Intelligent Detection of Tunnel Lining Cracks Based on Self-attention Mechanism and Convolution Neural Network
The inherent localization of the convolutional operation makes it difficult to capture the global feature repre-sentation of the cracks.To address this problem,proposes a high-precision tunnel lining crack intelligent detection algo-rithm,ST-YOLO,based on the YOLOv5 network framework by fusing the self-attention mechanism and convolutional neural networks(CNN).In ST-YOLO,a double-branched feature extraction module was designed.In the first branch,Swin Transformer based on self-attention mechanism was used to extract the global features of cracks,while in the second branch,CSPDarkent based on convolution neural network was used to extract the local details of cracks.Furthermore,the convolution attention enhanced feature fusion module was used to fuse the crack features extracted from the two bran-ches on multiple scales,so as to achieve the complementarity of the global features and local features.Finally,the clas-sification confidence and location results of cracks were obtained by using the decoupled head of YOLOv5.To verify the effectiveness of the ST-YOLO,five target detection algorithms based on CNN,including SSD,YOLOv4,YOLOv5,Effi-cientDet and Faster-RCNN,were selected for comparative analysis.The results show that based on the tunnel lining crack data set constructed in this paper,the F1 score and AP of ST-YOLO model are 83.95％and 85.26％,respectively.Compared with the five comparison algorithms,ST-YOLO has higher recognition accuracy and better environmental adaptability,which is more suitable for the crack detection task for tunnels.
NETL
NSTL
万方数据
