CrackW-Net: A Novel Pavement Crack Image Segmentation Convolutional Neural Network

Image-based intelligent detection of road cracks with high accuracy and efficiency is vital to the overall condition assessment of the pavement. However, significant problems of continuous cracks interruption and background discrete noise misidentification are frequently observed in current semantic segmentation of pavement cracks, which mainly caused by traditional segmentation convolutional neural networks. This paper proposes a skip-level round-trip sampling block structure with the implementation of convolutional neural networks, thereby constructed a novel pixel level semantic segmentation network called CrackW-Net. After that, two datasets, including the widely recognized Crack500 dataset and a self-built dataset, were used to train two versions CrackW-Net, FCN, U-Net and ResU-Net. Meanwhile, comparative experiments are conducted among all these network models for crack detection. Results show that CrackW-Net without residual block performs the best in the task of pavement crack segmentation.

I. Introduction

With continues increasing of the demand for road maintenance, computer vision based evaluation of road conditions has become a research hotspot in the industry [1]–[3]. The research of applying computer vision technologies for pavement damage condition assessment has been developed for decades. In general, there are two branches of methodologies in this research field. The first branch is based on the traditional image processing techniques [4]. This branch of methodologies is represented by the development of various filters such as edge detectors, which is based solely on the characteristics of the optical images themselves [5]. Despite the extensive research of these methods, the shortcomings of them are obvious. It is usually very difficult for these methods to detect irregular targets under complex background environments and illumination conditions, because the methods cannot understand the meaning of the contents in the images [6].