Automatic Crack Detection on Concrete Structure Using a Deep Convolutional Neural Network and Transfer Learning

Crack formation is one of the first indicators of degradation of the concrete structure. Therefore, they must be detected early to take the necessary precautions to avoid further degradation. The current traditional inspection methods present multiple difficulties, not only in terms of accessibility and personal safety, but also in terms of cost, effort and time. Consequently, applying machine learning techniques to detect cracks has drawn much attention recently. In this paper, we propose to use a pre-trained Convolutional Neural Network architecture to automatically detect and segment cracks through transfer learning. The model is trained on a dataset containing images with multiple crack forms in different brightness conditions. It is evaluated using several performance metrics. The model yields relatively good performance with an area under the receiver operating characteristic and an F1-score of 93.57% and 73.13%, respectively.