GSC-YOLO: a lightweight network for cup and piston head detection

摘要

In order to improve the detection effect of the cup and piston head during the operation of the automatic cupping machine and reduce the blurring effect of the robotic arm movement on the collected images, we propose a network called GSC-YOLO. Firstly, we introduce GhostNet as the backbone network and SimSPPF as the spatial pyramid pooling method to improve the network lightweight, and then we add the CA attention mechanism to strengthen the feature extraction ability of the network and expand the receptive field. Secondly, we add a small target detection layer to enhance the detection ability of the piston head. Finally, the data enhancement is carried out by affine transformation and motion blur processing of the dataset, which improves the robustness of the model for long-distance and vibration environment detection. The experimental results show that in the comparison model, GSC-YOLO has the smallest size, the best detection effect, and faster detection speed, where, the detection accuracy is 0.967, the recall is 0.945, and the inference speed of each image on the CPU and GPU is 175.1 ms and 12.4 ms, so this model can be used for real-time detection of cup bodies and piston heads.

关键词