A STUDY OF THE INTELLIGENT RECOGNITION OF CONCRETE-STRUCTURE CRACKS BASED ON YOLOv7 AND C2E-Net

Cracks, a common problem in concrete structures, severely compromise their safety and reliability. Accurate and efficient crack identification across diverse environmental conditions is pivotal for ensuring the safe operation and health monitoring of buildings. Accordingly, this paper innovatively proposes a novel intelligent crack-recognition model for concrete structures by integrating YOLOv7 and C2E-Net. First, crack features are extracted via the CA, ELG, and CCF modules of C2E-Net. These features are then inputted into an enhanced YOLOv7 model for fusion, which improves the recognition capability of tiny cracks. Additionally, the Inner-CIOU loss is introduced to optimize small-target detection, addressing the issue that traditional loss functions struggle to accurately identify small targets such as tiny cracks. Following the model’s refinement, we carry out ablation experiments. These experiments aim to assess the contribution of each functional module in the model’s structure and to uncover how the internal components work together. Finally, to validate the model’s performance, this paper selects SSD, Faster-RCNN, and Ghost-YOLO as contrast models and employs multiple evaluation metrics, including the Dice coefficient, IoU, detection accuracy, recall rate, and F1 score, for comprehensive analysis. The experimental results demonstrate that the C2E-Net-YOLOv7 model for concrete-structure-crack intelligent recognition performs exceptionally well during both the training and testing phases, with a Dice coefficient of 0.83, IoU of 0.81, detection accuracy of 91 %, recall rate of 89 %, and F1 score of 0.91. Compared to the traditional SSD model, the Dice coefficient increases by 50.91 %, IoU by 88.37 %, and detection accuracy by 59.65 %; relative to the Ghost-YOLO model, the Dice coefficient improves by 9.21 %, IoU by 22.73 %, and detection accuracy by 5.81 %. The experiments indicate that the YOLOv7 model with the C2E-Net module introduced achieves performance improvements in concrete-structure crack-detection tasks, breaking through the bottlenecks of traditional models and demonstrating remarkable superiority and effectiveness. This model offers an efficient and accurate method for intelligent crack recognition, helping to detect safety hazards promptly and ensuring the long-term stability and safety of buildings.