Research on low contrast surface defect detection method based on improved YOLOv7

Aiming at the difficulty of defect detection caused by the low contrast between defects such as scratches, deformation and foreign bodies on the surface of parts and the background, and the defects are greatly affected by the surrounding light, an accurate recognition method of low contrast defects based on improved YOLOv7 is proposed. A fusion Mosaic and MixUP online data enhancement method is proposed to expand the training sample data. The GAM attention module is added to the backbone network to enhance the feature extraction ability of low contrast defects, and SIoU loss function is used to focus on the accuracy of the model to accelerate the convergence speed of the model, and the fast suspected defect location is realized based on multi-camera. After focusing on the suspected defect position, the defect features are enhanced and accurately identified by rotating the 6RSS mechanism. Experiments show that the SIoU-YOLOv7-GAM algorithm shows better performance than the original YOLOv7 algorithm, and the average accuracy and recall rate are increased by 2.92 % and 5.02 %, respectively. The proposed multi-camera focusing detection method has a high recognition accuracy for low-contrast defects on the surface, and can eliminate the problem of defect error recognition to achieve accurate detection of low-contrast defects on the surface of parts. Authors