文献类型：期刊文献

英文题名：CSCLN-DDTE: Cross subject contrastive learning network with domain diversity and templates enhancement for SSVEP-BCI frequency recognition

作者：He, Xinjie[1];Xu, Ren[2];Lau, Andrew Ty[3];Wu, Xiao[1];Chen, Yixin[1];Chen, Weijie[1];Wang, Xingyu[1];Cichocki, Andrzej[4,5,6];Jin, Jing[1,7]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]GTEC Med Engn GmbH, A-4521 Schiedlberg, Austria;[3]Shanghai Lansheng Brain Hosp Investment Co Ltd, Shanghai 200336, Peoples R China;[4]Nicolaus Copernicus Univ, Dept Informat, PL-87100 Torun, Poland;[5]Polish Acad Sci, Syst Res Inst, PL-01447 Warsaw, Poland;[6]RIKEN, Brain Sci Inst, Lab Adv Brain Signal Proc, Wako, Japan;[7]East China Univ Sci & Technol, Sch Math, Shanghai 200237, Peoples R China

年份：2025

卷号：332

外文期刊名：KNOWLEDGE-BASED SYSTEMS

收录：;EI(收录号：20254819601032);WOS:【SCI-EXPANDED(收录号:WOS:001631172200003)】；

基金：This work was supported by the Grant National Natural Science Foundation of China under Grant 62176090 and STI 2030-major projects 2022ZD0208900; in part by Shanghai Municipal Science and Technology Major Project under Grant 2021SHZDZX, This research is also supported by Project of Jiangsu Province Science and Technology Plan Special Fund in 2022 (Key research and development plan industry foresight, fundamental research fund for the central universities JKH01231636 and key core technologies) under Grant BE2022064-1.

语种：英文

外文关键词：Brain computer interface (BCI); Steady-state visual evoked potential (SSVEP); Contrastive learning; Domain diversity; Templates enhancement

摘要：Deep learning and spatial filter methods based on individual calibration data have been proven to enhance the decoding performance of SSVEP-BCIs systems. However, such methods require online calibration and have limited system availability. In order to facilitate a plug-and-play system, a cross subject contrastive learning network with domain diversity and templates enhancement (CSCLN-DDTE) was proposed for SSVEP frequency recognition. Firstly, inspired by the student-teacher (S-T) network framework, a cross-subject comparative learning framework (CSCLF) was designed that can utilize the teacher's prior knowledge to guide students from different domains to learn specific discriminative features of SSVEP task-related components. Next, a periodic repeating task-related component template was constructed based on SSVEP time-domain periodic characteristics to enhance teachers' guidance in the TFEN module. Finally, a reconstruction of channel correlation data enhancement method is introduced to increase the domain diversity of training samples in the SFEN module. Offline tests on two public datasets show that the proposed CSCLN-DDTE outperforms state-of-the-art (SOTA) methods such as FBCCA, ttCCA, DNN, EEGNet and TST-CSFR. Proposed method can achieve an average information transfer rates (ITRs) of 166.27 +/- 31.09 bit/min and 128.39 +/- 30.35 bit/min on Benchmark dataset and Beta dataset respectively. Our approach achieves the best recognition performance without calibration trials and promotes the practical application of SSVEP-BCI systems.