Enhancing Road Safety: A Deep Learning Approach to Automated Crack Detection

Recently, the concept of transfer learning has been extensively utilized, demonstrating its efficacy across various tasks. Motivated by its success in image classification, this research presents a comparative analysis of road crack detection using transfer learning and Artificial Neural Networks (ANNs). Three different architecture for road crack detection were explored. First two were based on transfer learning and third one was based on Artificial Neural Network (ANN). Transfer learning with VGG16, VGG19 for road crack detection was employed and the performance of these models was evaluated and compared with the performance of ANN based model. Before exploring these model’s architecture, this work also involves other steps: collecting crack and non-crack images from three different sources and preprocessed them. The dataset consists of 42,266 crack and non-crack images. Then, the model based on VGG16 was trained on this dataset and achieved 98.55% validation accuracy. VGG19 and ANN based models were also trained on same dataset and achieved 99.01% and 98.94% validation accuracy respectively. The learning rate of these three models was set to 0.001.