Anomaly detection based on memory-augmented in computer room laboratory
Jianxia Wang, Wenying Liu, Yunfeng Xu · 2022 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI) · 2022
Computer room laboratory is a place to store servers, its security is very import, the way manual monitoring to monitor the abnormal of the computer room monitoring video is time-consuming and laborious, it is efficient to detect the abnormal conditions in the monitoring video of the computer room laboratory based on deep learning. Therefore, this paper proposes anomaly detection based on memory-agumented in computer room laboratory, and collected and processed the anomaly detection dataset of computer room laboratory. The model proposed in this paper uses the automatic encoder as the basic network and the anomaly discrimination method based on reconstruction, but the automatic encoder “generalizes”, in this paper, memory enhancement module is applied in automatic encoders, that suppresses the reconstruction of abnormal sample. In order to extract spatiotemporal features efficiently, interaction-reduced channel-separated convolution network is used as the video feature extraction network, which makes the memory module saves memory items with better performance and accurately reconstruct normal sample. The detection accuracy of the model of this paper is 90.68% in the anomaly detection dataset of computer room laboratory.