TY - GEN
T1 - Spectro-Spatio-Temporal EEG Representation Learning for Imagined Speech Recognition
AU - Ko, Wonjun
AU - Jeon, Eunjin
AU - Suk, Heung Il
N1 - Funding Information:
This work was supported by Institute for Information & Communications Technology Promotion (IITP) grant funded by the Korea government under Grant 2017-0-00451 (Development of BCI based Brain and Cognitive Computing Technology for Recognizing User’s Intentions using Deep Learning) and Grant 2019-0-00079 (Department of Artificial Intelligence, Korea University).
Publisher Copyright:
© 2022, Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - In brain–computer interfaces, imagined speech is one of the most promising paradigms due to its intuitiveness and direct communication. However, it is challenging to decode an imagined speech EEG, because of its complicated underlying cognitive processes, resulting in complex spectro-spatio-temporal patterns. In this work, we propose a novel convolutional neural network structure for representing such complex patterns and identifying an intended imagined speech. The proposed network exploits two feature extraction flows for learning richer class-discriminative information. Specifically, our proposed network is composed of a spatial filtering path and a temporal structure learning path running in parallel, then integrates their output features for decision-making. We demonstrated the validity of our proposed method on a publicly available dataset by achieving state-of-the-art performance. Furthermore, we analyzed our network to show that our method learns neurophysiologically plausible patterns.
AB - In brain–computer interfaces, imagined speech is one of the most promising paradigms due to its intuitiveness and direct communication. However, it is challenging to decode an imagined speech EEG, because of its complicated underlying cognitive processes, resulting in complex spectro-spatio-temporal patterns. In this work, we propose a novel convolutional neural network structure for representing such complex patterns and identifying an intended imagined speech. The proposed network exploits two feature extraction flows for learning richer class-discriminative information. Specifically, our proposed network is composed of a spatial filtering path and a temporal structure learning path running in parallel, then integrates their output features for decision-making. We demonstrated the validity of our proposed method on a publicly available dataset by achieving state-of-the-art performance. Furthermore, we analyzed our network to show that our method learns neurophysiologically plausible patterns.
KW - Brain–computer interface
KW - Convolutional neural network
KW - Electroencephalogram
KW - Imagined speech
UR - http://www.scopus.com/inward/record.url?scp=85130250402&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-02444-3_25
DO - 10.1007/978-3-031-02444-3_25
M3 - Conference contribution
AN - SCOPUS:85130250402
SN - 9783031024436
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 335
EP - 346
BT - Pattern Recognition - 6th Asian Conference, ACPR 2021, Revised Selected Papers
A2 - Wallraven, Christian
A2 - Liu, Qingshan
A2 - Nagahara, Hajime
PB - Springer Science and Business Media Deutschland GmbH
T2 - 6th Asian Conference on Pattern Recognition, ACPR 2021
Y2 - 9 November 2021 through 12 November 2021
ER -