Deep learning-based algorithm for multi defect detection in tunnel lining
A tunnel lining surface defect detection algorithm TDD-YOLO was proposed,for the problems of insufficient global information extraction and low detection accuracy of existing object detection algorithms in tunnel lining defect detection.The algorithm was based on the YOLOv7 framework.Firstly,MobileViT was used as the backbone feature extraction network to improve the global and local information extraction capability of the network.Secondly,Coordinate attention(CA)module was added after the upsampling and downsampling of the feature pyramid network to highlight the feature information of defects and remove the interference of background information.Finally,a convolutional module called TP Block was proposed to further improve the feature extraction ability of the network with less computation.Five algorithms,SSD,Faster-RCNN,EfficientDet,YOLOv5 and YOLOv7,were selected for comparison and analysis,in order to verify the effectiveness of the proposed algorithm.Results showed that the F1 of TDD-YOLO algorithm was 77.43％,which had an improvement of 15.58％,17.36％,12.19％,6.32％,and 6.14％,respectively,compared with the above five contrast algorithms.The mAP was 77.52％,which had an improvement of 15.20％,14.24％,9.44％,7.44％,and 6.39％,respectively.The TDD-YOLO algorithm has the highest defect recognition accuracy and the best overall performance,which is suitable for defect detection task of actual tunnel projects.
万方数据
维普
