Industrial Anomaly Classification Method Based on EMD Metric Loss
Xiaoyuan Liu, Jiarui Liu, Jinhai Liu, Xiangkai Shen, ZHANG Baojin · 2021 China Automation Congress (CAC) · 2021
Anomaly Classification is critical for the efficient and secure operation of industrial equipment. However, the accuracy of many present anomaly classification methods tend to decline, particularly when there are multiple anomalies. In this paper, an anomaly classification method on the basic of Earth Mover’s Distance (EMD) metric Loss is presented. Firstly, the anomaly classification is defined as a case of optimal image matching. Then, a novel EMD metric loss based on triplet is designed for extracting discriminative local features. Finally, a simple classifier support vector machines (SVM) can be adopted to realize the multi-anomalies classification. It is proved that the presented anomaly classification method has better performance than the traditional methods.