| Abstract [eng] |
Concrete cracks pose a significant challenge to the stability of structures, and in some cases, repairs may be costly or even necessitate complete rebuilding to prevent building collapse. However, detecting and repairing damage in a timely manner can prevent many accidents. Unfortunately, detecting cracks can be difficult, particularly in hard-to-reach areas or dangerous locations. Nuclear power plants, subsea pipelines, and dams, among other facilities, require regular external and internal structural checks, which can be expensive and challenging. Recently, machine learning based approaches have been proposed for asset integrity inspections using UAV. This paper explores the use of machine learning models to detect concrete cracks, which could simplify the process and reduce the cost of regular check-ups. However, we observed that in some cases machine learning models underperforms or can be deceived by poor quality data or data with significant displacement. In this research work, we propose image enhancement techniques to improve model predictions. We believe that our proposed mathematical and machine learning algorithms will make crack detection easier, safer, and more cost-effective for regular check-ups. |
