Crack Detection Method on Surface of Tunnel Lining

Abstract

Crack detection on surface of tunnel lining is one of the most important tasks in concrete structure inspection field. Naked eye inspection method is widely used in general but it needs huge resources. To solve the issue, many methods have been proposed based on convolutional neural network but they show disconnected crack results with thin or blurred crack image. To overcome this problem, we propose a multiscale feature fusion method for crack detection. Experientially, results show that performance of our method was improved over the previous methods.