Inconsistent boll maturity and uncertain environmental factors prevent large cotton mechanical harvesters from multiple batches harvesting with high quality. It is a challenge to develop harvesting robots for small patches and scattered cultivation of cotton, especially in rainy and windy cotton planting areas. To address the problem of cotton harvest in breeding fields of limited size, a novel design of cotton multi-batch harvesting robot is presented and evaluated in the field. First, after the analysis of the growth characteristics of cotton bolls, a five-picking-zones (FPZ) based end-effector, a T-bot-based three degree-of-freedom robotic arm, and cotton seed removal unit are designed to ensure the convenient picking of cotton bolls in multiple growth postures. Second, a robotic visual perception system based on depth camera and YOLOv8 detection model is developed to localize cotton bolls in complex field environments. Then, a picking sequence planning method and ROS2-based cotton-picking control process are proposed. Finally, an integrated cotton selective harvesting robot prototype is built to validate by picking assessment criteria. The test results show that the single picking time by FPZ of the end-effector ranges from 2.3 to 2.88 s, and its harvest integrity rate ranges from 97% to 98.2%. The optimal configuration of end-effector based on multivariate regression analysis results in 15 mm needle length, 15 mm needle spacing, and a motor speed of 215 rpm for the picking belt. Indoor and field experiments demonstrate the performance with an average picking time of 2.5 s indoors and 3.35 s in the field, about 78.5% of visible cotton targets were detected, of which 86.3% of the detected cotton targets were correctly identified, 86.7% of the harvest integrity rate after identification, and 94.4% of the one-shot harvesting rate during the field trials. This research highlights the potential of robotic picking to improve cotton breeding and offers a sustainable alternative to manual labor for size-limited cotton harvest.

Executable approach pose estimation for robotic apple harvesting remains challenging in natural orchards because near-spherical fruit geometry, weak growth-axis observability, and branch-and-leaf occlusion make approach-direction estimation unstable. This study proposes a growth-axis-aware apple pose estimation method that integrates growth-axis representation with spatial geometric constraints. An Oriented Deformable Box (ODB) annotation-level representation is designed from stem-calyx cues to encode the apple growth axis in 2D supervision. AppleAxisNet jointly outputs rotated detections and instance masks, with ARConv and KFIoU incorporated to improve local orientation modeling and angular regression stability. Instance masks are used to construct frustum-constrained RGB-D point clouds, and adaptive DBSCAN with prior seed points is applied to estimate the 3D fruit centroid. The 2D growth-axis orientation is then transformed into a 3D directional constraint and fused with the 3D observation vector to generate executable approach poses. AppleAxisNet achieved mAP50 values of 86.32% and 88.85% for rotated detection and instance segmentation, respectively, with an inference time of 14.2 ms. The mean 2D growth-axis angle error was 12.32 degrees, and the mean 3D centroid errors were 3.9, 4.2, and 5.6 mm at near, medium, and far ranges, respectively. In robotic tests, 94.41% of generated poses fell within the feasible approach range, and harvesting success rates of 91.00% and 80.75% were obtained on a 7-DOF manipulator and a four-arm orchard harvesting robot, respectively. These results demonstrate that the proposed method can generate accurate and executable apple approach poses in complex orchard environments.

The viable primordia in edible mushroom cultivation are the most critical factor determining yield and quality, yet its accurate quantification remains challenging due to morphological variability and low-contrast backgrounds. This study proposes an enhanced RSL-YOLOv8m-seg framework that significantly improves segmentation robustness through multi-scale feature enhancement and adaptive fusion mechanisms. Validated on a Hypsizygus marmoreus dataset containing 200,000 annotated instances, the model achieves 68.1% mAP@0.5-surpassing the baseline by 1.9 percentage points-while maintaining stable performance in industrial cultivation environments. The synergistic module integration elevates recall to 48.5% and true positive rate to 89.2%, with computational cost constrained within industrial deployment limits. Convex hull contour fitting is utilized to precisely quantify the aspect ratios of primordia. Subsequently, viable primordia are defined using an aspect ratio criterion greater than four. To the best of the researchers' knowledge, this represents the first application of deep learning for detection and analysis of Hypsizygus marmoreus primordia, providing reliable technical support for intelligent cultivation regulation.

The rapid and accurate measurement of pesticide droplet distribution in the field is essential for optimal pesticide application. To meet this need, a fast in-situ cloud-based multi-site measuring detector for canopy droplet distribution was developed. As a new approach to droplet deposition measurement in crop protection, it required extensive comparison with the conventional water-sensitive paper (WSP) method, focusing on differences between the two methods in terms of the consistency of droplet parameters in experimental tests. The study showed that the average time for droplet data acquisition was reduced to 35 s. The smallest droplet size detected by this detector was 3.86 mu m, which was 13.68 times smaller than the 52.8 mu m detected by WSP. When the WSP coverage was below 30%, the Dv0.1, volume median diameter (VMD), and Dv0.9 values measured by the detector and WSP showed similar overall trends. For coverages exceeding 30%, discrepancies increased, and WSP consistently reported larger droplet sizes than the detector, indicating that high coverage amplified the effects of droplet overlap and lateral spreading on size measurements. In parallel, droplet overlap and adhesion on WSP reduced the number of identifiable droplets, resulting in underestimated droplet density, whereas the detector continued to report higher droplet densities under laboratory conditions. In field tests, this detector showed consistency with WSP in the hierarchical droplet distribution in soybean, maize, and cotton canopies. Overall, the developed detector demonstrated promising performance in terms of consistency with WSP, and outperformed WSP in response to droplet estimation and wider spraying coverage estimation.

BACKGROUND:Uniform spray delivery within dense crop canopies is limited by upper-canopy interception, which restricts droplet transport to inner, middle and lower layers. This study evaluated a flexible canopy-opening strategy designed to form a temporary transport pathway. Soybean (Glycine max (L.) Merr.) was used as the test crop, and explicit finite element modelling, high-speed imaging and field water-sensitive paper (WSP) trials were combined to assess canopy-opening response, stem loading and WSP-measured coverage. RESULTS:Model-predicted maximum opening distance had a mean absolute percentage error of 7.37%. Within the tested range, opening depth was the factor most strongly related to maximum opening distance, which was used as a proxy for canopy opening. Increasing opening depth from 10% to 30% increased maximum opening distance from 187.15 to 385.91 mm. At 30% opening depth and 0.75 m s-1, middle- and lower-adaxial WSP-measured coverage reached 52.25% and 20.77%, compared with 18.56% and 5.23% without opening. Adaxial and abaxial interlayer coefficients of variation decreased by 62.37% and 40.15%, respectively. The simulated stem loading remained below the measured bioyield threshold. CONCLUSION:Flexible canopy opening improved spray access to poorly covered inner canopy layers by temporarily increasing canopy openness. Under the tested soybean conditions, 30% opening depth at 0.75 m s-1 produced the highest mean middle- and lower-canopy WSP-measured coverage and the greatest interlayer coverage uniformity within the tested range. These findings support flexible canopy opening as a practical approach for improving WSP-measured coverage and spray delivery potential at difficult-to-reach canopy positions. © 2026 Society of Chemical Industry.

Frequent rainfall and strong winds necessitate timely and phased cotton harvesting, thereby driving advances in robotic harvesting. However, the irregular orientation and dense distribution of cotton bolls pose a major challenge to achieving efficient harvesting with a low missed-pick rate. To address this challenge, a multi-orientation sequential picking planning method is proposed, integrating an on-the-move picking strategy and a multi-surface picking end-effector capable of harvesting cotton bolls from multiple orientations. First, the OrientCot-YOLO model is proposed for the orientation identification and localization of cotton bolls with high inference speed to form a robust foundation for next multi-surface picking plan. Then, a directional hybrid weighting method is proposed to optimize on-the-move picking sequence planning, which integrates directional and three-dimensional distance weighting for target prioritization, and identifies the optimal picking surface through normal vector matching with cosine similarity. Finally, an optimal picking path planning method based on multi-surface picking end-effector is developed. The method integrates multi-objective optimization and speed dynamic compensation to calculate the end-effector pre-picking point, and employs RRT-Connect with NURBS smoothing to generate the optimal picking path. In the gazebo simulation experiment (506 cotton bolls), compared with the Random Traversal and Euclidean Distance methods, the MOSP method improved the picking efficiency by 34.9% and 15.8%, respectively, while reducing the omission rate by 22.34% and 13.85%. In field trials (200 bolls), the omission rate was 13.7% and the picking efficiency was 6.9 s/boll, including efficiency losses caused by failed picks due to branch movement. This method improves picking efficiency and reduces the omission rate of cotton picking, providing an effective solution for on-the-move picking in complex and dense environments.

Robotic tomato harvesting in dense greenhouse clusters remains difficult because local visibility, accessibility, and contact safety change during observation and execution. Existing harvesting pipelines often separate perception from execution, which can lead to redundant views or failed harvesting attempts. This study proposes LARIS, an execution-oriented framework that integrates local-state reasoning with feasibility-guided observation–harvesting decision-making. LARIS formulates an integrated local execution-state representation to describe target geometry, neighboring local-structure evidence, free-space evidence, and residual uncertainty. Based on this state, it switches between supplementary observation and harvesting according to execution sufficiency rather than visual completeness alone, and adapts approach, contact, and retraction margins to residual uncertainty. In greenhouse experiments, LARIS achieved a harvesting success rate of 79.6% (133/167; 95% CI: 72.9–85.0%), with 1.60 ± 1.33 supplementary observations and a cycle time of 8.71 ± 1.33 s per attempt. In indoor reconstructed-cluster experiments, LARIS achieved a success rate of 84.1% (127/151; 95% CI: 77.4–89.1%). These results indicate that LARIS can improve harvesting reliability and efficiency under the tested greenhouse conditions with dense tomato clusters.

Tomatoes grow in clusters, shading each other and ripening non-uniformly even within the same cluster. This variability poses a challenge in designing an end-effector to efficiently pick fruits individually with minimal damage. A novel hybrid bending-twisting-pulling end-effector for robotic single-fruit tomato picking is designed, and the interaction dynamics between the tomato and the fruit-stem are analyzed. First, a bionic robotic tomatopicking model based on the force levels in manual operations is proposed by analyzing fruit-stem characteristics during manual picking. Next, a compact rigid-flexible hybrid end-effector, utilizing a compliant deformed belt for wrap-around clamping, is designed to apply hybrid bending-twisting-pulling forces on the stem's abscission layer to detach the fruit. Then, a multi-body dynamics model involving the stem, tomato, and end-effector is established to analyze the interaction dynamics during picking. Finally, performance criteria are presented to evaluate the end-effector's harvesting efficiency. Simulation results showed that clamping only 1/3 of the tomato's surface can still roll the target tomato from the cluster into the end-effector. Successful picking of tomatoes in various postures was achieved through the application of hybrid forces at the abscission layer, with peak bending torque ranging from 71.6 to 262.68 N center dot mm, peak twisting torque from 18.46 to 32.36 N center dot mm, and peak pulling force from 25.18 to 31.64 N. Field tests showed a tomato clamping success rate of 93.3 %, and the picking success rates for tomatoes in four different postures were 93.3 % (vertical), 90.0 % (front), 76.7 % (left-front), and 83.3 % (right-front). The average single picking time of the end-effector was 1.8 s when the flexible belt ran at 282.7 mm/s, and the fruit damage rate was 1.9 %, demonstrating the effectiveness and feasibility of this end-effector in addressing the aggregated growth characteristics and picking requirements of tomatoes.

In China, there are around 36.7 million hectares of saline–alkali lands that hold utilization potential. Precision fertilization stands as a vital measure for enhancing the quality of saline–alkali soil and promoting a significant increase in crop yields. The performance of the fertilization device is a decisive factor in determining the effectiveness of fertilization. To optimize the fertilizer utilization rate in coastal saline–alkali soils and substantially reduce fertilizer waste, it is imperative to transport fertilizers to the deep soil layers and execute layered variable-rate fertilization. In light of this, a chisel-type variable-rate layered electronically controlled deep-fertilization device specifically designed for saline–alkali soils has been developed. Extensive experimental research on its fertilization performance has also been carried out. Drawing on the principles of soil dynamics, this paper meticulously investigates the structures of key components and the operating parameters of the fertilization device. Key parameters such as the penetration angle of the fertilizer shovel, the penetration clearance angle, the curvature of the shovel handle, the angle between the fertilizer baffle and the fertilizer pipe wall, the angle between the fertilizer pipe and the horizontal plane, and the forward speed are precisely determined. Moreover, this study explores the quantitative relationship between the fertilizer discharge amount of the fertilizer applicator and the effective working width. Simultaneously, this research mainly focuses on analyzing the impact of the forward speed on the operational effect of layered and variable-rate fertilization. Through a series of field experiments, it was conclusively determined that the optimal fertilization effect was attained when the forward speed was set at 6 km/h. Under this condition, the average deviation in the fertilization amount was merely 2.76%, and the average coefficients of variation in the fertilizer amount uniformity in each soil layer were 7.62, 6.32, 6.06, and 5.65%, respectively. Evidently, the experimental results not only successfully met the pre-set objectives, but also fully satisfied the design requirements. Undoubtedly, this article can offer valuable methodological references for the research and development of fertilization devices tailored for diverse crops cultivated on saline–alkali lands.

Green asparagus has the characteristic of growing in clusters, making it inevitable for harvest targets to overlap with weeds and immature asparagus in the field. Extracting stem details in complex spatial positions information presents a significant challenge in identifying suitable harvest targets and high-precision cutting-points. This paper explored YS3AM (Yolo-SAM-3D-Adaptive-Modeling) method for green asparagus detection and 3D adaptive-section modeling using a depth camera, which could furnish harvesting path planning for the selective harvesting robots. Firstly, the model was developed and deployed to extract bounding boxes for individual asparagus stems within clusters. Secondly, the green asparagus stems within these bounding boxes were segment and generate binary mask images. Thirdly, high-quality depth images were obtained using pixel block completion. Finally, based on the cylinder, an adaptive-section 3D reconstruction method fusion with mask and depth was proposed, with a novel evaluation method applied to assess modeling accuracy. The experimental detection results of 1,095 test images demonstrated that the Precision was 98.75%, the Recall was 95.46%, the F1 score was 0.97, and the mAP was 97.16%. The modeling accuracy of 103 asparagus stems under sunny (54) and cloudy (49) conditions was estimated. The average RMSEs of length and bottom depth were 0.74 and 1.105. The detection and modeling for each stem approximately demanded 22 ms. The results of this paper indicated that the 3D model effectively represented the spatial distribution of green asparagus, and further accurately identification of suitable harvest targets and stem cutting-points. This model provided essential spatial pathways for end-effector path planning, thereby fulfilling the operational requirements for efficient green asparagus harvesting robot.

Green asparagus grows in clusters, which can cause overlaps with weeds and immature stems, making it difficult to identify suitable harvest targets and cutting points. Extracting precise stem details in complex spatial arrangements is a challenge. This paper explored the YS3AM (Yolo-SAM-3D-Adaptive-Modeling) method for detecting green asparagus and performing 3D adaptive-section modeling using a depth camera, which could benefit harvesting path planning for selective harvesting robots. Firstly, the model was developed and deployed to extract bounding boxes for individual asparagus stems within clusters. Secondly, the stems inside these bounding boxes were segmented, and binary masks were generated. Thirdly, high-quality depth images were obtained through pixel block completion. Finally, a novel 3D reconstruction method, based on adaptive section modeling and combining the mask and depth data, is proposed. And an evaluation method is introduced to assess modeling accuracy. Experimental validation showed high-performance detection (1095 field images demonstrated, Precision: 98.75%, Recall: 95.46%, F1: 0.97) and robust 3D modeling (103 asparagus stems, average RMSE: length 0.74, depth: 1.105) under varying illumination conditions. The system achieved 22 ms per stem processing speed, enabling real-time operation. The results demonstrated that the 3D model accurately represents the spatial distribution of clustered green asparagus, enabling precise identification of harvest targets and cutting points. This model provided essential spatial pathways for end-effector path planning, thereby fulfilling the operational requirements for efficient green asparagus harvesting robots.

For an efficient, low-damage selective harvesting robot for white asparagus, a parallel dual-arm control method based on moving-looking-harvesting coordination and asynchronous spears harvesting cooperation is proposed. This approach aims to enable continuous, non-stop harvesting along the ridge while minimizing damage to the spears. Firstly, a parallel harvesting mode with independent harvesting areas for the dual arms was designed to reduce collisions between the robotic arms and simplify control complexity. Secondly, an efficient dual-arm cooperative harvesting algorithm, aimed at load balancing for multiple randomly distributed asparagus, was proposed. Then, a coordinated moving-looking-harvesting control strategy was developed to synchronize the robot's movement, asparagus identification, and the harvesting operation using the two end-effectors. Finally, a prototype of the selective harvesting system was constructed, and its performance was evaluated in the field. The simulation analysis of the cooperative harvesting algorithm indicated that the shortest-time-based first-see- harvest (ST-FSH) path planning strategy outperformed two alternative methods. The dual-arm harvesting saved 45.17 % of the time and increased the harvest success rate by 3.19 % compared with the single-arm harvesting, while maintaining workload balance. Field trials demonstrated an asparagus recognition rate of 82.6 %, with an average detection time of 33 ms, a successful harvest rate of 92.3 % for recognized asparagus, and average robotic arm movement time and end-effector harvest time of 1.7 s and 5.7 s, respectively. The system achieved an asparagus damage rate of 7.2 %. The results confirm the feasibility of the proposed efficient, low-damage harvesting strategy, providing a solid foundation for the development of selective harvesting robots for white asparagus.

There are approximately 36.7 million hectares of saline alkali land available in China. To enhance the comprehensive utilization value of coastal saline alkali land and boost crop yields in such areas, it is essential to conduct research on optimizing the operational performance of high-performance soil contact components in light of the soil characteristics of coastal saline alkali land. Discrete element simulation can be employed to investigate the operational mechanisms of various key components. Nevertheless, at present, there is a dearth of discrete element models for the key physical parameters and soil structure of coastal saline alkali land soil. In this article, typical coastal saline alkali field soil was sampled, and the physical properties of the saline alkali soil, including salt content, moisture content, particle size distribution, and particle size, as well as intrinsic parameters such as soil compaction, density, Poisson’s ratio, and shear modulus, were measured. The Hertz Mindlin with Bonding contact model was employed. Physical experiments on soil accumulation angles at different depths were carried out using the cylindrical lifting method. Subsequently, by means of the discrete element method and the BBD experimental design method, a response surface model was established, and an optimization analysis was performed on the optimal parameters for the soil–soil collision recovery coefficient, static friction coefficient, and dynamic friction coefficient at each depth. Test benches for measuring the collision recovery coefficient, static friction coefficient, and rolling friction coefficient of saline alkali soil at -65Mn were set up, calculation formulas for each parameter were derived, and the contact parameters between soil at different depths and 65Mn were obtained. The results of the sliding friction angle test on different depths of saline alkali soil at -65Mn were further verified using the discrete element method, with a maximum error of 3.11%, which falls within the allowable range. This suggests that the calibration results of the discrete element simulation parameters for the interaction between soil and contact components are reliable, providing data and model support for future research on enhancing the operational performance of high-performance contact components.

Efficiency of harvesting robots remains a major bottleneck for increasing productivity, and the usual approach is to improve the parallelism between fruit separation and fruit crating. However, during robotic harvesting, achieving continuous and efficient fruit crating with low damage after fruit separation still needs to be urgently addressed, which is often overlooked but it is critical for maintaining fruit quality and harvesting efficiency. This study focuses on the dynamic mechanical damage analysis and optimization of a double-buffered post-picking crating process based on multiscale finite element model of tomatoes. First, by comparing the Burgers model and Multiscale Finite Element (MSFE) model, a framework for evaluating and optimizing tomato mechanical damage is presented. Then, the dynamic impacts and deformations of tomatoes during crating were simulated using the MSFE model, and critical damage thresholds were determined. A double-buffered cushioning system is introduced into the design of the tomato crating unit to constrain the impact velocities that may cause mechanical damage to the tomatoes. Based on the MSFE model and Response Surface Methodology (RSM), the doublebuffered cushioning structure was optimized for efficient picking, low-damage crating process, and the crating process was validated with a customized test bench. Simulation tests the crating process determined safe impact velocities for tomato-to-tomato and tomato-to-container impacts, with maximum thresholds of 1530 mm/s and 911 mm/s, respectively. The optimal crating scheme with a primary buffer height of 40 mm, a secondary buffer angle of 30 degrees, and the use of silicone pads as a cushioning material resulted in impact velocities of 1446.2 mm/s between the tomatoes at average mass and 676.45 mm/s with the container, which is well below the damage threshold. The results of the prototype tests showed that when fed at one per second, the damage rate of tomatoes was significantly reduced, the damage rate of only 2 %, and the mean crating time for tomatoes was 1.82 s which meets the robotic picking efficiency and low-damage requirements.

In China, there are approximately 36.7 million hectares of available saline–alkali land. The quality of land preparation significantly influences the yield of crops grown in saline–alkali soil. However, saline–alkali soil is highly compacted, and, currently, the market lacks land-preparation products specifically tailored to the unique characteristics of saline–alkali land. The soil crushing performance of existing power harrows fails to meet the requirements for high-quality land preparation, thus affecting crop planting yields. Consequently, it is imperative to conduct research on the design and performance improvement of the soil crushing components of power harrows for saline–alkali land. This paper centers on the key soil crushing component, the harrow blade, and conducts research from the perspectives of kinematics and dynamics. Initially, the ranges of key structural and motion parameters are determined, such as the angle of the harrow blade cutting edge, the thickness of the of the harrow blade cutting edge, and the ratio of the circumferential speed to the forward speed. Subsequently, through simulation tests integrating the Discrete Element Method (DEM) and the Box–Behnken Design (BBD), the optimal parameter combination is identified. The impact of the forward speed and the rotational speed of the vertical-shaft rotor on soil disturbance is analyzed. The relationship between soil disturbance and soil heaping is explored, and an optimal forward speed of around 6 km/h is determined. Field tests are conducted to verify the cause of soil heaping. The test results show that the soil crushing rates are all above 85%, with an average soil crushing rate of 88.66%. These test results have achieved the predetermined objectives and meet the design requirements.

The segmentation and localization of Agaricus bisporus is a precondition for its automatic harvesting. A. bisporus growth clusters can present challenges for precise localization and segmentation because of adhesion and overlapping. A low-cost image stitching system is presented in this research, utilizing a quick stitching method with disparity correction to produce high-precision panoramic dual-modal fusion images. An enhanced technique called Real-Time Models for Object Detection and Instance Segmentation (RTMDet-Ins) is suggested. This approach utilizes SimAM Attention Module’s (SimAM) global attention mechanism and the lightweight feature fusion module Space-to-depth Progressive Asymmetric Feature Pyramid Network (SPD-PAFPN) to improve the detection capabilities for hidden A. bisporus. It efficiently deals with challenges related to intricate segmentation and inaccurate localization in complex obstacles and adhesion scenarios. The technology has been verified by 96 data sets collected on a self-designed fully automatic harvesting robot platform. Statistical analysis shows that the worldwide stitching error is below 2 mm in the area of 1200 mm × 400 mm. The segmentation method demonstrates an overall precision of 98.64%. The planar mean positioning error is merely 0.31%. The method promoted in this research demonstrates improved segmentation and localization accuracy in a challenging harvesting setting, enabling efficient autonomous harvesting of A. bisporus.

Timely harvesting white asparagus depend on the detection of two forms: the emerging unearthed spear tips and the soil leaks raised by earthed spears. Accurate detection of the small spear tips and the invisible spears in the complex backgrounds of field ridges remains a challenge, especially the soil leaks with pattern similarity to the drought-induced cracks. In this paper, a novel lightweight model named HGCA-YOLO (Hyperparameter evolution-Ghost module-Coordinate Attention mechanism You Only Look Once algorithm) is proposed for accurate detection of two forms of targets. Firstly, the baseline network is determined by adopting the hyperparameter evolution to converge the network faster and obtain better parameters for spear detection. Then, the Ghost module and coordinate attention mechanism are introduced in the baseline network to decrease the complexity of the model as well as to enhance the sensitivity to the target location. In addition, the TTA (Test Time Augmentation) is introduced to the network inference to handle the targets in strongly varying environment. Finally, a dataset covered spear tips and soil leaks acquired from natural ridge is constructed, and the detection test and field experiments are conducted. The experimental results show that the accuracy of the proposed method achieved mAP 0.952 and mF1 0.924. Compared to the baseline network, this method reduced the parameters, GFLOPs and model size by 46.2%, 48.4% and 44.7%. In particular, the success rate of detection achieved 87% in the field test. This lightweight model effectively extracted the features of the cracking pattern of invisible asparagus and the small spear tips, which improved the accuracy of spear detection for selective robotic harvesting of white asparagus.

This study addresses challenges related to imprecise edge segmentation and low center point accuracy, particularly when mushrooms are heavily occluded or deformed within dense clusters. A high-precision mushroom contour segmentation algorithm is proposed that builds upon the improved SOLOv2, along with a contour reconstruction method using instance segmentation masks. The enhanced segmentation algorithm, PR-SOLOv2, incorporates the PointRend module during the up-sampling stage, introducing fine features and enhancing segmentation details. This addresses the difficulty of accurately segmenting densely overlapping mushrooms. Furthermore, a contour reconstruction method based on the PR-SOLOv2 instance segmentation mask is presented. This approach accurately segments mushrooms, extracts individual mushroom masks and their contour data, and classifies reconstruction contours based on average curvature and length. Regular contours are fitted using least-squares ellipses, while irregular ones are reconstructed by extracting the longest sub-contour from the original irregular contour based on its corners. Experimental results demonstrate strong generalization and superior performance in contour segmentation and reconstruction, particularly for densely clustered mushrooms in complex environments. The proposed approach achieves a 93.04% segmentation accuracy and a 98.13% successful segmentation rate, surpassing Mask RCNN and YOLACT by approximately 10%. The center point positioning accuracy of mushrooms is 0.3%. This method better meets the high positioning requirements for efficient and non-destructive picking of densely clustered mushrooms.

In air-assisted spraying, assisted airflow leads to flexible leaf deformation and affects distribution of droplet deposition in canopies. However, analyzing droplet deposition behavior during dynamic changes in canopies is challenging. To address this issue, a method was developed, called two-stage simulation, with one stage involving a fluid-structure interaction of assisted airflow with plant leaves and the other stage involving a discrete particle tracking simulation of droplet deposition within deformed plant canopies under the air-liquid interaction of assisted airflow and droplets. First, a representative three-dimensional (3D) plant model is developed through 3D point cloud scanning, agricultural planting parameters in the field, and plant growth characteristics. Subsequently, the deformed plant model is derived from the results of the fluid-structure interaction simulation. Finally, discrete particle tracking simulations of droplet deposition in canopies of deformed plants under air-liquid interaction are conducted. The accuracy of the simulation is verified by examining airflow distribution in the canopy and the deposition of droplet particles. The airflow verification results indicate accuracy, with a coefficient of determination of 0.8684 and a root mean squared error of 0.1463 for the linear fitted equation between the simulated and measured values. The normalized mean absolute error between the simulated and measured values is 17.2 %, indicating a favorable match between the two. The variance analysis results indicated that there is no significant difference (P > 0.05) between the simulated and measured values of droplet deposition density in the upper, middle, and lower layers. Utilizing the validated computational fluid dynamics (CFD) model, we analyze the deposition characteristics of droplets under varying airflow velocities and spray flow rates. The results highlight a direct correlation between the liquid distribution and generated airflow pattern with increased droplet deposition and drift risk observed under strong airflow and high spray flow rates. This study offers a novel approach to uncovering droplet deposition patterns through CFD simulation.