Using A CNN to Solve the Problem of Asphalt Pavement Crack Detection.

How to quickly and accurately detect cracks in asphalt pavement for pavement maintenance is of great significance for ensuring the safety of vehicles and personnel. Current research mainly focuses on feature extraction or edge detection based on digital image processing. However, these features cannot cover all the cases encountered in complicated road situations. In recent years, convolutional neural networks have received more attention due to their ability to automatically extract features from image data, and there are studies describing the application of convolutional neural networks (CNNs) for pavement crack detection. However, while these studies apply CNNs for the classification of small image patches, they lose the structural information of cracks. This work designs an end-to-end context aware convolutional network architecture that has a large receptive field to leverage the contextual information in large areas to help classify central areas. The experimental results showed that our architecture is robust to image noise and is more accurate than traditional methods. Most importantly, the predicted cracks have better structural preservation.