Identification and Assessment of Coating Diseases on Railway Steel Bridges Based on UAV Images

The prolonged service time and natural environment of railway steel bridges lead to blistering, peeling and rust of the coating surface, which pose a potential threat to the safety of railway operations and require accurate and timely disease detection and assessment. UAVs, with their flexible aerial perspective and high operational safety, provide a powerful alternative for coating inspection. However, the current deep learning-based image recognition methods are often faced with challenges such as small target scale, multiple discrete, irregular, non-significant features of some disease edges, and complex background interference in non-coated areas when applied to this scenario. For this reason, this paper constructs a system for detecting and evaluating coating diseases on railway steel bridges. Firstly, the system effectively suppresses the background noise interference by extracting the foreground region of the coating and image slicing preprocessing technology. Subsequently, a multi-stage target aggregation algorithm based on image area-level reasoning is developed, which achieves accurate aggregation of multiple discrete similar targets in close neighbourhoods and the same disease target in adjacent sub-images; and further quantitatively evaluates the deterioration level of the disease by using the percentage of disease area.