A Surface Fatal Defect Detection Method for Magnetic Tiles based on Semantic Segmentation and Object Detection: IEEE ITAIC(ISSN:2693-2865)

The detection of surface defects is extremely important in manufacturing process of industrial products. However, missed detection of fatal defects during the application of machine vision technology has become a key problem in the field of surface defect detection. This paper takes surface fatal defect images of magnetic tiles as the research object, innovatively proposes a tandem method based on semantic segmentation and object detection. To avoid the confusion of model due to complex textures and boundaries on the product surface, our approach applies semantic segmentation method for fatal defects and completes the optimization of segmentation network which achieved a high mIoU of 90.43%. The comparative test results show that by segmenting the specific region where the fatal defect is located before applying object detection, detection accuracy of the fatal crack has been significantly improved, and the missed detection rate (False Negative) has been greatly reduced from 32.89% to 7.89%, indicating that this method can effectively identify fatal defect. It is of great significance for enterprises to increase productivity, improve product quality, and reduce costs.