Debonding defect quantification method of building decoration layers via UAV-thermography and deep learning

Abstract

The falling offs of building decorative layers (BDLs) on exterior walls are quite common, especially in Asia, which presents great concerns to human safety and properties. Presently, there is no effective technique to detect the debonding of the exterior finish because debonding are hidden defect. In this study, the debonding defect identification method of building decoration layers via UAV-thermography and deep learning is proposed. Firstly, the temperature field characteristics of debonding defects are tested and analyzed, showing that it is feasible to identify the debonding of BDLs based on UAV. Then, a debonding defect recognition and quantification method combining CenterNet (Point Network) and fuzzy clustering is proposed. Further, the actual area of debonding defect is quantified through the optical imaging principle using the real-time measured distance. Finally, a case study of the old teaching-building inspection is carried out to demonstrate the effectiveness of the proposed method, showing that the proposed model performs well with an accuracy above 90%, which is valuable to the society.