Crack Identification Algorithm Based on MASK Dodging Principle and Deep Learning

As a large public infrastructure, roads are related to the economic operation of the entire country and the safety of society. With the dramatic increase in the number of vehicles, the demand for transportation and the number of highway miles, road diseases have also increased dramatically. If inspection and maintenance work cannot be performed in time, pavement cracks will rapidly deepen and widen over time, which will cause serious safety accident. At present, automatic crack identification methods proposed in academic research mainly focus on ideal clean images in the experimental stage, ignoring the interference factors such as shadows, stains and uneven illumination, which are common in actual road surfaces, so their recognition accuracy is greatly reduced when applied in actual situations. In this paper, we propose a road crack identification algorithm combining the illumination homogenization algorithm and the deep convolutional neural network method. Experiments show that our proposed method can improve the accuracy of crack recognition in actual road images, and Dice similarity coefficient of crack identification is improved to 0.7937.