A neural network ensemble method for effective crack segmentation using fully convolutional networks and multi-scale structured forests

Abstract

Crack image segmentation has recently become a major research topic in nondestructive inspection. However, the image segmentation methods are not robust to variations such as illumination, weather, noise and the segmentation accuracy which cannot meet the requirements of practical applications. Therefore, a neural network ensemble method is proposed for effective crack segmentation in this paper, which consists of fully convolution networks (FCN) and multi-scale structured forests for edge detection (SFD). In order to improve the accuracy of crack segmentation and reduce the error mark under complex background, a new network model based on FCN model is proposed to address the problems that lose local information and the capacity of partial refinement, which are frequently encountered in FCN model in the crack segmentation. In addition, SFD is combined with the half-reconstruction method of anti-symmetrical bi-orthogonal wavelet to overcome the limitation of crack edge detection. Finally, the result of the two maps is merged after resizing to the original image dimensions. Qualitative and quantitative evaluations of the proposed methods are performed, showing that they can obtain better results than certain existing methods for crack segmentation.