Cross-category general industrial defect detection method, system, terminal and medium

Abstract

This invention provides a cross-category general industrial defect detection method and system. The method includes: providing a pre-trained feature extraction network to obtain initial features and corresponding reference features of input samples; providing an industrial defect detection training model, which, based on the reference features and the initial features of training samples, obtains residual features and performs feature perturbation, and is trained through cross-class domain invariance constraints and normal residual feature distribution to obtain a trained industrial defect detection model; using the initial features of the sample to be detected and its corresponding reference features as input to the trained industrial defect detection model to obtain an anomaly score for the sample to be detected, which is then used for industrial defect detection. This invention is category-independent, not limited by cross-class boundaries, and can be used for new categories without retraining and fine-tuning, making it suitable for industrial defect detection with diverse product categories and continuously evolving new technologies.

Images