Concrete Defect Detection Based on Improved YOLOv7 Network

This paper proposes an enhanced YOLOv7-based algorithm for intelligent concrete defect detection to improve recognition accuracy in complex scenarios. To address the inefficiencies and interference issues of traditional methods, a novel S<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>C-Attention mechanism is proposed, reconstructing features across style, spatial, and channel dimensions to enhance precise localization and semantic differentiation for small-scale defects. Additionally, the SPPCSPC module is optimized with multi-branch parallel structures and dynamic feature balancing, improving multi-scale feature fusion for medium and large-scale defects. Integrated into YOLOv7-Ultra, the model achieves an mAP of 87.69%, a 5.46% improvement over the baseline YOLOv7’s 82.23%