An Application of Thermography in Building Facade Inspection based on YOLOv3

In various cities around the world, the inspection of old buildings is a crucial task. The thermography is one of the most essential Non-destructive Techniques (NDT) in building inspection. The traditional visual inspection, hammer tapping inspection and other methods are inefficient a nd require direct contact with the facade compared to thermography. You Look Only Once version 3 (YOLOv3) is a popular one-stage deep learning object detection model. This paper proposes an efficient approach for building inspection using the YOLOv3 to process the on-site infrared images of the building facade. The infrared images could directly detect the defects in the concrete structure remotely while the visible RGB images could only detect the surface defects. With the using the deep learning technology, the health condition of the building could be fast and precisely diagnosed. The real on-site infrared images are collected and tested in this research to demonstrate this application in real building inspection. The experimental results demonstrate the trained model is able to detect the defects in the on-site infrared images.