Firstly, the current cucumber pruning operation is carried out manually. Secondly, the existing target detection algorithm is not accurate enough to efficiently detect and identify the cucumber pruning target. In response to the above issues, this article proposes the LW-YOLO algorithm based on improved YOLOv5s-7.0. Initially, FCSP is constructed to replace the C3 module in the baseline model so as to reduce parameters, speed up detection, and improve detection accuracy. Then MCACB is designed to realize downsampling, reduce parameters, effectively aggregate feature information, and improve detection performance. Finally, the EW K-means clustering algorithm is proposed to estimate the prior boxes, which further improves the detection accuracy and robustness of the model. The experimental results show that the proposed algorithm has the smallest model size, only 8.39MB; the detection accuracy is the highest, with map@.5% reaching 97.1; less computing resources are needed, and GFLOPs is only 10.0 GB; and it runs faster, with 294.1 FPS. The various indicators of the LW-YOLO algorithm are not only better than the baseline model YOLOv5s-7.0 but also better than the latest models YOLOv6, YOLOv7, YOLOv7-tiny, and YOLOv8, which meet the detection and recognition requirements of cucumber pruning robots and can be deployed.
|