YOLO v3-based Concrete Wall Damage Detection with Single Class Classification

Structural damage monitoring is crucial for maintaining the longevity and safety of civil structures. It provides valuable insights into the health, serviceability, integrity, and safety of structures, and helps identify potential damage at an early stage. To ensure the continuous performance of a structure, it is important to monitor the occurrence, formation, and propagation of damage. The goal of this research is to build an automated system that can efficiently and precisely find cracks in concrete walls, which is essential for guaranteeing the security and longevity of structures. The proposed method employs transfer learning on a YOLO v3 model that is fine-tuned using a dataset of labelled concrete wall surface images. The trained model is then utilized to detect and locate cracks in new images. The efficacy of the outlined approach is assessed using experiments on real-world datasets of concrete wall surface photographs. The results show that it achieves good accuracy and efficiency in detecting cracks with an average loss of 0.0784 during the training phase. This research has the potential to be applied in several fields, including civil engineering, construction, and infrastructure maintenance.