A Machine Learning Based Approach to Crack Detection in Asphalt Pavements

Asphalt roads are always prone to surface cracks because of the stress produced by automobiles, drastic climatic changes and improper laying of the roads. These cracks reduce the load-bearing capacity of the roads, penetrate deeper into the structures and lead to severe degradation of the roads. There are several existing methods to detect the cracks, including destructive methods and Non-Destructive Testing (NDT) methods. Non-destructive methods such as Ground Penetrating Radar based techniques are preferred now-a-days but suffer from the necessity of high-cost equipment. In this study, twelve models: six deep learning based models and six models designed by combining image processing techniques and machine learning, are proposed and evaluated for their effectiveness in detecting cracks on the road surface. Five different measures are employed for performance evaluation. It is observed from the results that deep learning based models are well suited to detection of cracks on the road surface.