Copper Tube Surface Defect Detection Method Based On Improved YOLO Fusion

The rapid and accurate identification of copper tube surface defects in refrigeration equipment is important to the quality and efficiency of the whole copper tube manufacturing process.At present,the widely used artificial visual method is subjective and arbitrary,but there are some difficulties in the online detection of copper tube surface defects in industrial field,such as few data samples,similarity of some defect categories and strict real-time requirements.In view of the above difficulties,YOLO is selected as the main research method based on the comparative analysis of the current multi-target detection and recognition methods,and its anchor-based and anchor-free mechanisms are studied respectively.A surface defect detection method of copper tube based on improved YOLO fusion was proposed.Aiming at the problems of large number of model parameters and difficult adjustment of hyperparameters in surface defect detection algorithm,a surface defect detection method based on feature-enhanced YOLO(FE-YOLO)is proposed by analyzing the characteristics of the YOLOV4 algorithm based on anchor-based mechanism and the surface defect data set of industrial parts.In this method,the model structure is adjusted and optimized by lightweight,and a dense multi-scale weighted feature pyramid is designed to enhance the spatial position correlation between multi-scale detection layers.The convergence efficiency of the model is improved by the prior box selection method based on statistics and the mixed training method based on the loss function of the optimized boundary regression box.The validity of the method is verified by two industrial benchmark data sets,and experimental results show that this method is universal and suitable for large target defect detection.Aiming at the problem that there are many small target defects in copper tube surface defect data set and it is difficult to detect,the principle of YOLOX algorithm based on anchor-free mechanism and the characteristics of copper tube surface defect data set are analyzed to optimize the classification loss function and regression loss function,and a copper tube surface defect detection method based on loss function optimization YOLOX is proposed.Combined with Io U threshold selection strategy and minimum feature points selection strategy in label allocation for small target defect recognition,effectively detection and recognition of copper tube surface defects are carried out.Experimental results show that this method has strong pertinence and is suitable for small target defect detection.Finally,in order to make full use of the advantages of the proposed methods,the copper tube surface defect method based on improved YOLO model fusion is established based on multiple model fusion strategies to fuse feature enhanced YOLO and loss function optimized YOLOX models.The experimental results show that the proposed improved YOLO fusion method can effectively solve the problems of strong subjective arbitrariness and low detection efficiency in the traditional manual visual method,which has a clear significance to improve the automation level and production benefit of copper tube production process.Figures 50,Tables 18,References 81...