Steel Surface Defect Detection Based on Improved YOLOV5 Algorithm

Abstract

Abstract When the traditional target detection method based on deep learning identifies steel surface defects, the recognition accuracy is low due to the imbalance of classification and regression tasks and the loss of feature information. To solve this problem, an improved YOLOV5 algorithm is proposed for steel surface defect detection. Firstly, the multi-branch prediction of regression and classification is decoupled, and three different outputs of regression, classification, and confidence are obtained through two different convolutions at the output end. Then, the features of different levels of the backbone network are adaptively weighted and fused, and the weighted coefficients of features of different depths are calculated by the softmax function, and then weighted and fused. Compared with the YOLOV5 algorithm, the experimental results show that the detection accuracy of the proposed algorithm is improved by 2.0%.