基于FC-ResNet网络的隧道衬砌裂缝像素级分割方法

Pixel-Level Segmentation Method for Tunnel Lining Cracks Based on FC-ResNet Network

为提升隧道定期巡检中裂缝的检测精度和检测效率,以ResNet作为主干特征提取网络,借鉴U-net“编码-解码”和优化网络结构特征层等方法,提出一种用于隧道衬砌裂缝检测的FC-ResNet算法,实现对衬砌裂缝的像素级分割。为验证本算法的有效性和可靠性,采用CrackSegNet和U-net进行对比验证。结果表明:该算法的检测性能表现优异,测试集的像素准确率、平均交并比及F1-score分别为99.2%、87.4%、0.87,均优于CrackSegNet和U-net,且该算法的单张图片检测时间为122 ms,优于CrackSegNet,与模型结构简洁的U-net基本持平。基于提出的FCResNet算法开发隧道衬砌裂缝智能识别系统,实现对实际隧道工程衬砌裂缝准确、快速的智能化识别。

To improve the detection accuracy and efficiency of cracks during regular tunnel inspections,this study proposes an FC-ResNet algorithm for tunnel lining crack detection by using ResNet as the backbone feature extraction network,incorporating U-net's"encoder-decoder"structure and optimizing network feature layers.The algorithm achieves pixel-level segmentation of lining cracks.To verify its effectiveness and reliability,a comparative validation was conducted using CrackSegNet and U-net.The results show that the proposed algorithm demonstrates excellent detection performance,with a pixel accuracy,mean Intersection over Union(mIoU),and F1-score of 99.2%,87.4%,and 0.87,respectively,on the test set.These results are superior to those of CrackSegNet and U-net,and the detection time per image is 122 ms,better than CrackSegNet and comparable to the simpler U-net.Based on the FC-ResNet algorithm,an intelligent recognition system for tunnel lining cracks was developed,enabling accurate and fast intelligent recognition of cracks in actual tunnel engineering linings.