YOLOv11-PC: an intelligent detection method for concrete structure defects

Abstract Concrete structures are extensively applied in a wide range of infrastructure projects. However, concrete structures are susceptible to cracks, water leakage, and water seepage due to prolonged exposure to severe environmental conditions. Accurate detection of these defects is vital for preserving structural integrity and safety. To solve this problem, we introduce the YOLOv11-PC algorithm for detecting defects in concrete structures. We introduce pinwheel-shaped convolution to enhance feature extraction of low-contrast targets, the local importance-based attention mechanism to optimize feature selection, and Shape-IoU as a boundary box loss function to improve target localization precision. The experimental results demonstrate that the precision of the YOLOv11-PC model is 95.9%, recall is 94.2%, and mAP@0.5–0.95 is 76.3%, significantly surpassing existing methods. This research offers an effective technological solution for the intelligent detection and repair of concrete structures.