Residual plastic film on the peanut field surface has limited the subsequent large-scale production in farmland. Existing segmentation of residual plastic film is also confined to low accuracy and slow detection speed. An accurate and rapid detection is often required to assess the residual plastic film pollution. In this study, a segmentation approach was proposed for the residual plastic film on the peanut field surface using HCM-UNet. Both performance and efficiency were optimized to detect the residual plastic film on the peanut field surface under complex environments. Furthermore, MobileNetV3 was integrated into the UNet model as a lightweight backbone feature extraction network in order to enhance the overall performance of the HCM-UNet model. Meanwhile, a Haar wavelet downsampling (HWD) module was adopted to mitigate the accuracy degradation caused by the lightweight model. The loss of details was then reduced during downsampling. Secondly, the multi-scale attention aggregation (MSAA) architecture was introduced to fuse the multi-scale features, in order to improve the detection performance for the small-target residual plastic film. Finally, the convolutional block attention module (CBAM) was incorporated into the decoder to strengthen the focus on the edges of residual plastic film, thereby further improving the segmentation accuracy. A total of 1 340 images were collected from the peanut field surface. The dataset of residual plastic film was then constructed to expand the image quantity to 4 880 after data augmentation. Subsequently, the dataset was divided into 3 416 training samples, 976 validation samples, and 488 test samples in a ratio of 7:2:1. A series of ablation experiments was conducted to verify the effectiveness of the HCM-UNet model. The experimental results demonstrated that the detection accuracy was significantly improved with the MSAA feature fusion module and the CBAM attention mechanism. In addition, the mIoU of HCM-UNet was 1.30, 0.87, 2.15, and 1.90 percentage points higher than that of UNet, respectively, in the four scenarios of "Postharvest-sunny", "After plowing - sunny", "Postharvest - cloudy", and "After plowing - cloudy". The mIoU of HCM-UNet was 1.22 and 2.93 percentage points higher than that of UNet, respectively, under sunny and cloudy lighting conditions. Furthermore, the mIoU of HCM-UNet was 1.38 and 0.86 percentage points higher than that of UNet, respectively, in the two operation periods of post-harvest and after plowing. The mIoU of HCM-UNet increased by 4.38, 9.48, 3.27, 3.84, 1.65, and 5.85 percentage points, respectively, compared with the mainstream lightweight models, such as Deeplabv3, PSPNet, UNet, Segformer, Mask2former, and HRNet. Overall, the HCM-UNet model achieved an mIoU of 85.72%, an mPA of 84.26%, and an F1 score of 83.68%, with a model parameter count of 43.22 M and an inference time of 127.17 ms, indicating excellent performance in both accuracy and speed. Visual analysis confirmed that the improved model also exhibited high stability and robustness under different scenarios. There was an intelligent and high-precision detection of the residual plastic film on the peanut field surface. In conclusion, the HCM-UNet model can provide a promising solution to the accurate and rapid detection of the residual plastic film on the peanut field surface under complex environments. The finding can accurately capture the distribution, coverage, and fragmentation of the residual plastic film at different stages. Agricultural operations can be adjusted to improve the recycling machinery of the residual plastic film. Intelligent technical support can also offer to accurately assess and efficiently control the residual plastic film pollution.

A winnowing machine has been used to screen the residual film and impurities for resource utilization as post cotton harvesting. However, the manual cleaning hinders its further development after screen hole clogging, due mainly to the high labor costs and low efficiency. In this study, a multi-region dual-column unclogging system was proposed to realize the automatic cleaning operations. Firstly, the YOLOv8s model was selected for the identification after the comparative experiments on the YOLO series. Then, the pruning experiments were conducted to determine a pruning rate of 40% followed by 200 epochs of fine-tuning. The YOLOv8s-prune model was also established after verification. Finally, the identification performance of the YOLOv8s-prune was evaluated to compare with the various object detection. The results showed that the YOLOv8s model achieved the optimal performance on the validation set: precision (P) of 99.4%, recall (R) of 99.2%, mAP_0.5 of 99.5%, mAP_0.5-0.95 of 94.7%, and a model size of 21.4 MB. After that, the YOLOv8s-prune model was achieved in a precision (P) of 99.3%, recall (R) of 99.0%, mAP_0.5 of 99.5%, mAP_0.5-0.95 of 92.1%, and a model size of 10.9 MB. There was a decrease of 0.1, 0.2, 0, and 2.6 percentage points, respectively, while the model size was reduced by 50%, compared with the YOLOv8s before optimization. The YOLOv8s-prune model significantly reduced the number of parameters, computational load, and model size, compared with the rest object detection. On the test set, it was achieved in a precision (P) of 98.9%, a recall (R) of 99.0%, and mAP_0.5 of 99.4%. The coordinate information was also identified by YOLOv8s-prune. An identification control system was constructed to take a microcontroller (STM32F103C8T6) as the controller. The coordinate information was also divided into four intervals corresponding to four nozzles. The motion module traveled 360 mm per cycle was realized in the queue-based positioning between the nozzles and screen holes. Positional accuracy compensation was implemented for the motion module, according to the difference between the identified X-axis coordinate value of the leftmost screen hole and its actual coordinate value. The performance tests showed that the average recognition rate of the model was 98.3% within 0-10 min, 95.4% within 10-20 min, and 92.7% within 20-30 min, respectively, during sieving operations. The average recognition rate also exceeded 94% for the screen holes. The queue positioning test indicated that the absolute positioning deviation of the motion module ranged from a maximum of 25.6 mm to a minimum of 7 mm, with a qualification rate exceeding 94%, thus fully meeting the requirements for the cyclic operations. Orthogonal experiments on the whole machine revealed that the influencing factors on the unblocking rate were ranked in the descending order of: unblocking wind speed, operating distance, and unblocking angle. The maximum unblocking rate of 93.535% was achieved under the optimal conditions. Furthermore, the optimal combination of the rounded parameter was: unblocking wind speed of 5 m/s, unblocking angle of 11°, and operating distance of 120 mm. The average unblocking rate was 90.56% with an error of 2.975 percentage points, compared with the theoretical prediction, thus meeting the operational requirements of the film residue winnower. The findings can also provide a strong reference to develop the automatic unblocking for the film residue winnowers.

This study addressed several significant challenges encountered in the visual calibration of RGBD cameras and Delta parallel robots used in cotton topping devices. Specifically, it tackled the issues of the separation between the camera's field-of-view and the operational space, as well as the problem of insufficient calibration accuracy. An innovative approach known as Disjoint Area Visual Calibration (DAHEC) was proposed to address these issues effectively. The study explored the principles and procedures of visual calibration technology and, based on cotton planting patterns and topping requirements, selected an eye-to-hand visual calibration method. The DAHEC method was specifically developed to handle situations where the camera's field-of-view is separate from the operational space. By leveraging the principles of projective geometry and least-squares estimation, the DAHEC method simplifies the calibration process and enhances accuracy. An experimental setup was constructed to integrate the RGBD camera with a Delta parallel robot mounted on a conveyor belt, establishing a comprehensive visual relationship system. A visualization program was developed using Python and OpenCV, and comparative experiments were conducted against the traditional TSAI visual calibration method. Detailed statistical analysis was performed on average positioning errors, dispersion characteristics, and offset errors from these comparative experiments. The results indicated that the DAHEC method achieved an offset error of (4.72±0.86) mm, whereas the TSAI method had an offset error of (7.97±1.46) mm, demonstrating a clear advantage of the DAHEC method over TSAI. To further optimize the calibration process, a three-factor experiment was designed using Box-Behnken design theory, with lighting intensity, arm cumulative movement, and camera calibration board distance as experimental factors. The main objective was to ensure that the offset error remained within the acceptable range of the disk knife radius. Orthogonal tests were conducted to analyze the impact of these factors on offset error and determine the optimal working parameters for the cotton topping knife. The results revealed that the most significant factors affecting offset error were the camera calibration board distance, arm cumulative movement, and lighting intensity. The optimal working parameters were identified as a lighting intensity of 800 lux, arm cumulative movement of 99 times, and a camera calibration board distance (distance from the RGBD camera to the cotton top bud) of 300-560 mm. Under these optimized parameters, the topping verification tests showed an average offset error of 9.76 mm, which fell within the acceptable range of the disk knife radius. The topping rate was 93.75%, while the missed topping rate was 6.25%. The average time per topping was 1 800 ms, meeting the stringent requirements for efficient and precise topping. This research not only validated the effectiveness of the new DAHEC method through comparative and orthogonal experiments but also provided a scientific basis for setting working parameters in the practical application of cotton topping devices. The findings are of significant practical importance for advancing agricultural mechanization and automation, thereby enhancing agricultural productivity and crop yields. Future research will focus on exploring adaptive calibration technologies to accommodate varying environmental conditions and extending this method to other robotic applications in agriculture. By further refining and validating these methods, researchers aim to improve the reliability and applicability of robotic systems in agricultural tasks, ultimately contributing to sustainable agricultural practices and global food security.

Chopped straws can help replenish soil nutrients, improve soil structure, and increase the amount of organic matter contained in soil. The grinding ability of crop straws is influenced by the frictional characteristics of the materials involved in the grinding process. Studying the frictional properties of peanut stem, residual film, and external contact material is essential to understanding the grinding action of peanuts. This study discussed the frictional properties between films and between residual film and external contact materials. A physical test was conducted using a friction coefficient detector. The results showed that the average value of the dynamic sliding friction coefficient (f_k) was 0.34, the average value of the static sliding friction coefficient (f_s) between the film and the 40Cr steel plate (as the external contact material) was 0.38, and the average f_s value between the films and between the residual film and the external contact material was 0.36. Based on the Box-Behnken test, second-order response models were established for the static rolling stability angle (μ_e) and the static sliding friction coefficient (f_s). On the basis of establishing the static rolling stability angle (μ_e) and static sliding friction coefficient (f_s) of the evaluation index, the different friction characteristics between straw and external contact materials were investigated under varying moisture content, external contact materials and particle sizes. The study results can provide a basis for the development of equipment that can be used for peanut straw crushing and membrane separation.

This study constructed a numerical model using the discrete element software EDEM to address the current lack of calibrated contact parameters for peanut seedling membranes and the absence of precise simulation model parameters for mechanized separation. The Hysteretic Spring Contact Model (HSCM) was employed to calibrate the contact parameters of peanut seedling membranes. The angle of repose of peanut seedling membranes was determined through image processing combined with the least squares method. Through central composite design (CCD), a second-order response model linking the contact parameters to the angle of repose was established. Optimization was achieved by using the angle of repose obtained from physical tests as the objective. Secondary simulation tests were conducted with the calibrated parameters, revealing a relative error of 1.37% between the simulated and physical angles of repose. This confirmed the effectiveness of the parameters in calibrating peanut seedling membrane characteristics. The findings offer theoretical and empirical support for discrete element simulations of peanut seedling membrane separation and peanut straw pulverization processes.