A Pixel-Level Segmentation Convolutional Neural Network Based on Global and Local Feature Fusion for Surface Defect Detection

Surface defect detection (SDD) is a fundamental task in smart industry to ensure the product quality. Due to the complexity and diversity of the industrial scenes and the low contrast and tiny sizes of the defect, it is still difficult to accurately segment the defect. To overcome these issues, this research studied the pixel-level segmentation convolutional neural network based on global and local defect information for surface defect detection. Firstly, the low- and high-level features are extracted as the multi-scale network (MMPA-Net) to enrich the defect features information. Secondly, the global and local feature fusion with the global mapping branch module is developed to gradually refine the defect details to promote the detection of defects with different sizes and shapes. Thirdly, the deep supervision is applied to the global feature map and multi-scale prediction maps to train MMPA-Net. MMPA-Net has been conducted on three public SDD datasets, and the results show that MMPA-Net has achieved state-of-the-art results on the intersection of the union (IoU) by comparing with other DL methods (NEU-Seg: 86.62%, DAGM 2007: 87.94%, MT: 84.23%).