ADCapsNet: An Efficient and Robust Capsule Network Model for Anomaly Detection

Abstract

With the rapid development of the industrial internet of things(IIoT), the anomalies will cause significant damage to the ordinary operation of the industry. Anomaly detection work has increasingly become a hot spot. Although many related kinds of research exist, some problems still need to be solved. This paper proposes an efficient and robust semi-supervised capsule network (ADCapsNet) for anomaly detection by changing the convolution structure to better extract the features of the data and adding a new SecondaryCaps layer to better extract spatial relationships. Besides, we optimize the vector selection for dynamic anomaly detection routing and propose the scoring operation, the modified probability mechanism. The modified probability mechanism can widen the score gap between positive and negative samples. This model can accurately identify and output the spatial relationships. Extensive experiments on four datasets show that the ADCapsNet has good performance in anomaly detection.