Convolutional Neural Networks for Crack Detection on Flexible Road Pavements

Flexible road pavements deteriorate primarily due to traffic and adverse environmental conditions. Cracking is the most common deterioration mechanism; surveying thereof typically conducted manually using internationally defined classification standards. In South Africa, use of high-definition video images has been introduced, which allows for safer surveying. However, still a tedious manual process. Automation detection defects such as cracks would allow faster analysis networks potentially reduce human bias error. This study performs comparison six state-of-the-art convolutional neural network models purpose crack detection. The are pretrained on ImageNet dataset, fine-tuned new real-world binary dataset consisting 14000 samples. effects augmentation also investigated. Of trained, five achieved accuracy above 97%. highest recorded was 98%, by ResNet VGG16 models.