Metal surface Feature Fusion for Defect Recognition

Metal sheets have been widely used in various industrial production and production life aspects. However, the surface of metal plates is prone to multiple defects, which affect the electromagnetic characteristics and coating effect of metal plates. Currently, mainstream inspection uses YOLO, PaddleClas, and other methods to detect flat metal surfaces. These algorithms are all very potent in their classification capabilities, but as generalized defect handling algorithms, they cannot be specific to each type of defect in a real problem. In this paper, we propose a multi-information fusion method for defect recognition. We consider both low-level texture features and high-level convolutional features of metal surfaces for defect recognition. We use a convolutional neural network as the main body and add extracted and processed prompts during training. Finally, we test the proposed method extensively on Camshaft and Kaggle metal surface defects datasets, and the experimental results demonstrate that our method achieves a state-of-the-art recognition performance that surpasses previous defect recognition methods.