深度线性判别分析用于两级脑控字符拼写解码

doi:10.19665/j.issn1001-2400.2020.04.015

摘要/Abstract

摘要：

为了充分利用视听觉感知通道,实现高效的脑控字符拼写,提出一种基于区域的两级拼写范式。该范式的第一级基于运动视觉诱发电位进行目标区域选择,并引入码分多址方法进行区域编码,以提高其选择速率;第二级基于混合运动视觉诱发电位和听觉P300对目标字符进行编码,充分利用视听觉混合效应,改善目标字符选择的准确率。为了对采集的脑电信号进行有效的目标字符解码,提出一种结合深度线性判别分析的脑电信号分类识别算法。实验结果表明,深度线性判别分析算法在两级脑电信号的分类识别中平均分类准确率分别为61.7%和74%,明显高于传统方法和两种卷积神经网络方法的准确率。因此,该算法可有效地改善视听混合诱发两级脑机接口的指令解码性能。

关键词: 脑机接口, 拼写范式, 运动视觉诱发电位, 深度学习

Abstract:

In order to make full use of visual and auditory perception channels and realize efficient brain-controlled character spelling, a two-level spelling paradigm based on the region is proposed. In the first level of the paradigm, the target region is selected based on the motion-onset visual evoked potential, and the code division multiple access method is introduced to improve the selection rate. In the second level, the target character is encoded based on the hybrid motion-onset visually evoked potential and auditory P300 to make full use of the visual and auditory hybrid effect to improve the accuracy of target character selection. In order to decode the collected EEG signals effectively, a classification recognition algorithm for EEG signals combined with a deep linear discriminant analysis is proposed. Experimental results show that the average classification accuracy of the deep linear discriminant analysis algorithm in the classification recognition of two-level EEG signals is 61.7% and 74%, respectively, which is obviously higher than that of the traditional method and the two convolutional neural network methods. Therefore, the algorithm can effectively improve the decoding performance of the two-level brain computer interface induced by the audio-visual hybrid.

Key words: brain computer interface, spelling paradigm, motion-onset visual evoked potential, deep learning

中图分类号:

TP391

郭柳君,张雪英,陈桂军. 深度线性判别分析用于两级脑控字符拼写解码[J]. 西安电子科技大学学报, 2020, 47(4): 109-116.

GUO Liujun,ZHANG Xueying,CHEN Guijun. Deep linear discriminant analysis for two-stage brain-controlled character spelling decoding[J]. Journal of Xidian University, 2020, 47(4): 109-116.

图/表 7

图1

图2

表1

图3

图4

图5

表2

参考文献 16

[1]	GEMBLER F, STAWICKI P, SABOOR A, et al. Dynamic Time Window Mechanism for Time Synchronous VEP-based BCIs—Performance Evaluation with a Dictionary-supported BCI Speller Employing SSVEP and c-VEP[J]. PloS One, 2019,14(6):1-18.
[2]	CHEN X, WANG Y, NAKANISHI M, et al. High-speed Spelling with a Noninvasive Brain-computer Interface[J]. Proceedings of the National Academy of Sciences of the United States of America, 2015,112(44):E6058-E6067. doi: 10.1073/pnas.1508080112 pmid: 26483479
[3]	XU M, XIAO X, WANG Y, et al. A Brain-Computer Interface Based on Miniature-event-related Potentials Induced by Very Small Lateral Visual Stimuli[J]. IEEE Transactions on Biomedical Engineering, 2018,65(5):1166-1175. doi: 10.1109/TBME.2018.2799661 pmid: 29683431
[4]	许强, 李伟, 占荣辉, 等. 一种改进的卷积神经网络SAR目标识别算法[J]. 西安电子科技大学学报, 2018,45(5):177-183.
	XU Qiang, LI Wei, ZHAN Ronghui, et al. Improved Algorithm for SAR Target Recognition Based on the Convolutional Neural Network[J]. Journal of Xidian University, 2018,45(5):177-183.
[5]	CECOTTI H, GRASER A. Convolutional Neural Networks for P300 Detection with Application to Brain-computer Interfaces[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011,33(3):433-445. pmid: 20567055
[6]	LI J, YU Z L, GU Z, et al. A Hybrid Network for ERP Detection and Analysis Based on Restricted Boltzmann Machine[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2018,26(3):563-572. doi: 10.1109/TNSRE.2018.2803066 pmid: 29522400
[7]	PODMORE J J, BRECKON T P, AZNAN N K N, et al. On the Relative Contribution of Deep Convolutional Neural Networks for SSVEP-based Bio-signal Decoding in BCI Speller Applications[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019,27(4):611-618. doi: 10.1109/TNSRE.7333
[8]	DORFER M, KELZ R, WIDMER G. Deep Linear Discriminant Analysis[CP/OL]. [2019-11-16].https://arxiv.org/pdf/1511.04707v5.pdf.
[9]	GAO S, WANG Y, GAO X, et al. Visual and Auditory Brain-computer Interfaces[J]. IEEE Transactions on Biomedical Engineering, 2014,61(5):1436-1447. doi: 10.1109/TBME.2014.2300164
[10]	CHEN J, LI Z, HONG B, et al. A Single-stimulus, Multitarget BCI Based on Retinotopic Mapping of Motion-onset VEPs[J]. IEEE Transactions on Biomedical Engineering, 2019,66(2):464-470. doi: 10.1109/TBME.2018.2849102 pmid: 29993456
[11]	LIU M, WU W, GU Z, et al. Deep Learning Based on Batch Normalization for P300 Signal Detection[J]. Neurocomputing, 2018,275(1):288-297. doi: 10.1016/j.neucom.2017.08.039
[12]	朱莉, 赵俊, 傅应锴, 等. 一种红外热图像目标区域分割的深度学习算法[J]. 西安电子科技大学学报, 2019,46(4):1-8.
	ZHU Li, ZHAO Jun, FU Yingkai, et al. Deep Learning Algorithm for the Segmentation of the Interested Region of an Infrared Thermal Image[J]. Journal of Xidian University, 2019,46(4):1-8.
[13]	MA T, LI H, YANG H, et al. The Extraction of Motion-onset VEP BCI Features Based on Deep Learning and Compressed Sensing[J]. Journal of Neuroscience Methods, 2017,100(275):80-92.
[14]	陈瑞. 基于mVEP的脑—机接口关键技术研究[D]. 成都: 电子科技大学, 2015.
[15]	官金安, 汪鹭汐, 赵瑞娟, 等. 基于3D卷积神经网络的IR-BCI脑电视频解码研究[J]. 中南民族大学学报(自然科学版), 2019,38(4):538-546.
	GUAN Jinan, WANG Luxi, ZHAO Ruijuan, et al. Research on IR-BCI EEG Video Decoding Based on 3D Convolutional Neural Network[J]. Journal of South-Central University for Nationalities(Natural Science Edition), 2019,38(4):538-546.
[16]	王斐. 视听觉脑机接口及其临床应用研究[D]. 广州: 华南理工大学, 2017.

试次类型	第一级		第二级
试次类型	训练集	测试集	训练集	测试集
目标	448	266	300×4	167
非目标	452	235	1200	668
总数	900	501	2400	835

被试	分类准确率/%				F1值/%
被试	传统方法	3D CNN	CNN	DeepLDA	传统方法	3D CNN	CNN	DeepLDA
1	60.5	61.8	55.9	63.5	28.0	33.1	29.2	36.6
2	66.5	71.7	68.1	72.0	22.7	40.4	42.2	41.2
3	64.0	61.3	61.2	64.6	23.5	25.1	32.8	33.0
4	74.5	64.2	85.9	88.0	27.3	21.9	68.1	72.5
5	73.3	82.8	75.8	77.8	20.6	52.3	45.1	48.8
6	71.9	64.1	65.9	68.0	17.0	32.7	34.5	35.7
7	72.6	78.9	83.6	84.4	25.4	51.9	62.3	64.5
平均值±标准差	69.0±5.4	69.3±8.7	70.9±11.3	74.0±9.6	23.5±3.9	36.8±12.0	44.9±15.0	47.5±15.4