits were within 9. 02 m2, with elongated shapes ( aspect ratio was 2. 05 ± 1. 13) suggesting lateral erosion dominance, while depth exhibited bimodal distribution reflecting different formation mechanisms. The research established the first 3D morphometric threshold system for loess collapse-pits, providing critical data for soil conservation and geohazard prevention.

Abstract:Water content is one of the most important indicators for evaluating the quality of fresh corn cobs, which affects the quality grading of fresh corn. Because of the special physical characteristics of fresh corn cobs, such as the rod-like shape with different thickness and the unevenness of kernel rows on the surface, the near-infrared spectroscopic (NIRS) acquisition of fresh corn cobs and the prediction of water content modeling analysis have a great impact on the quality of fresh corn cobs, and therefore it is necessary to carry out the NIRS modeling study of the water content of fresh corn cobs in different collection areas and points. Firstly, the 360° spectra of the first, middle and last regions of the cob were collected by using a homemade NIR NDT device, with 60° intervals between each region and six point locations. Secondly, the outliers were rejected by Z-score, and combined with no preprocessing (NONE), standard normal variate ( SNV ), multiplicative scatter correction ( MSC ), first order derivative (1D), second order derivative (2D) and Savitzky-Golay smoothing (SG). Then, the sample set portion based on joint x-y distance (SPXY) algorithm was used to divide it into correction set and prediction set. Finally, the partial least squares regression ( PLSR) was used to establish a global prediction model containing data from all regions and a local prediction model for data from different regions, respectively. The effects of different numbers of collection sites in the middle part of the cob on the prediction model were further explored, and the prediction models were constructed under different numbers of collection sites (1, 2, 3, 4, 5 and 6), respectively. The results showed that the modeling results of the global prediction model, R2 p, RMSEP and RPD, were 0. 905, 0. 011% and 3. 270, respectively; the local prediction model had the best modeling effect and stronger generalization ability in the mid-section of the cob, with the modeling results, R2 p, RMSEP and RPD, being 0. 955, 0. 007% and 4. 884, respectively; when the number of collection sites was 5, the predictive accuracy of the model was optimal, with R2 p and R2 c values of 0. 967 and 0. 974, respectively. The research result showed that the scheme of choosing the mid-section of fresh maize cob and five collection sites could establish the best predictive model.

Abstract:The precise detection of high-grade kernels during walnut shell-breaking is significant for optimizing the shelling process. Traditional methods rely on manual inspection, which is inefficient and has unstable accuracy. Focusing on internal structural changes, an X-ray imaging-based method for detecting high-grade kernels was proposed. A rotating experimental platform integrating mechanical loading control and X-ray imaging was established. The shelling process was controlled by adjusting the loading speed and deformation threshold, and dual-view imaging was enabled via a rotating device. Then image processing and the YOLO v8n-seg-DN model were applied to detect and segment the kernels, and two image features were extracted from the segmented regions: the maximum kernel area proportion (Pmax) and the relative kernel block count index (RIkbc). Correlation analysis was performed between the image features and the proportion of high-grade kernels. Ridge regression was employed to predict the proportion of high-grade kernels from the extracted image features. A three-factor five-level central composite design experiment was conducted to evaluate the effects of loading speed (10 ~ 50 mm/ min), extrusion deformation (5 ~ 9 mm), and walnut equivalent diameter on high-grade kernel proportion. The results showed that the experimental platform was stable and controllable, with the average errors of loading speed and deformation within 5% . Pmax was significantly positively correlated with high-grade kernel proportion ( the correlation coefficient was 0. 931), RIkbc was significantly negatively correlated (the correlation coefficient was - 0. 926), and there was significant collinearity between Pmax and RIkbc (the VIF was 13. 067). The model validation set yielded R2 of 0. 92, RMSEp of 8. 89% , and RPD of 3. 46, indicating high accuracy. Loading speed, extrusion deformation, and walnut equivalent diameter all had significant negative effects on high-grade kernel proportion, and the predicted values closely matched the measured values, demonstrating that the developed model can effectively substitute traditional detection methods. When the loading speed was 10 ~ 30 mm/ min, the extrusion deformation was 5 ~ 6 mm, and the walnut equivalent diameter was 33 ~ 35 mm, the proportion of high-grade kernel remains above 70% , providing a technical approach for optimizing walnut low-damage shelling processes.

Abstract:The PTO torque load application test of a tractor is a key approach for evaluating power take- off performance and the reliability of field operations. Because field loads are strongly stochastic and non- repeatable due to operating-condition disturbances, extracting representative load characteristics and reproducing them in a controllable and repeatable test-bench environment is of great significance. To address the insufficient loading accuracy caused by the strong time-varying nonlinearity and dynamic parameter drift of a torque hydraulic loading system, an improved LSTM-MPC hybrid optimization control method for high-precision reproduction of PTO torque load spectra was proposed. A prediction- model-based receding-horizon MPC framework was developed, where the LSTM was employed to capture the input-output dynamics, and numerical perturbation was adopted for gradient approximation, combined with projected gradient descent to achieve online optimization, to enhance spectrum-tracking accuracy while maintaining real-time feasibility. In addition, an AMESim hydraulic loading system model incorporating refined characteristics of the swash-plate axial piston pump and the pilot-operated relief valve was established to provide a detailed representation of the internal nonlinear behaviors and to serve as the control plant. Based on an AMESim-Matlab co-simulation platform, the proposed method was compared with mainstream FNN-MPC, conventional MPC, and open-loop control. The results showed that the LSTM MPC achieved a coefficient of determination R2 of 0. 970 9 with a maximum overshoot of 5. 32% , demonstrating clear advantages in both prediction accuracy and dynamic response and offering a solution for improving the loading accuracy of tractor PTO torque spectra.

Abstract:During frequent start-stop operations and variable working conditions, pump-turbine runners are prone to complex vibration phenomena, and their modal characteristics are directly related to the operational safety and stability of the unit. To investigate the influence of rotation effects on the modal characteristics of disk structures, a fluid-structure interaction numerical approach based on acoustic- structure coupling was employed to analyze the modal behavior of a disk under different rotational speeds and fluid environment. The results showed that, in air, rotation effects caused nodal diameter modes of the disk to split from standing waves into forward and backward traveling wave modes. As the number of nodal diameters increased, the modal frequencies became more sensitive to variations in rotational speed. In contrast, nodal circle modes were scarcely affected by rotation effects. For coupled nodal diameter- nodal circle modes, frequency splitting occurred, and the coupled mode shapes were not completely dominated by rotation effects, resulting in modal displacements that firstly decreased and then increased. However, due to the weak equivalent effect of rotation in low-density air, the disk frequencies exhibited only slight variations with rotational speed, and the frequency splitting phenomenon was not pronounced. Due to the combined influence of relative rotation between the disk structure and the surrounding water as well as rotation effects, the nodal diameter modes of a rotating disk in water exhibited pronounced frequency splitting. Specifically, at a rotational frequency of 8 Hz, the backward traveling wave frequency of the (2,0) mode was increased by 8. 11 Hz, accounting for approximately 5. 9% of the stationary disk frequency. The forward traveling wave frequency of the ( 3, 0 ) mode was decreased by 9. 92 Hz, accounting for approximately 2. 9% of the stationary disk frequency. Unlike the rotating disk in the air, the frequencies of the forward and backward traveling waves varied linearly with rotational speed, and the frequency difference between them was increased linearly as the rotational speed increased. Meanwhile, both the frequency reduction rate and the added mass coefficient also showed linear growth with rotational speed. Different fluid environments exerted distinct influences on the modal characteristics of rotating disks. In seawater, the frequency splitting difference between the forward and backward traveling waves increased slightly to 16. 33 Hz. In aerated water, the speed of sound was 1 000 m/ s, and the density was 980 kg / m3, the natural frequencies of the disk was increased slightly, and the frequency splitting difference between the forward and backward waves was similar to that in pure water, at 16. 03 Hz. The findings can provide theoretical support for vibration characteristic analysis and structural optimization design of rotating disk-like structures such as pump-turbine runners.

Abstract:A force compensation algorithm based on Kane dynamics was proposed to address the problem of insufficient use of six-dimensional force sensors for robot force compensation algorithms. This method converted dynamic forces into generalized resultant forces that required compensation, and derived force compensation in a unified manner to compensate for the zero position value of the robot sensor in real-time without external forces. Firstly, the Kane’ s dynamic equation of the articulated collaborative robot was established and the velocity recursive algorithm between each joint was derived. Then, a joint type collaborative robot force compensation model was established, the force compensation algorithm was derived, and the expression for force compensation was obtained. Finally, the AUBO-I5 articulated collaborative robot was used for simulation and experimental verification, and the simulation, experimental, and theoretical results were compared, verified, and error were analyzed to obtain the force compensation comparison curve and error analysis results of the base six-dimensional force sensor during the robot??s motion process. The results showed that the comparison curves of the three directions of force / torque were consistent, and the relative error was below 5. 6% , which verified the correctness of the theoretical analysis of the force compensation algorithm. The results can provide a solid theoretical basis and valuable reference for future research in the fields of human-machine collaboration and collision detection.