Research on Recognition Method of Cashmere and Wool Mixed Fiber Based on DFS-YOLOv5

Aiming at the actual needs of simultaneous recognition of multiple fibers of cashmere and wool, a method for automatic fiber recognition using deep learning object detection algorithm is proposed. In this paper, using the YOLOv5 algorithm as the basic structure, this paper improves its detection head module, and uses the decoupled head to divide the classification prediction and bounding box positioning prediction into two independent branches, introduces the anchor-free strategy to directly predict the four parameters of the bounding box, alleviates the influence of bounding box uncertainty caused by rotating objects, and uses the SimOTA method in YOLOX to reset the positive and negative sample matching strategy, and finally proposes the DFS-YOLOv5 algorithm. The model was trained on a self-made 2000 cashmere and wool mixed fiber image dataset, and the ablation experiment and comparative analysis of various general target detection models show that the model mAP value is the highest, reaching 90.2%, and 2.9% higher than that of the mAP before improvement, which verifies the effectiveness of the method and the reliability of the algorithm, and promotes the practical application of the automatic detection technology of cashmere and wool.