Development of Noise Reduction Techniques Using Adversarial Generative Networks and Their Application to Biological Signals

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K20334
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: University of Tsukuba
• Principal Investigator: HORIE Kazumasa 筑波大学, 計算科学研究センター, 助教 (60817112)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 機械学習 / ノイズ除去 / 生体信号処理

Outline of Final Research Achievements

In this study, I developed and evaluated noise reduction methods using deep learning models such as adversarial generative networks (GANs) and autoencoders. The proposed methods have a practical advantage in that they do not require noise-removed and noise-contaminated biological signal samples during training. However, I have still not achieved improvements in recognition accuracy during the study period. Nevertheless, the insights gained regarding the model structure and noise removal performance in this research are expected to be valuable for future research and development. I also intend to continue developing the method.

Free Research Field

深層学習

Academic Significance and Societal Importance of the Research Achievements

生体信号認識は，直感的に操作可能な入力インタフェースや自動診断に必要な技術の一つである．これらのシステムにおいては認識精度を低下させるノイズは可能な限り取り除くのが望ましい．しかしながら，認識において重要な波形とノイズの主要な周波数成分が一致することがある，ノイズ単体での計測が困難であるといった理由から，従来手法の適用が困難だった．
本研究で得られた，深層学習によるノイズ除去の知見は生体信号認識システムの有用性向上につながる可能性がある．