A study on drone-based detection and recognition of concrete surface cracks in tunnels using advanced imaging and machine learning techniques

The purpose of this thesis is to use drones and machine learning algorithms for automating crack detection in tunnel systems. With the high resolution RGB cameras and LiDAR sensor in drones, you get the imagery and structural data required to inspect tunnels. The images are then fed through CNNs together with SVMs for detecting and classification cracks in concrete and other surfaces. With this automated mechanism, the process will no longer need manual effort, and the inspection will be more precise and safer. The study shows the efficiency of this hybrid approach, which has 92% detection rate, much better than traditional inspection. And it is also very good at reducing false positives, and produces more trustworthy results. Crack severity is sorted into hairline, medium and deep cracks to make the process of maintenance and repairs easier. According to the results, paired with drones and machine learning, tunnel inspections become more effective, and data collection and analysis greatly enhanced. This method has potential use cases in infrastructure monitoring and could possibly be used for other structural damage detection tasks in high-dimensional domains.