An improved multi-label classification algorithm based on YOLOV5s

At present, the optimization of multi-label classification algorithm based on deep learning is not perfect. Based on this, this paper studies the multi-label classification algorithm based on detection. Based on YOLOv5s, an improved algorithm for image multi- label classification is proposed, which integrates SimAM attention- free module and GAM attention-mechanism effectively. Retain channel and spatial information to enhance cross-dimensional interaction, and improve the flexibility and effectiveness of the module. The experimental results show that the mean average accuracy (mAP) of the improved algorithm on the COCO dataset reaches 54.8%, which is 2.2% higher than that of the original YOLOv5s algorithm, and achieves good results in multi-label classification tasks.