Research on Fabric Defect Detection Method Based on YOLOv5

Traditional machine vision-based fabric defect detection techniques rely on manually extracting defect features, and have the problem of missed or false detections for small target defect detection. This article proposes a fabric defect detection method based on YOLOv5. By performing data augmentation on the dataset, labeling the data using Labellmg, and iterative training of the YOLOv5 network model, the optimal model obtained is used to test the test set. And compared with experimental results using Faster-RCNN and SSD models under the same conditions. The experimental results show that the YOLOv5 model has higher accuracy compared to the Faster-RCNN and SSD models. The model map reaches 64.12%, and the detection speed can meet the requirements of real-time detection. Replace the PA-Net part in the feature fusion structure of the YOLOv5 model with Bi-FPN for model optimization to improve the detection accuracy of small defects. The optimized model's map is improved by 4.16%.