Defect Detection in Industrial Welds based on STCM R-CNN

• 1、School of Computer Science(National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing 100876

Abstract：For industrial weld defect detection based on deep learning, due to the complex defect images, there are many irrelevant features in the images, and the detection network based on a single threshold cannot deal with defects of different sizes better. For the above problems, this paper proposes a weld defect detection algorithm utilizing Swin Transformer as the backbone and a three-level cascade structure network as the detection head. In contrast to using residual networks to extract features, Swin Transformer can force the model to pay attention to the defect itself, and the cascade detection network can partially counteract the differences in the samples\' distribution. In addition, this paper introduced the data augmentations of flipping and crop-paste to enhance the size of the dataset. Experiments show that the STCM R-CNN algorithm proposed in this paper has an obvious improvement in accuracy over the classical two-stage algorithm for weld defect detection.

Keywords： computer vision defect detection Swin Transformer cascade detection network data augmentation.