For crack detection, the image input into the neural network should meet the following requirements. (1) The crack pixel is brighter than the surrounding pixels. (2) There is consistency in crack size. (3) The crack is continuous. (4) The crack is close to the edge of the road. Cracks with disordered shapes and low contrast with surrounding pixels may fail to meet these requirements.
The trained model is used to detect the initial crack pixels. The fatigue cracking of concrete pavement is a long-term process with a length of several years. It is difficult to find cracks after the initial cracks. This paper uses the MFF module to improve the image quality. The crack pixel is brightened, which increases the chances of detecting the crack.
If the crack pixel is darker than the surrounding pixels, it is possible to detect cracks. However, if the crack pixel is darker than the surrounding pixels, it is less likely that cracks have been detected. Hence, the crack pixel should be brighten to increase the probability of detection. The MFF module is used to change the grayscale value of the crack pixel to the maximum value. After the MFF module, the crack pixel is brighter than the surrounding pixels.
To detect and extract edge cracks, we use the proposed method to detect cracks that are similar to wheel-path fatigue cracking. The proposed method can effectively detect cracks with good continuity and crack pixel contrast with the background, and it can improve the continuity of the crack area by enhancing the continuity of the crack. It provides a relatively complete solution for crack image detection.
As a post-processing task, crack image classification is the task of classifying the detected crack images to a specific crack category. For this task, we propose a crack image classification method using HFS. The proposed method classifies the collected crack image into several categories, using the multiple attribute decision-making of HFS to analyze each crack image and classifying the crack image to a specific category.