Improving the Performance of Anomaly Detector based on Geometric Transform-based Deep Neural Networks
Hyunsoo Kim, Dong‐Joong Kang · 2021 21st International Conference on Control, Automation and Systems (ICCAS) · 2021
The geometric transformation (GT) based anomaly detection is a recent famous method to detect the abnomal sample among many normal data. However, the training process of the GT method is unstable and rather inaccurate for use in the industrial field. The goal of this paper is to suggest a method to improve the performance of GT based anomaly detector. Specifically, this paper proposes a few methods such as loss and activation function redefinition for training efficient GT model, so that it can be used in the real industrial field.