Aiming at the problems of steel surface defect detection being susceptible to background interference and large scale information gap,an improved steel surface defect detection algorithm MN-Yolo was proposed.A global perception GC module was introduced,effectively utilizing contextual information to enhance the feature extraction capability.A multi-scale fusion module MSF was designed to integrate multi-scale feature information at a deeper level and the network detection performance was improved.The WDloss was introduced to improve the false detection and missed detection of small target defects using NWD to measure the similarity of BBoxes.Validation on NEU-DET shows that compared with the original Yolov7-tiny algorithm,the mAP,Precision and F1 of the improved algorithm are improved by 4.7％,5％,and 5.5％,respectively,and the FPS reaches 141,verifying the effectiveness of the improved algorithm.
维普
万方数据