Research Of Tunnel Cracks Recognition And Analysis Based On Deep Learning Algorithm Cascade

The existence of cracks will affect the stability of the tunnel.Hence,it is of vital importance to obtain timely identification and treatment of tunnel cracks in ensuring railway traffic safety.As the amount of tunnel images collected by high-speed cameras is extremely large,manual method is unable to meet the actual needs in recent years while traditional image processing methods become powerless to achieve a satisfying speed and accuracy.This thesis studies tunnel cracks recognition and analysis algorithms based on deep learning.It applies image classification and semantic segmentation methods to the field of tunnel image analysis.A tunnel crack analysis system is designed and implemented,which improves the efficiency of tunnel detection work.This thesis firstly establishes a tunnel image dataset in line with deep learning standards,which satisfies the training,testing,evaluation and optimization of image classification model and semantic segmentation model.Secondly,tunnel cracks classification algorithm is designed and implemented.The classification model is constructed using ResNetl 8.Training and testing process are finished on Pytorch framework.The classification model achieves 98%accuracy,87%precision and 8 1%recall on tunnel image dataset.Then,tunnel cracks segmentation algorithm is designed and implemented.SPPNet is proposed to improve the effect of feature aggregation process of tunnel cracks.Also,optimization schemes such as weighted loss function are introduced to further improve the accuracy.The result of the segmentation model is that,cracks segmentation IoU is 42.16%,Mean IoU is 41.48%,and mean Pixel Accuracy is 98.05%.Finally,classification and segmentation algorithms cascade strategy is proposed and a system software for tunnel cracks analysis is inplementated.