Lightweight algorithm for maintenance parts detection based on YOLOv8-OBB

In the environment of intelligent machine shop, there are maintenance tools with different shapes and placement angles. Real-time detection of the location of these maintenance tools can provide key guidance information for intelligent sorting and sorting operations. Therefore, an improved YOLOv8n-OBB algorithm is proposed in this paper, which replaces the original pyramid layer with SimSPPF to improve the convolution speed. The C2f layer in the neck network is substituted with the MSGELAN module to improve the network’s capacity for image feature learning. Experimental results reveal that the proposed algorithm achieves a 0.9% increase in mAP0.5 and a 1.6% increase in mAP0.95. Additionally, the model’s parameters, computational load, and size are reduced by 35.6%,15.7% and 32.9%, respectively. It shows the dual advantages of accuracy and computational efficiency.