CrackNet: A new deep learning-based strategy for automatic classification of road cracks after earthquakes

Abstract

Highways are one of the most preferred transport options. Timely maintenance of highways prevents higher maintenance costs in the future. Especially detecting deterioration on highways due to major earthquakes is of great importance. Because humanitarian and logistical material aid is provided to the earthquake areas through highways. Therefore, there is a need for system applications that automatically detect asphalt deterioration. In this study, the images of asphalt cracks that occurred in five different major cities in Turkey after two major earthquakes that occurred consecutively in the Elbistan region were analyzed. These cracks were labeled as major and minor by experts from the construction department. In the next stage, asphalt cracks were categorized with a new deep learning-based model. In the study, data reliability was increased with gradient-based preprocessing steps. In the feature extraction stage, a multi-scale and multi-input customized ConvMixer (MSMICM)-based model was used. In the classification stage, a new weighted-reliefF-subspace-SVM (WRSS) algorithm was developed. This proposed approach achieved 94.2% classification performance.