Crack-MA: Automatic Pavement Crack Detection Based on Deep Learning

Automated pavement crack detection has always been a significant challenge in road maintenance and infrastructure management due to the complex and varied conditions of real-world pavements, including differences in texture, weathering, and lighting. Traditional methods of crack detection often rely on manual inspection, which is both time-consuming and prone to human error. Recent advances in computer vision and machine learning have enabled more efficient and accurate automated detection, but the deployment of these models in real-world scenarios still faces obstacles related to processing power, memory constraints, and user accessibility. This paper introduces a novel integrated system aimed at overcoming these challenges by enhancing both the efficiency and accuracy of pavement crack detection. At the core of the system is a user-friendly mobile application, Crack-MA, designed to make road quality assessment easily accessible to a broad range of users, including road maintenance personnel, urban planners, and even the general public. The app leverages the power of artificial intelligence (AI) to assess and detect cracks in pavement surfaces directly from mobile devices, making it a practical tool for on-the-go evaluations. An U-net architecture (teacher model) is employed to perform end-to-end training on the raw crack images. The teacher model is compressed with the distillation and channel pruning method. Finally, the pruned model is embedded into the mobile phone with Tensorflow Lite. The method is tested on public databases CFD and compared with existing methods. Experimental results show that it outperforms the other methods.