Due to high-speed operation, the production performance and indicators of rice plug tray precision seeders, such as tray qualification rate, over-sowing rate, and missed-sowing rate, are difficult to monitor in real time. In addition, failures such as air-suction pipe duct damage cause significant fluctuations in these indicators, which severely affect production continuity. To address these challenges, an intelligent solution integrating rice plug tray seeding recognition and counting with air-suction duct fault prediction was proposed. At the system architecture level, the overall high-performance monitoring framework for rice plug tray precision seeding was designed, in which the relationship between pneumatic suction holes and seeding qualification rate was analyzed, and the operating principles were systematically explained. Based on this design, an advanced detection approach was adopted, where the YOLOv8+LADH+NWD (You Only Look Once version 8 - Lightweight Asymmetric Dual-Head - Normalized Gaussian Wasserstein Distance) algorithm was employed to achieve seeding recognition in rice plug trays. By optimizing both the model architecture and the loss function, the algorithm’s recognition accuracy and counting efficiency were significantly enhanced under high-frequency operational conditions. Real-time statistical counting of production indicators such as tray qualification rate, over-sowing rate, and missed-sowing rate was achieved through the recognition results. Through statistical analysis of these indicators, it was observed that failures such as air-suction duct damage lead to considerable fluctuations in production performance. Such faults were identified as one of the primary factors influencing equipment stability and operational efficiency. To realize effective prediction of air-suction duct failures in rice plug tray precision seeders, a predictive model based on the Bi-Gate Convolutional Unit with Multi-Head Residual Self-Attention (BiGCU-MHResAtt) was proposed. This model was designed to address the issue of frequent failures that result in extensive short-term loss or absence of production data. By incorporating adaptive learning strategies and cross-condition robustness mechanisms, the model was further enhanced to ensure stability across diverse operational environments. By leveraging the integration of gated convolutional mechanisms with residual self-attention, robust temporal and contextual dependencies were captured, enabling accurate failure forecasting even in environments with noisy and incomplete datasets. On the basis of these methods, a high-performance monitoring system for rice plug tray precision seeding was developed. The proposed system was validated through comprehensive experimental studies, which confirmed the accuracy and reliability of both the recognition and prediction models. The system successfully enabled accurate recognition of rice plug tray seeding, efficient real-time counting of production indicators, and reliable fault prediction of air-suction ducts during high-speed operations. This achievement not only mitigated the risks associated with production instability but also contributed to the advancement of intelligent agricultural machinery. The presented research demonstrates the potential of integrating advanced computer vision algorithms with predictive modeling in agricultural machinery monitoring. By combining YOLOv8-based lightweight detection with BiGCU-MHResAtt-driven failure prediction, a holistic monitoring solution was achieved, capable of ensuring production stability in high-frequency seeding environments. The proposed framework therefore provides not only technical innovation but also a practical foundation for scaling intelligent seeding systems to large-scale agricultural production. The contribution is expected to provide significant support to the intelligentization of rice seeding machinery, promoting greater automation, reliability, and sustainability in modern agricultural practices.

With the nature of the high wind and sand in western China, the Chinese wolfberry recognition shows a strong relationship with the sandy noise and needs a high-accuracy algorithm. To address this issue, this study aimed to develop an algorithm for accurately detecting and recognizing wolfberries. YOLOv8, an algorithm promoted by Ultralytics, supports image classification, object detection, and instance segmentation tasks. To enhance the performance of the original YOLOv8 model, a novel YOLOv8 algorithm incorporating FasterNet, RepBiFPN, and Lightweight Asymmetric Dual-Head was proposed. Firstly, thousands of Chinese wolfberry images were collected from the Ningxia Academy of Agriculture and Forestry Science, China, and random noises were added to simulate the wind and sand conditions typical of spring. Secondly, leveraging the advantages of YOLOv8n, such as its high speed and accuracy, this research innovatively integrated the FasterNet block into the C2f module of YOLOv8 to improve the effective handling of data uncertainty and noise. Additionally, an innovative RepViT+BiFPN, a new detective head, and a Lightweight Asymmetric Dual-Head were introduced to improve the training efficiency of the YOLOv8 network. Finally, to evaluate the effectiveness of improved YOLOv8 for the recognition of wolfberry, the dataset of wolfberry images was divided into a training set, a validation set, and a testing set to assess the performances of different models. Experiment results demonstrate that the YOLOv8-FasterNet+LADH+RepBiFPN model outperforms other models in terms of mAP@0.50-0.95, achieving a 4.5% improvement on the validation set compared to the original YOLOv8n. This research addresses the high-speed and accurate recognition of the Chinese wolfberry under strong winds and sand noise through algorithmic improvements and integration, which can facilitate the automation and intelligence of Chinese wolfberry harvesting and contribute to the advancement of agricultural mechanization.

In order to realize the efficient and high-quality mechanical picking for Chinese wolfberry, firstly, the forced reciprocating vibration picking principle of the Chinese wolfberry branch was studied, and the mechanical model of vibration picking was established based on the simplified cantilever model, and the response analysis and solution of all positions for the branch were carried out. At the same time, the critical mechanical model of fruit detachment under the condition of fruit hanging on branches was established, and the theoretical values of inertia force for each component of the branch were obtained. Secondly, through actual measurement and finite element modeling, the natural frequency and forced vibration response simulation for each component of the branch of Chinese wolfberry terminal branch model were both studied, and the relationship between single-point periodic excitation force and high-quality fruit shedding parameters was obtained. Thirdly, according to the conclusion of the picking model, a test bench with many groups of adjustable parameters was built. Finally, the last branch of fruit-hanging Chinese wolf berry for Ningqi No.1 was taken as the experimental object, a four-level orthogonal experiment was designed with three factors: frequency, amplitude and entrance angle. Meanwhile, the net picking rate, damage rate and false picking rate were taken as the evaluating indicators, referring to the comprehensive scores of the three factors. It was concluded that the primary and secondary relations of factors affecting the picking effect are frequency, amplitude and entrance angle, and the best operation parameters are frequency of 20 Hz, amplitude of 15 mm, and entrance angle of 45°, then, a hand-held vibration picker with setting parameters was trial-produced, and the optimal parameter combination was verified in the Chinese wolfberry planting base of the National Chinese wolfberry Engineering and Technology Research Center. The results showed that the net picking rate of ripe Chinese wolfberry was 96.13%, the damage rate of fruit was 1.13%, and the false picking rate was 3.23%, mechanized picking efficiency was 30.28 kg/h, which is 6.65 times that of manual picking. The experimental results are consistent with the simulation results. The research results can provide an important basis for the creation and operation standards of large-scale Chinese wolfberry vibration harvesting equipment.

Real-time detection of kiwifruits in natural environments is essential for automated kiwifruit harvesting. In this study, a lightweight convolutional neural network called the YOLOv4-GS algorithm was proposed for kiwifruit detection. The backbone network CSPDarknet-53 of YOLOv4 was replaced with GhostNet to improve accuracy and reduce network computation. To improve the detection accuracy of small targets, the upsampling of feature map fusion was performed for network layers 151 and 154, and the spatial pyramid pooling network was removed to reduce redundant computation. A total of 2766 kiwifruit images from different environments were used as the dataset for training and testing. The experiment results showed that the F1-score, average accuracy, and Intersection over Union (IoU) of YOLOv4-GS were 98.00%, 99.22%, and 88.92%, respectively. The average time taken to detect a 416×416 kiwifruit image was 11.95 ms, and the model’s weight was 28.8 MB. The average detection time of GhostNet was 31.44 ms less than that of CSPDarknet-53. In addition, the model weight of GhostNet was 227.2 MB less than that of CSPDarknet-53. YOLOv4-GS improved the detection accuracy by 8.39% over Faster R-CNN and 8.36% over SSD-300. The detection speed of YOLOv4-GS was 11.3 times and 2.6 times higher than Faster R-CNN and SSD-300, respectively. In the indoor picking experiment and the orchard picking experiment, the average speed of the YOLOv4-GS processing video was 28.4 fps. The recognition accuracy was above 90%. The average time spent for recognition and positioning was 6.09 s, accounting for about 29.03% of the total picking time. The overall results showed that the YOLOv4-GS proposed in this study can be applied for kiwifruit detection in natural environments because it improves the detection speed without compromising detection accuracy.

In order to find out the matching principle of excitation force and energy consumption of reciprocating vibrating Chinese wolfberry picking device, the energy consumption mechanism of reciprocating vibrating Chinese wolfberry picking device is studied. According to the structural characteristics and operating principle of the picking device, the no-load and load movement and force characteristics of the crank bearing and vibrating component are analyzed, and the theoretical model of energy consumption of the reciprocating vibration picking device is jointly constructed, and the simulation analysis is carried out. The results show that the vibrating component and load mass have a significant influence on torque, the load air resistance phase has a significant effect on torque, and the load air resistance and friction coefficient have no significant influence on torque. Subsequently, by building an AC servo motor torque detection system and a torque sensing detection system, verification experiments are carried out, the maximum torque of the preset system is 1.3 N∙m, the rated power is 400 W, the motor frequency is 20 Hz, the amplitude is 15 mm, and the total mass of the vibrating component is 0.143 kg. Test results show that, the no-load operation, the change trend of detected torque is consistent with simulation, the torque model is verified to be accurate. The maximum torque of simulation and detection are 0.52 N∙m and 0.57 N∙m respectively, and the error between test and simulation is 9.6%. For load operation, the maximum torque of five groups of branch loads of 20 g, 60 g, 100 g, 140 g, 180 g and 220 g are detected to be 0.73 N∙m, 0.74 N∙m, 0.75 N∙m, 0.82 N∙m and 0.83 N∙m, respectively, and the relationship model between load and torque is obtained by fitting. The research results can provide a theoretical basis which can configure a suitable motor in the reciprocating vibration Chinese wolfberry picking device with a certain load limit.