Metric structure of kernel Grassmannian representation and its application to brain signal processing

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K00302
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: The University of Electro-Communications
• Principal Investigator: Washizawa Yoshikazu 電気通信大学, 大学院情報理工学研究科, 准教授 (10419880)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 生体信号処理 / グラスマン多様体 / 脳信号処理 / ブレインコンピュータインターフェース

Outline of Final Research Achievements

We have introduced a metric structure on Grassmannian defined by pattern data, and applied it to brain signal processing. Our experimental results using open benchmark dataset shows that the proposed method exhibited significantly higher classification performance than conventional methods. Especially, the combination of the proposed method and CSSSP method, which is the extension of common spatial pattern (CSP) filter method, exhibited the best performance. Furthermore, we also proposed a method to select electrodes location and design a classifier simultaneously in brain signal classification problems using criteria of classification performance and sparse regularization based on the Bayes estimation.

Free Research Field

生体信号処理