Research on Fatigue Crack Detection Method of Asphalt Concrete Pavement Based on Machine Learning

Abstract

Based on the original research results, this paper proposes a method for fatigue crack detection of asphalt concrete pavement based on machine learning. Obtain the asphalt concrete pavement fatigue crack data set composed of public data sets and manually collected picture data, and divide the crack pictures into three types: horizontal crack pictures, longitudinal crack pictures and reticular crack pictures. The crack image is pre-processed including grayscale processing, histogram equalization operation processing, noise preprocessing, and Gamma correction processing. After the preprocessing operation, the subsequent target segmentation, extraction and positioning of the crack image are performed. Since there are fractures in the extracted fracture image, it is necessary to splice multiple fracture connected region targets into a complete fracture connected region. On this basis, the fatigue crack detection network model of asphalt concrete pavement is built based on machine learning. The collected data are input into the model to train the model, and the fatigue crack detection of asphalt concrete pavement is realized. The experimental results show that the detection accuracy of this method is high for all kinds of fracture types, and the detection effect is ideal when the model training iterations to 10000 times.