Towards identification of long-term building defects using transfer learning

Detecting long-term issues on building wall surfaces, such as cracks, flakes, and roof defects, is important for timely maintenance before they become risky and expensive. This study proposes an approach named TILT to identify building defects accurately using transfer learning. The model was evaluated on real-world onsite images and achieved 98.33% accuracy with VGG16 and 79.13% accuracy with ResNet50. The results show that VGG16 performed better overall for identifying long-term structural defects.

In International Journal of Structural Engineering