Multi-stage distribution adaptation model for cross-subject motor imagery EEG decoding

doi:10.11772/j.issn.1001-9081.2024111698

Journal of Computer Applications ›› 2025, Vol. 45 ›› Issue (12): 4064-4072.DOI: 10.11772/j.issn.1001-9081.2024111698

• Frontier and comprehensive applications • Previous Articles

Multi-stage distribution adaptation model for cross-subject motor imagery EEG decoding

Min HE¹^,², Tianjian LUO¹^,²

^1.College of Computer and Cyber Security，Fujian Normal University，Fuzhou Fujian 350117，China
^2.Digital Fujian Environmental Monitoring IoT Laboratory （Fujian Normal University），Fuzhou Fujian 350117，China

Received:2024-12-04 Revised:2025-04-15 Accepted:2025-04-16 Online:2025-04-22 Published:2025-12-10
Contact: Tianjian LUO
About author:HE Min， born in 1999， M. S. candidate. His research interests include brain-computer interface， pattern recognition， EEG signal analysis.
LUO Tianjian， born in 1990， Ph. D.， associate professor. His research interests include brain-computer interface， pattern recognition， EEG signal analysis.
Supported by:
National Natural Science Foundation of China(62106049);Natural Science Foundation of Fujian Province(2022J01655)

多阶段分布适应的跨被试运动想象脑电解码模型

何敏¹^,², 罗天健¹^,²

^1.福建师范大学计算机与网络空间安全学院，福州 350117
^2.数字福建环境监测物联网实验室（福建师范大学），福州 350117

通讯作者: 罗天健
作者简介:何敏（1999—），男，福建南平人，硕士研究生，主要研究方向：脑机接口、模式识别、脑电（EEG）信号分析
罗天健（1990—），男，湖北黄冈人，副教授，博士，主要研究方向：脑机接口、模式识别、EEG信号分析。
基金资助:
国家自然科学基金资助项目(62106049);福建省自然科学基金资助项目(2022J01655)

Abstract

Abstract:

Motor Imagery ElectroEncephaloGraph （MI-EEG） signals play a significant role in non-invasive Brain-Computer Interface （BCI） and have been utilized in clinical rehabilitation training widely. As one of the subjective paradigms， MI-EEG has high sample collection costs and large individual differences with complex time variability and low signal-to-noise ratio， so that constructing cross-subject MI-EEG decoding models have become a critical research focus. However， most of the recent cross-subject decoding models adopt the single-stage adversarial learning strategy， and only consider to learn deep representations with marginal and conditional distribution minimization， which constrain the MI-EEG decoding performance seriously. Therefore， a Multi-Stage Distribution Adaptation （MSDA） model was proposed for cross-subject MI-EEG decoding. Firstly， sample covariance was employed to align marginal distribution differences between subjects. Secondly， marginal distribution-invariant deep representations were obtained through pre-trained feature extractor and domain discriminator. Finally， a joint distribution-invariant mapping of deep representations was constructed using L²-distance， and such mapping and classifiers were trained alternately to learn joint distribution-invariant deep representations and used for cross-subject MI-EEG decoding. In MSDA model， distribution adaptation between subjects were conducted in three stages， including sample’s marginal distribution， deep representations’ marginal distribution and deep representations’ joint distribution， thereby addressing the challenge of single-stage distribution adaptation effectively. Experimental results on the BCI competition IV-2a and BCI Competition IV-2b public datasets demonstrate that MSDA model surpasses the latest decoding models in both accuracy and Kappa coefficient. The above indicates that MSDA model enhances the learning ability of cross-subject domain-invariant deep representations， which offers a new option for building MI-BCI.

Key words: Motor Imagery (MI), ElectroEncephaloGraph (EEG) signal, cross-subject, domain-invariant deep representation, domain adaptation

摘要：

运动想象脑电（MI-EEG）信号在无创脑机接口（BCI）中扮演重要角色，已被广泛应用于临床辅助康复训练。作为一个主动刺激范式，MI-EEG的样本采集成本高且个体差异大，同时具有复杂时变性和低信噪比等特点，使得构建跨被试MI-EEG解码模型成了研究关键。然而，当前绝大多数的跨被试解码模型采用单阶段的对抗学习策略，并且仅考虑学习边缘分布或条件分布最小化的深度表征，进而严重制约了MI-EEG解码性能。因此，提出一种多阶段分布适应（MSDA）的跨被试MI-EEG解码模型。首先，采用样本协方差对齐被试者之间的样本边缘分布差异；其次，通过预训练特征提取器和域鉴别器获取边缘分布不变的深度表征；最后，通过L²距离构建深度表征的联合分布不变映射，并交替训练联合这些映射和分类器学习联合分布不变的深度表征用于跨被试MI-EEG解码。MSDA模型分别从样本边缘分布、深度表征边缘分布和深度表征联合分布这3个阶段进行被试者之间的分布适应，从而有效地应对单阶段分布适应的挑战。在BCI Competition IV-2a和BCI Competition IV-2b公开数据集上的实验结果表明，MSDA模型在准确率和Kappa系数这2个指标均超越了近期提出的解码模型。可见， MSDA模型提升了跨被试域不变深度特征的学习能力，为构建MI-BCI提供了新的选择。

关键词: 运动想象, 脑电信号, 跨被试, 域不变深度表征, 域适应

CLC Number:

TP242

Min HE, Tianjian LUO. Multi-stage distribution adaptation model for cross-subject motor imagery EEG decoding[J]. Journal of Computer Applications, 2025, 45(12): 4064-4072.

何敏, 罗天健. 多阶段分布适应的跨被试运动想象脑电解码模型[J]. 《计算机应用》唯一官方网站, 2025, 45(12): 4064-4072.

Figures/Tables 14

References 25

[1]	肖晓琳，辛风然，梅杰，等. 自适应脑机接口研究综述［J］. 电子与信息学报， 2023， 45（7）： 2386-2394.
	XIAO X L， XIN F R， MEI J， et al. A review of adaptive brain-computer interface research［J］. Journal of Electronics and Information Technology， 2023， 45（7）： 2386-2394.
[2]	陆蓉蓉，高天昊，胡义茜，等. 基于运动想象的脑机接口系统联合不同末端效应器对慢性期脑卒中患者上肢运动功能改善的初步研究［J］. 中国康复医学杂志， 2024， 39（8）： 1104-1110.
	LU R R， GAO T H， HU Y Q， et al. Improving upper limb motor function in chronic stroke patients using a brain-computer interface system based on motor imagery combined with different end effectors： a preliminary study［J］. Chinese Journal of Rehabilitation Medicine， 2024， 39（8）： 1104-1110.
[3]	杨帮华. 运动想象脑机接口在神经系统疾病康复中的应用［J］. 康复学报， 2023， 33（6）： 477-485.
	YANG B H. Application of motor imagination brain-computer interface in rehabilitation of neurological diseases［J］. Rehabilitation Medicine， 2023， 33（6）： 477-485.
[4]	申佳华，代成龙，李光辉. AFBCNet：一种基于自适应滤波器组的运动想象卷积神经网络［J］. 传感技术学报， 2023， 36（11）： 1714-1723.
	SHEN J H， DAI C L， LI G H. AFBCNet： a convolutional neural network for motor imagery EEG based on adaptive filter bank［J］. Chinese Journal of Sensors and Actuators， 2023， 36（11）： 1714-1723.
[5]	胡莹，刘燕，程晨晨，等. 基于自适应时频共空间模式结合卷积神经网络的多任务运动想象脑电分类［J］. 生物医学工程学杂志， 2022， 39（6）： 1065-1073， 1081.
	HU Y， LIU Y， CHENG C C， et al. Multi-task motor imagery electroencephalogram classification based on adaptive time-frequency common spatial pattern combined with convolutional neural network［J］. Journal of Biomedical Engineering， 2022， 39（6）： 1065-1073， 1081.
[6]	徐慧，何宏，张慧敏，等. 黎曼流形切平面空间中运动想象脑电信号的迁移学习［J］. 中国生物医学工程学报， 2023， 42（6）： 659-667.
	XU H， HE H， ZHANG H M， et al. Transfer learning of motor imagery EEG signals in Riemannian manifold tangent space［J］. Chinese Journal of Biomedical Engineering， 2023， 42（6）： 659-667.
[7]	蒋云良，周阳，张雄涛，等. 基于域间Mixup微调策略的跨被试运动想象脑电信号分类算法［J］. 智能系统学报， 2024， 19（4）： 909-919.
	JIANG Y L， ZHOU Y， ZHANG X T， et al. Cross-subject motor imagery EEG classification based on inter-domain Mixup fine-tuning strategy［J］. CAAI Transactions of Intelligent Systems， 2024， 19（4）： 909-919.
[8]	高云园，薛云峰，张聪睿，等. 基于集成式张量域自适应的运动想象脑电分类［J］. 中国生物医学工程学报， 2024， 43（4）： 399-407.
	GAO Y Y， XUE Y F， ZHANG C R， et al. MI-EEG classification based on ensemble tensor domain adaptation［J］. Chinese Journal of Biomedical Engineering， 2024， 43（4）： 399-407.
[9]	ZHANG L， GAO X. Transfer adaptation learning： a decade survey［J］. IEEE Transactions on Neural Networks and Learning Systems， 2022， 35（1）： 23-44.
[10]	ZHAO H， ZHENG Q， MA K， et al. Deep representation-based domain adaptation for nonstationary EEG classification［J］. IEEE Transactions on Neural Networks and Learning Systems， 2020， 32（2）： 535-545.
[11]	HONG X， ZHENG Q， LIU L， et al. Dynamic joint domain adaptation network for motor imagery classification［J］. IEEE Transactions on Neural Systems and Rehabilitation Engineering， 2021， 29： 556-565.
[12]	曹铉，罗天健. 运动想象脑电信号的跨被试动态多域对抗学习方法［J］. 计算机应用， 2024， 44（2）： 645-653.
	CAO X， LUO T J. Dynamic multi-domain adversarial learning method for cross-subject motor imagery EEG signals［J］. Journal of Computer Applications， 2024， 44（2）： 645-653.
[13]	SHE Q， CHEN T， FANG F， et al. Improved domain adaptation network based on Wasserstein distance for motor imagery EEG classification［J］. IEEE Transactions on Neural Systems and Rehabilitation Engineering， 2023， 31： 1137-1148.
[14]	PHUNRUANGSAKAO C， ACHANCCARAY D， HAYASHIBE M. Deep adversarial domain adaptation with few-shot learning for motor-imagery brain-computer interface ［J］. IEEE Access， 2022， 10： 57255-57265.
[15]	LI H， ZHANG D， XIE J. MI-DABAN： a dual-attention-based adversarial network for motor imagery classification［J］. Computers in Biology and Medicine， 2023， 152： No.106420.
[16]	ZHANG D， LI H， XIE J. MI-CAT： a Transformer-based domain adaptation network for motor imagery classification［J］. Neural Networks， 2023， 165： 451-462.
[17]	ZHANG D， LI H， XIE J， et al. MI-DAGSC： a domain adaptation approach incorporating comprehensive information from MI-EEG signals［J］. Neural Networks， 2023， 167： 183-198.
[18]	AL-SAEGH A， DAWWD S A， ABDUL-JABBAR J M. CutCat： an augmentation method for EEG classification［J］. Neural Networks， 2021， 141： 433-443.
[19]	ALTAHERI H， MUHAMMAD G， ALSULAIMAN M. Physics-informed attention temporal convolutional network for EEG-based motor imagery classification ［J］. IEEE Transactions on Industrial Informatics， 2022， 19（2）： 2249-2258.
[20]	TANG X， LI X， LI W， et al. Transfer learning： rotation alignment with Riemannian mean for brain-computer interfaces and wheelchair control［J］. IEEE Transactions on Cognitive and Developmental Systems， 2021， 15（2）： 487-498.
[21]	CHEN S， HARANDI M， JIN X， et al. Domain adaptation by joint distribution invariant projections［J］. IEEE Transactions on Image Processing， 2020， 29： 8264-8277.
[22]	QUANZ B， HUAN J， MISHRA M. Knowledge transfer with low-quality data： a feature extraction issue［J］. IEEE Transactions on Knowledge and Data Engineering， 2012， 24（10）： 1789-1802.
[23]	KANAMORI T， HIDO S， SUGIYAMA M. A least-squares approach to direct importance estimation ［J］. Journal of Machine Learning Research， 2009， 10： 1391-1445.
[24]	BICKEL S， BOGOJESKA J， LENGUAUER T， et al. Multi-task learning for HIV therapy screening［C］// Proceedings of the 25th International Conference on Machine Learning. New York： ACM， 2008： 56-63.
[25]	GONG M， ZHANG K， LIU T， et al. Domain adaptation with conditional transferable components［C］// Proceedings of the 33rd International Conference on Machine Learning. New York： JMLR.org， 2016： 2839-2848.

方法	准确率/%									平均准确率±标准差/ %	Kappa
方法	A01	A02	A03	A04	A05	A06	A07	A08	A09	平均准确率±标准差/ %	Kappa
DRDA	83.19	55.14	87.43	75.28	62.29	57.15	86.18	83.61	82.00	74.70±12.96	0.663
CutCat	82.64	60.76	94.79	81.25	47.22	67.36	89.58	82.56	78.13	76.03±14.97	0.678
DAFS	81.94	64.58	88.89	73.61	70.49	56.60	85.42	79.51	81.60	75.85±10.47	0.678
CAT	90.62	54.51	91.32	72.57	63.19	62.85	87.15	85.07	84.03	76.81±13.80	0.690
DABAN	88.54	55.56	91.32	77.43	60.42	58.68	87.15	83.68	82.64	76.16±14.06	0.676
IWDAN	83.29	63.97	90.30	76.94	69.34	60.08	89.31	82.35	82.81	77.60±10.85	0.695
DAGSC	90.28	61.46	93.75	76.39	71.18	60.76	94.79	84.38	83.68	79.63±12.99	0.728
DJDAN	86.46	68.75	93.06	85.42	72.57	63.54	95.49	85.76	83.68	81.64±10.94	—
DMDAN	84.75	61.11	92.36	69.44	62.85	57.64	88.89	85.42	86.11	76.50±13.57	0.687
MSDA	86.56	65.63	93.44	75.31	82.19	69.38	91.25	88.13	84.06	81.77±9.69	0.754

方法	准确率/%									平均准确率±标准差/ %	Kappa
方法	A01	A02	A03	A04	A05	A06	A07	A08	A09	平均准确率±标准差/ %	Kappa
DRDA	83.19	55.14	87.43	75.28	62.29	57.15	86.18	83.61	82.00	74.70±12.96	0.663
CutCat	82.64	60.76	94.79	81.25	47.22	67.36	89.58	82.56	78.13	76.03±14.97	0.678
DAFS	81.94	64.58	88.89	73.61	70.49	56.60	85.42	79.51	81.60	75.85±10.47	0.678
CAT	90.62	54.51	91.32	72.57	63.19	62.85	87.15	85.07	84.03	76.81±13.80	0.690
DABAN	88.54	55.56	91.32	77.43	60.42	58.68	87.15	83.68	82.64	76.16±14.06	0.676
IWDAN	83.29	63.97	90.30	76.94	69.34	60.08	89.31	82.35	82.81	77.60±10.85	0.695
DAGSC	90.28	61.46	93.75	76.39	71.18	60.76	94.79	84.38	83.68	79.63±12.99	0.728
DJDAN	86.46	68.75	93.06	85.42	72.57	63.54	95.49	85.76	83.68	81.64±10.94	—
DMDAN	84.75	61.11	92.36	69.44	62.85	57.64	88.89	85.42	86.11	76.50±13.57	0.687
MSDA	86.56	65.63	93.44	75.31	82.19	69.38	91.25	88.13	84.06	81.77±9.69	0.754

方法	准确率/%									平均准确率±标准差/ %	Kappa
方法	B01	B02	B03	B04	B05	B06	B07	B08	B09	平均准确率±标准差/ %	Kappa
DRDA	81.37	62.86	63.63	95.94	93.56	88.19	85.00	95.25	90.00	83.98±12.67	0.680
CutCat	75.31	60.00	60.31	97.19	82.81	82.50	74.69	88.13	85.00	78.44±12.33	0.569
DAFS	70.31	73.57	80.31	94.69	95.00	83.75	93.73	95.00	75.31	84.63±10.20	0.733
CAT	86.11	65.97	61.46	98.26	93.75	89.24	86.11	95.69	90.97	85.28±12.93	0.706
DABAN	84.03	63.43	62.50	98.61	94.44	88.19	86.81	94.79	90.63	84.83±13.17	0.684
IWDAN	84.66	66.57	68.04	96.78	94.32	82.61	88.47	93.96	90.10	85.06±11.05	0.740
DAGSC	86.88	63.13	63.75	98.75	95.63	89.06	87.50	95.93	91.25	85.76±13.28	0.710
DJDAN	83.44	58.57	59.06	98.13	96.56	84.38	86.25	92.81	87.81	83.00±14.64	—
DMDAN	76.56	73.57	82.19	95.94	98.12	83.13	91.56	95.00	82.50	86.50±8.90	0.730
MSDA	80.94	70.00	85.94	97.50	98.13	82.19	94.06	95.63	88.44	88.09±9.35	0.764

方法	准确率/%									平均准确率±标准差/ %	Kappa
方法	B01	B02	B03	B04	B05	B06	B07	B08	B09	平均准确率±标准差/ %	Kappa
DRDA	81.37	62.86	63.63	95.94	93.56	88.19	85.00	95.25	90.00	83.98±12.67	0.680
CutCat	75.31	60.00	60.31	97.19	82.81	82.50	74.69	88.13	85.00	78.44±12.33	0.569
DAFS	70.31	73.57	80.31	94.69	95.00	83.75	93.73	95.00	75.31	84.63±10.20	0.733
CAT	86.11	65.97	61.46	98.26	93.75	89.24	86.11	95.69	90.97	85.28±12.93	0.706
DABAN	84.03	63.43	62.50	98.61	94.44	88.19	86.81	94.79	90.63	84.83±13.17	0.684
IWDAN	84.66	66.57	68.04	96.78	94.32	82.61	88.47	93.96	90.10	85.06±11.05	0.740
DAGSC	86.88	63.13	63.75	98.75	95.63	89.06	87.50	95.93	91.25	85.76±13.28	0.710
DJDAN	83.44	58.57	59.06	98.13	96.56	84.38	86.25	92.81	87.81	83.00±14.64	—
DMDAN	76.56	73.57	82.19	95.94	98.12	83.13	91.56	95.00	82.50	86.50±8.90	0.730
MSDA	80.94	70.00	85.94	97.50	98.13	82.19	94.06	95.63	88.44	88.09±9.35	0.764

消融条件		平均准确率±标准差/%		Kappa
TCN 模块	对抗预训练	BCIIV-2a	BCIIV-2b	BCIIV-2a	BCIIV-2b
√	√	81.77±9.69	88.09±9.35	0.754	0.764
×	√	77.50±8.29	86.76±9.22	0.702	0.733
√	×	78.54±10.03	87.43±9.66	0.714	0.747

Multi-stage distribution adaptation model for cross-subject motor imagery EEG decoding

多阶段分布适应的跨被试运动想象脑电解码模型

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 25

Related Articles 15

Recommended Articles

Metrics

消融情形	平均准确率±标准差/%		Kappa
消融情形	BCIIV-2a	BCIIV-2b	BCIIV-2a	BCIIV-2b
情形1）	79.28±11.24	87.78±9.14	0.724	0.756
情形2）	79.93±10.16	87.14±9.70	0.731	0.743
情形3）	81.77±9.69	88.09±9.35	0.754	0.764

[1]	Chuang WANG, Lu YU, Jianwei CHEN, Cheng PAN, Wenbo DU. Review of open set domain adaptation [J]. Journal of Computer Applications, 2025, 45(9): 2727-2736.
[2]	Bo FENG, Haizheng YU, Hong BIAN. Domain adaptive semantic segmentation based on masking enhanced self-training [J]. Journal of Computer Applications, 2025, 45(7): 2132-2137.
[3]	Meirong DING, Jinxin ZHUO, Yuwu LU, Qinglong LIU, Jicong LANG. Domain adaptation integrating environment label smoothing and nuclear norm discrepancy [J]. Journal of Computer Applications, 2025, 45(4): 1130-1138.
[4]	Xuan CAO, Tianjian LUO. Dynamic multi-domain adversarial learning method for cross-subject motor imagery EEG signals [J]. Journal of Computer Applications, 2024, 44(2): 645-653.
[5]	Siqi YANG, Tianjian LUO, Xuanhui YAN, Guangju YANG. Transfer kernel learning method based on spatial features for motor imagery EEG [J]. Journal of Computer Applications, 2024, 44(11): 3354-3363.
[6]	Yang WANG, Hongliang FU, Huawei TAO, Jing YANG, Yue XIE, Li ZHAO. Cross-corpus speech emotion recognition based on decision boundary optimized domain adaptation [J]. Journal of Computer Applications, 2023, 43(2): 374-379.
[7]	Daili CHEN, Guoliang XU. Cross-domain person re-identification method based on attention mechanism with learning intra-domain variance [J]. Journal of Computer Applications, 2022, 42(5): 1391-1397.
[8]	Xiaolong LIU, Shitong WANG. Open set fuzzy domain adaptation algorithm via progressive separation [J]. Journal of Computer Applications, 2021, 41(11): 3127-3131.
[9]	YUAN Yuan, WU Wen, WAN Yi. Single image shadow detection method based on entropy driven domain adaptive learning [J]. Journal of Computer Applications, 2020, 40(7): 2131-2136.
[10]	XIAO He, LIU Zhiqin, WANG Qingfeng, HUANG Jun, ZHOU Ying, LIU Qiyu, XU Weiyun. Mass and calcification classification method in mammogram based on multi-view transfer learning [J]. Journal of Computer Applications, 2020, 40(5): 1460-1464.
[11]	Fei LUO, Pengfei LIU, Yuan LUO, Simeng ZHU. Motor imagery EEG feature extraction method based on multi-feature fusion [J]. Journal of Computer Applications, 2020, 40(2): 616-620.
[12]	ZHENG Zongsheng, HU Chenyu, JIANG Xiaoyi. Deep transfer adaptation network based on improved maximum mean discrepancy algorithm [J]. Journal of Computer Applications, 2020, 40(11): 3107-3112.
[13]	TAN Yao, RAO Wenbi. Heterogeneous compound transfer learning method for video content annotation [J]. Journal of Computer Applications, 2018, 38(6): 1547-1553.
[14]	HU Min, LI Chong, LU Rongrong, HUANG Hongcheng. Performance analysis of motor imagery training based on 3D visual guidance [J]. Journal of Computer Applications, 2018, 38(3): 836-841.
[15]	TANG Song, CHEN Lijuan, CHEN Zhixian, YE Mao. Domain adaptation image classification based on target local-neighbor geometrical information [J]. Journal of Computer Applications, 2017, 37(4): 1164-1168.