A method for concrete crack detection based on improved YOLOv8s

Cracks are one of the main defects in concrete structures, and timely detection and management of cracks are crucial for maintaining the stability and safety of these structures. Addressing the issues of low accuracy and efficiency in existing deep learning object detection models applied to concrete crack identification, this paper proposes a method based on an improved YOLOv8s for concrete crack detection. Initially, the C3STR is used to replace part of C2f in the backbone network of YOLOv8s to enhance the extraction of complex global features; secondly, a CA attention mechanism is introduced to improve the feature extraction across different layers; finally, the SPPFCSPC replaces the SPPF spatial pyramid pooling module to enhance feature fusion across various spatial hierarchies. The experimental results demonstrate that compared to conventional YOLOv8s, this method enhances detection precision by 3.09% to 94.34%, with a detection speed of 224 frames per second. Additionally, it allows for the quantitative calculation of crack length and average width.