A Method for Surface Defect Detection Based on Multiscale Feature Fusion and Pyramid Attention

Abstract

The two-stage defect detection model needs to pay attention to the results of the segmentation network and the classification network, and the results of the segmentation network will have an impact on the classification network. Previous models ignored shallow features in the segmentation network and used relatively simple classification networks that could not make good use of the features of the segmentation network. This paper proposes a surface defect detection algorithm based on multi-scale feature fusion and pyramid attention(MFFPA). First, a multi-scale feature fusion module is added to the segmentation network to fuse shallow features and extract more comprehensive feature information; then a pyramid attention module is added to the classification network to increase the receptive field of the model and enhance the discriminative ability of the model. The method proposed in this article was verified on four datasets, and the experimental results show that the added module can effectively improve the accuracy of the model.