Deep learning for crack detection on masonry façades using limited data and transfer learning

Crack detection in masonry façades is a crucial task for ensuring the safety and longevity of buildings. However, traditional methods are often time-consuming, expensive, and labour-intensive. In recent years, deep learning techniques have been applied to detect cracks in masonry images, but these models often require large amounts of annotated data to achieve high accuracy, which can be difficult to obtain. In this article, we propose a deep learning approach for crack detection on brickwork masonry façades using transfer learning with limited annotated data. Our approach uses a pre-trained deep convolutional neural network model as a feature extractor, which is then optimised specifically for crack detection. To evaluate the effectiveness of our proposed method, we created and curated a dataset of 700 brickwork masonry façade images, and used 500 images for training, 100 for validation, and the remaining 100 images for testing. Results showed that our approach is very effective in detecting cracks, achieving an accuracy and F1-score of up to 100% when following end-to-end training of the neural network, thus being a promising solution for building inspection and maintenance, particularly in situations where annotated data is limited. Moreover, the transfer learning approach can be easily adapted to different types of masonry façades, making it a versatile tool for building inspection and maintenance.