Anomaly Detection Based on Semi-Supervised Generative Adversarial Networks

Abstract

In order to solve the problem of uncivilized speech on campus, a detection method combining semi-supervised generative adversarial network and self-attention mechanism is proposed. The algorithm can use most of the unlabeled samples and a small part of the labeled samples to realize the abnormal detection of uncivilized speech, which solves the problems that the experimental samples are not easy to obtain and the cost of marking is high. In order to make full use of the latent representation learned by the encoder module in the auto-encoder, a new representation term is added to the anomaly score function part; in order to improve the problem of the traditional convolution local receptive field, a self-attention mechanism is introduced into the generative adversarial network Modules that help models learn long-range dependency representations of data. Experiments on self-made datasets show that the method proposed in this paper can improve the ability of model anomaly detection and improve the accuracy of anomaly detection.