Industrial surface defect detection and localization using multi-scale information focusing and enhancement GANomaly

Recently, deep learning-based methods have been widely applied in identifying and detecting surface defects in industrial products. However, in real industrial scenarios, there are challenges such as limited defect samples, weak defect features, diverse defect types, irregular background textures, and difficulties in locating defect regions. To address these issues, this paper proposes a new industrial surface defect detection and localization method called multi-scale information focusing and enhancement GANomaly (MIFE-GANomaly). Firstly, skip-connection is incorporated between the encoder and decoder to effectively capture the multi-scale feature information of normal sample images to enhance representation ability. Secondly, self-attention is introduced in both the encoder and decoder to further focus on the representative information contained in the multi-scale features. Finally, an improved generator loss function based on structural similarity is designed to address the visual inconsistencies, thereby improving the robustness of detecting irregular textures. Experimental results demonstrate that the proposed method achieves superior robustness and accuracy in anomaly detection and defect localization for complex industrial data. The effectiveness of the proposed approach is fully validated through a series of comparative ablation experiments.