Real-Time Tunnel Crack Analysis System via Deep Learning

Cracks in the tunnel become an unavoidable problem in tunnel construction and tunnel using. Cracks will affect the stability of the tunnel and have a negative impact on the operation of the train. It is a crucial part of rail safety as well as rail defects and train defects. Therefore, cracks in the tunnel must be identified and repaired in time. At present, the detection of tunnel cracks in the domestic railways relies on the manual inspection mainly. It is difficult to satisfy the requirements of the rapidity and the accuracy of railway inspection by the manual inspection due to the subjective judgment of the inspection personnel. At the same time, tunnel images have some complex situations such as water stains, scratches, structural seams, uneven illumination, and a lot of noise, which have brought bottlenecks to the development of traditional image processing methods. It is necessary to adopt more effective methods to detect the tunnel cracks in time. This paper builds the first tunnel crack dataset with semantic segmentation annotation and proposes an objective and fast tunnel crack identification algorithm using semantic segmentation in computer vision to construct a complete tunnel crack identification and analysis system. The system applies advanced semantic segmentation to the railway tunnel image analysis to achieve precise segmentation of tunnel crack locations, thereby saving the railway department a lot of manpower and material resources and improving efficiency.