433 / 2022-08-08 18:01:31
Unsupervised machine anomalous sound detection based on domain generalization technique
Unsupervised deep learning; Anomalous sound detection; Domain generalization; Sub-cluster AdaCos loss
终稿
Linke Zhang / Wuhan University of Technology
Yanwu Xu / Wuhan University of Technology
Ming Jin / Wuhan University of Technology
Yongsheng Yu / Wuhan University of Technology
For monitoring the mechanical equipment based on acoustic signal, the sound emitted during machine operation may be different due to the changes of machine operation status or malfunctions, and other factors, such as environmental noise, may also change the acoustic characteristics captured in the scene. Traditional machine anomalous sound detection systems may incorrectly label normal sounds as abnormal due to the presence of changes in acoustic features when classifying machine conditions. We propose an unsupervised machine anomalous sound detection system based on domain generalization techniques, which uses source domain data to learn common features across domains to provide generalization capability for model domain transfer conditions, multiple feature representations and specially designed subsystem architectures are used in a single neural network, combined with a domain blending method and a coordinate attention mechanism module to further improve domain generalization capability and anomaly recognition performance. Experiments are conducted on open source datasets and analyzed the comparing AUC and pAUC scores with two baseline evaluation systems, and the experimental results show that the anomaly detection performance of the proposed system in this paper is significantly improved under the domain generalization condition.
重要日期
  • 会议日期

    11月01日

    2022

    11月03日

    2022

  • 10月30日 2022

    初稿截稿日期

  • 11月09日 2022

    注册截止日期

主办单位
Qingdao University of Technology
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询