Crack Detection of Masonry Structure Based on Infrared and Visible Image Fusion and Deep Learning

From the standpoint of protecting and repairing the ancient city walls, this work aims to improve the feasibility and accuracy of crack detection in the brick wall background. Data sets of cracks in the surface of the ancient city walls were created, including RGB and thermal images. By using deep learning techniques, the best combination of data input type and network architecture were explored in the CNN-based training framework. The main contribution of this paper is: (a) a comprehensive dataset of cracks in the background of ancient city walls, including RGB images and infrared images; (b) a comparative analysis of crack detection results of different data fusion methods under different deep learning networks. Based on the results, the optimal data input and training network combination were identified for masonry wall crack identification, which enables an automatic crack damage detection for ancient city wall.