Photometric Stereo Multi-information Fusion Unsupervised Anomaly Detection Algorithm

• Lan Jianmin and Shi Jinjin
• received 04/01/2024; accepted 07/29/2024; posted 07/29/2024; Doc. ID 524199
• Abstract: Due to different materials, the product surface is susceptible to light, shadow,reflection, and other factors, coupled with the appearance of defects of various shapes and types,as well as dust, impurities, and other interfering influences, resulting in normal and abnormalsamples difficult to distinguish, is a common problem in the field of defect detection. Giventhis, this paper proposes an end-to-end photometric stereo multi-information fusionunsupervised anomaly detection model. First, the photometric stereo feature generator is usedto obtain normal, reflectance, depth, and other information to reconstruct the 3D topographicdetails of the object's surface. Second, a multi-scale channel attention mechanism is constructedto fully use the feature associations of different layers of the backbone network, and the limitedfeature information is used to enhance the defect characterization ability. Finally, the originalimage is fused with normal and depth features to find the feature variability between defectsand defects, as well as between defects and background, and the feature differences betweenthe source and clone networks are utilized to achieve multi-scale detection and improvedetection accuracy. In this paper, the model performance is verified on the PSAD dataset. Theexperimental results show that the algorithm in this paper has higher detection accuracycompared with other algorithms. Among them, the multi-scale attention mechanism and multiinformation fusion input improve the detection accuracy by 2.56% and 1.57%, respectively. Inaddition, the ablation experiments further validate the effectiveness of the detection algorithmin this paper.