A novel crack detection algorithm of underwater dam image

Abstract

A novel algorithm is introduced for the deficiencies of underwater dam image crack detection. The algorithm makes use of the intensity values of 2D image to generate a 3D spatial surface, which is regarded as a concave-convex ground with “pits” and “ditches”. The “pits” represent the noise pixels and the “ditches” represent the crack pixels. The cracks that are difficult to describe in 2D image can be regarded well as ditches in the 3D spatial surface. Then by analyzing the characteristics of ditches space curvatures, the space detected method is used to get the ditches information, which is mapped to 2D surface as the crack. Because the detected result contains some noise and fake cracks, so BP neural network is adopted to identify crack object. As a result, the crack information is detected successfully.