An Improved YOLOv7 Based Concrete Defect Detection Method

Concrete is widely used in numerous buildings, and timely detection of concrete defects is significant. Currently, the detection of concrete defects mainly relies on manual visual inspection methods. However, manual inspection has many work efficiency, accuracy, and safety issues. In recent years, with the rapid development of computer vision technology, many studies have begun to attempt concrete defect detection methods based on deep learning networks. This paper proposes a concrete defect detection method based on an improved YOLOv7 network, which enhances the detection performance of minor target defects such as cracks by integrating the ACmix attention mechanism and separation-merge operations. To increase the difficulty of defect detection, this paper constructs a concrete defect dataset containing various defect in various environments. The effectiveness of the improved method is verified through ablation experiments, and comparative experiments with Faster R-CNN, YOLOv4, SSD, and other target detection networks which demonstrated the advancement of the enhanced network.