Anomaly Detection Method Using Feature Reconstruction Based Knowledge Distillation

作     者：ZHU Xin-yu SI Zhan-jun ZHANG Ying-xue 朱新雨;司占军;张滢雪

作者机构：College of Light Industry Science and EngineeringTianjin University of Science and TechnologyTianjin 300457China College of Artificial IntelligenceTianjin University of Science and TechnologyTianjin 300457China 

出 版 物：《印刷与数字媒体技术研究》 (Printing and Digital Media Technology Study)

年 卷 期：2024年第4期

页      面：115-124,236页

学科分类：081203[工学-计算机应用技术] 08[工学] 0835[工学-软件工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Feature Reconstruction Anomaly Detection Distillation Mechanism Industrial Production 

摘      要：In recent years,anomaly detection has attracted much attention in industrial *** traditional anomaly detection methods usually rely on direct comparison of samples,they often ignore the intrinsic relationship between samples,resulting in poor accuracy in recognizing anomalous *** address this problem,a knowledge distillation anomaly detection method based on feature reconstruction was proposed in this *** distillation was performed after inverting the structure of the teacher-student network to avoid the teacher-student network sharing the same inputs and similar *** was improved by using feature splicing to unify features at different levels,and the merged features were processed and reconstructed using an improved *** experimental results show that the proposed method achieves better performance on the MVTec dataset,verifying its effectiveness and feasibility in anomaly detection *** study provides a new idea to improve the accuracy and efficiency of anomaly detection.