Steel Surface Defect Detection Based on Improved YOLOv7

In order to solve the problem of high misdetection and leakage rates in steel surface defect detection, we propose an improved algorithm based on YOLOv7. The algorithm introduces Swin- Transformer's sliding window encoder into the YOLOv7 structure to capture the contextual information and global dependencies of steel surface defects at different scales, so as to fully learn the semantic features of the target. Meanwhile, we utilize the ACmix attention module to improve the network's attention to small targets, which effectively solves the problem of the original model's omission detection of small targets. Finally, the lightweight convolutional technique GSConv is used in the feature fusion part of the model to reduce the computational complexity of the model. The experimental results show that on the NEU-DET dataset, the detection accuracy of the algorithm reaches 78.5%, which is 4.00/0 higher than the original YOLOv7 algorithm. Compared with the traditional algorithm for detecting steel surface defects, this algorithm has higher detection accuracy and detection rate, which can play an important role in industrial applications.