Automated Asphalt Pavement Crack Detection and Classification using Deep Convolution Neural Network

Asphalt pavement defects on road surface contribute one of the most important factors for traffic accident. Research on asphalt pavement using image processing techniques have been carried but there are still have challenges to the presence of shadows, oil stains and water spot. Therefore, considering the abovementioned issues, this study proposed a fully automated pavement crack detection and classification using deep convolution neural network (DCNN). First, the image of pavement cracks with dimension of 1024x768 pixels, will segmented into patches (32x32 pixels) to prepare training dataset. Next, the trained DCNN with different numbers of layers and different size of filters are employed in network. Upon the evaluation of proposed method, with respect to accuracy and processing time, the result found that the size of filters and convolution layers has an influence on the network performance. The experimental results achieved a high performance with overall accuracies above 94.25%.