Concrete Exterior Wall Crack Detection Using a Coordinate Attention-Enhanced YOLOv11 Model

Abstract

Concrete exterior wall cracks are important indicators of surface deterioration and potential durability problems in building structures. However, automatic detection remains challenging because cracks are often slender, low-contrast and easily confused with rough surface textures, stains, pores and shadows. In this study, an improved YOLOv11 model with Coordinate Attention, termed CA-YOLOv11, is developed for concrete exterior wall crack detection. The Coordinate Attention module is embedded after the SPPF module and before the neck to introduce horizontal and vertical positional information into high-level crack feature representations before multi-scale feature fusion. A dataset containing 4,028 concrete exterior wall crack images is used to train and evaluate YOLOv8, YOLOv11 and CA-YOLOv11 under the same experimental conditions. The results show that CA-YOLOv11 achieves mAP50, mAP50-95, Precision, Recall and Mean IoU values of 0.845, 0.746, 0.942, 0.805 and 0.588, respectively, while maintaining efficient inference with a latency of 5.810 ms and an FPS of 172.300. Visual comparisons further show that CA-YOLOv11 produces more continuous predictions for weak and slender crack regions. These findings indicate that coordinate-sensitive spatial feature enhancement improves crack recognition and localisation without substantially increasing computational cost.