融合注意力的多维特征图卷积运动想象分类

doi:10.3778/j.issn.1673-9418.2103079

计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (9): 2050-2060.DOI: 10.3778/j.issn.1673-9418.2103079

融合注意力的多维特征图卷积运动想象分类

李珍琦¹^,², 王晶¹^,²^,⁺(), 贾子钰¹^,², 林友芳¹^,²

1.北京交通大学计算机与信息技术学院,北京 100044
2.北京交通大学交通数据分析与挖掘北京市重点实验室,北京 100044

收稿日期:2021-03-23 修回日期:2021-06-03 出版日期:2022-09-01 发布日期:2021-06-23
通讯作者: + E-mail: wj@bjtu.edu.cn
作者简介:李珍琦（1996—）,女,河北唐山人,硕士研究生,主要研究方向为深度学习、时间序列分析与挖掘、脑机接口等。
王晶（1987—）,女,江苏扬州人,博士,副教授,博士生导师,CCF会员,主要研究方向为时间序列分析与挖掘、脑机接口、异常检测等。
贾子钰（1993—）,男,河北唐山人,博士研究生,主要研究方向为深度学习、时间序列分析与挖掘、脑机接口等。
林友芳（1971—）,男,福建武平人,博士,教授,博士生导师,CCF高级会员,主要研究方向为数据挖掘、机器学习、强化学习、复杂网络、智能技术与系统等。
基金资助:
中央高校基本科研业务费专项资金(2021JBM007);国家自然科学基金(61603029)

Attention-Based Multi-dimensional Feature Graph Convolutional Network for Motor Imagery Classification

LI Zhenqi¹^,², WANG Jing¹^,²^,⁺(), JIA Ziyu¹^,², LIN Youfang¹^,²

1. School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China
2. Beijing Key Laboratory of Traffic Data Analysis and Mining, Beijing Jiaotong University, Beijing 100044, China

Received:2021-03-23 Revised:2021-06-03 Online:2022-09-01 Published:2021-06-23
About author:LI Zhenqi, born in 1996, M.S. candidate. Her research interests include deep learning, time series analysis and mining, brain-computer interface, etc.
WANG Jing, born in 1987, Ph.D., associate professor, Ph.D. supervisor, member of CCF. Her research interests include time series analysis and mining, brain-computer interface, anomaly detection, etc.
JIA Ziyu, born in 1993, Ph.D. candidate. His research interests include deep learning, time series analysis and mining, brain-computer interface, etc.
LIN Youfang, born in 1971, Ph.D., professor, Ph.D. supervisor, senior member of CCF. His research interests include data mining, machine learning, reinforcement learning, complex network, intelligent technology and system, etc.
Supported by:
Fundamental Research Funds for the Central Universities of China(2021JBM007);National Natural Science Foundation of China(61603029)

摘要/Abstract

摘要：

运动想象（MI）作为脑机接口（BCI）的重要应用,是运动康复训练的重要支撑。由于脑电的电极分布并非天然的欧式空间,对运动想象进行准确分类具有很大的挑战。而且现有方法仅仅考虑了脑电信号（EEG）中某一维度或者某两维度的信息,无法全面捕获脑电信号在时、频、空三个维度存在的内在特征。同时,脑电信号各维度上的动态关联强度影响了分类的鲁棒性。针对上述问题,提出了一种新颖的融合注意力的多维特征图卷积网络（AMFGCN）。首先,根据电极节点分布的非欧空间特性设计出图结构,充分表示电极间的空间相关性。其次,提出时-空、频-空的双分支框架,同时表示脑电信号在时域、频域和空间域上的信息。最后,通过融合注意力机制、图卷积和时间/频谱卷积从图表示中学习脑电信号的空间表示、时间依赖性和频率依赖性,并自适应捕获各维度上的动态关联强度。在四个公开脑机接口数据集上进行了实验,结果表明AMFGCN模型提高了分类性能,优于其他现有的运动想象分类方法。

关键词: 运动想象（MI）, 注意力机制, 图卷积网络, 多维特征, 脑电信号（EEG）

Abstract:

Motor imagery (MI), an important application of brain-computer interface (BCI), is critical for sports rehabilitation. Since the spatial position of electrodes is not Euclidean, accurate classification of MI is extremely challenging. Existing methods only consider the information of a single dimension or two dimensions in electroence-phalogram (EEG) signals, and cannot fully capture the inherent features of EEG signals in three dimensions of temporal, spectral and spatial. Meanwhile, dynamic correlations of EEG in each dimension affect the robustness of classification. To solve above problems, a novel model, named attention-based multi-dimensional feature graph convolutional network (AMFGCN), is proposed. Firstly, to model non-Euclidean spatial positions of electrodes and fully show spatial correlation between electrodes, a dedicated graph structure is designed. Secondly, a dual-branch framework of time-space and frequency-space dimensions is proposed, simultaneously decoding EEG signals from temporal, spectral and spatial domains. Thirdly, via fusing attention mechanisms into graph convolution and temporal/ spectral convolution, temporal, spectral and spatial correlations are then dynamically captured respectively. Experiments on four public BCI datasets illustrate superior performance of AMFGCN, which outperforms state-of-the-art competitive methods.

Key words: motor imagery (MI), attention mechanism, graph convolutional network, multi-dimensional feature, electroencephalogram (EEG)

中图分类号:

TP391

李珍琦, 王晶, 贾子钰, 林友芳. 融合注意力的多维特征图卷积运动想象分类[J]. 计算机科学与探索, 2022, 16(9): 2050-2060.

LI Zhenqi, WANG Jing, JIA Ziyu, LIN Youfang. Attention-Based Multi-dimensional Feature Graph Convolutional Network for Motor Imagery Classification[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(9): 2050-2060.

图/表 11

图1 AMFGCN模型框架图

Fig.1 Architecture of AMFGCN

图2 脑电信号的转换过程

Fig.2 Conversion process of EEG signals

图3 图的构建过程

Fig.3 Construction process of graph

表1 时间/频谱卷积结构

Table 1 Convolution structure of temporal/spectral

Layer	KernelSize/Stride	Kernel	Activation
时/频卷积	(1,5)/(1,1)	64	ReLU

表2 全局特征聚合的结构

Table 2 Structure of global feature aggregation

Layer	KernelSize/Stride	Kernel	Activation
全局空间聚合	$(N 2, 1)$ /1	64	ReLU
全局时/频聚合	$(1, T 2)$ /1	64	ReLU

表2 全局特征聚合的结构

Table 2 Structure of global feature aggregation

Layer	KernelSize/Stride	Kernel	Activation
全局空间聚合	$(N 2, 1)$ /1	64	ReLU
全局时/频聚合	$(1, T 2)$ /1	64	ReLU

表3 实验环境

Table 3 Experimental environment

名称	版本说明
操作系统	Ubuntu 16.04.2 LTS
内存	128 GB
CPU型号	Intel^®Xeon^® CPU E5-2683 v3 @2.00 GHz
GPU型号	Tesla K80
CUDA版本	10.1
cuDNN版本	7.5.1
Python版本	3.6.8
TensorFlow版本	1.13.1
Keras版本	2.1.6

表4 不同方法在数据集BCICIV-2a中的分类性能比较

Table 4 Performance comparison of different methods on BCICIV-2a dataset

Subject	Metric	FBCSP	Shallow ConvNet	EEGNet	Multi-branch-3D	MSFBCNN	CNN-LSTM	AMFGCN
A01	Acc	0.734 4	0.833 3	0.833 3	0.812 2	0.816 0	0.874 1	0.889 8
A01	Kappa	0.645 7	0.777 4	0.777 3	0.749 6	0.754 1	0.831 9	0.852 9
A02	Acc	0.560 8	0.677 5	0.638 0	0.679 5	0.641 5	0.773 9	0.823 3
A02	Kappa	0.414 1	0.569 2	0.516 8	0.572 7	0.522 0	0.697 7	0.764 4
A03	Acc	0.804 2	0.874 1	0.887 6	0.840 9	0.869 8	0.907 3	0.909 3
A03	Kappa	0.738 8	0.832 0	0.849 8	0.787 8	0.826 0	0.843 0	0.879 0
A04	Acc	0.576 8	0.740 0	0.624 1	0.656 9	0.681 4	0.827 7	0.835 9
A04	Kappa	0.435 9	0.653 3	0.499 2	0.542 5	0.574 2	0.770 0	0.781 0
A05	Acc	0.573 8	0.634 1	0.587 2	0.705 8	0.712 7	0.728 9	0.733 5
A05	Kappa	0.431 3	0.511 9	0.449 7	0.607 8	0.615 8	0.630 1	0.644 5
A06	Acc	0.494 8	0.768 7	0.585 1	0.675 7	0.633 7	0.825 1	0.835 5
A06	Kappa	0.326 3	0.691 7	0.446 2	0.567 6	0.511 1	0.766 5	0.780 5
A07	Acc	0.812 5	0.864 6	0.848 1	0.858 7	0.905 4	0.895 8	0.917 1
A07	Kappa	0.749 7	0.819 3	0.797 2	0.811 6	0.873 6	0.794 0	0.889 4
A08	Acc	0.735 2	0.859 8	0.821 2	0.849 4	0.778 7	0.851 7	0.878 0
A08	Kappa	0.646 3	0.812 8	0.760 8	0.799 1	0.704 7	0.768 4	0.837 3
A09	Acc	0.663 6	0.823 8	0.783 0	0.781 0	0.703 1	0.889 1	0.894 5
A09	Kappa	0.551 4	0.764 7	0.710 6	0.708 0	0.603 5	0.817 6	0.859 3
Mean	Acc	0.661 8	0.786 2	0.734 2	0.762 2	0.749 1	0.841 5	0.857 4
Mean	Kappa	0.548 8	0.714 7	0.645 3	0.683 0	0.665 0	0.768 8	0.809 8

表5 不同方法在数据集BCICIII-3a中的分类性能比较

Table 5 Performance comparison of different methods on BCICIII-3a dataset

Subject	Metric	FBCSP	Shallow ConvNet	EEGNet	Multi-branch-3D	MSFBCNN	CNN-LSTM	AMFGCN
K3b	Acc	0.913 9	0.967 4	0.962 5	0.948 1	0.963 9	0.911 8	0.984 0
K3b	Kappa	0.884 8	0.956 4	0.949 9	0.930 9	0.951 7	0.882 1	0.978 7
K6b	Acc	0.679 2	0.769 8	0.705 2	0.752 8	0.785 4	0.883 3	0.922 9
K6b	Kappa	0.572 1	0.691 1	0.607 3	0.670 4	0.713 5	0.840 6	0.896 7
L1b	Acc	0.803 1	0.829 2	0.802 1	0.807 8	0.822 9	0.916 7	0.957 3
L1b	Kappa	0.736 9	0.771 0	0.735 4	0.743 7	0.763 1	0.886 8	0.943 0
Mean	Acc	0.798 7	0.855 4	0.823 3	0.836 3	0.857 4	0.903 9	0.954 7
Mean	Kappa	0.731 3	0.806 2	0.764 2	0.781 7	0.809 4	0.869 9	0.939 4

表6 不同方法在数据集HaLT中的分类性能比较

Table 6 Performance comparison of different methods on HaLT dataset

Subject	Metric	FBCSP	Shallow ConvNet	EEGNet	Multi-branch-3D	MSFBCNN	CNN-LSTM	AMFGCN
1	Acc	0.476 5	0.847 6	0.874 0	0.867 2	0.899 9	0.707 3	0.911 4
1	Kappa	0.372 0	0.816 9	0.848 6	0.840 6	0.879 8	0.648 7	0.893 5
2	Acc	0.286 3	0.675 2	0.672 2	0.733 7	0.783 7	0.697 0	0.831 3
2	Kappa	0.143 8	0.610 0	0.606 5	0.680 4	0.736 3	0.596 4	0.797 4
3	Acc	0.368 2	0.793 3	0.823 6	0.798 8	0.820 9	0.689 8	0.835 3
3	Kappa	0.241 7	0.751 6	0.788 1	0.758 5	0.780 7	0.607 4	0.802 3
4	Acc	0.359 9	0.748 3	0.769 4	0.769 9	0.814 9	0.771 6	0.827 1
4	Kappa	0.231 5	0.697 8	0.723 1	0.723 7	0.777 7	0.705 3	0.792 4
5	Acc	0.299 8	0.703 7	0.703 2	0.741 2	0.790 4	0.758 6	0.796 1
5	Kappa	0.159 0	0.644 4	0.643 7	0.689 3	0.748 2	0.710 3	0.755 2
6	Acc	0.383 0	0.845 8	0.893 3	0.854 9	0.891 4	0.878 7	0.878 1
6	Kappa	0.259 5	0.814 9	0.871 9	0.825 8	0.869 5	0.834 1	0.853 6
7	Acc	0.221 0	0.384 4	0.434 6	0.399 9	0.468 1	0.508 0	0.515 9
7	Kappa	0.064 6	0.261 0	0.321 3	0.279 7	0.361 4	0.412 1	0.419 0
8	Acc	0.213 5	0.391 6	0.442 5	0.388 6	0.454 1	0.491 8	0.493 7
8	Kappa	0.056 9	0.269 5	0.331 5	0.266 0	0.340 9	0.390 2	0.392 0
9	Acc	0.727 2	0.989 4	0.988 4	0.988 4	0.988 4	0.692 0	0.990 0
9	Kappa	0.671 7	0.987 3	0.986 0	0.986 0	0.986 0	0.583 7	0.987 9
10	Acc	0.418 0	0.796 8	0.810 3	0.817 4	0.853 4	0.586 2	0.853 7
10	Kappa	0.301 2	0.755 9	0.772 0	0.780 8	0.824 0	0.482 2	0.824 3
11	Acc	0.596 1	0.955 9	0.953 5	0.928 3	0.948 5	0.729 4	0.956 1
11	Kappa	0.515 0	0.947 0	0.944 1	0.913 9	0.938 2	0.635 1	0.947 3
12	Acc	0.444 3	0.829 6	0.849 3	0.859 0	0.894 5	0.852 1	0.904 8
12	Kappa	0.332 9	0.795 4	0.819 0	0.830 7	0.873 3	0.782 1	0.885 6
Mean	Acc	0.399 5	0.746 8	0.767 9	0.762 3	0.800 7	0.696 9	0.816 1
Mean	Kappa	0.279 2	0.696 0	0.721 3	0.714 6	0.759 7	0.615 6	0.779 2

表7 不同方法在数据集AHU-MIEEG中的分类性能比较

Table 7 Performance comparison of different methods on AHU-MIEEG dataset

Subject	Metric	FBCSP	Shallow ConvNet	EEGNet	Multi-branch-3D	MSFBCNN	CNN-LSTM	AMFGCN
S1	Acc	0.795 3	0.818 5	0.539 6	0.506 9	0.790 4	0.819 3	0.820 0
S1	Kappa	0.692 7	0.727 7	0.249 4	0.200 7	0.685 1	0.729 5	0.729 7
S2	Acc	0.438 4	0.393 0	0.537 2	0.520 4	0.412 1	0.580 7	0.593 2
S2	Kappa	0.158 0	0.092 1	0.247 2	0.221 0	0.118 7	0.373 0	0.389 9
S3	Acc	0.861 7	0.802 6	0.761 0	0.777 1	0.909 9	0.886 8	0.920 8
S3	Kappa	0.792 6	0.704 2	0.581 7	0.605 9	0.864 9	0.829 8	0.880 9
S4	Acc	0.689 5	0.723 2	0.541 4	0.537 5	0.553 4	0.814 3	0.808 1
S4	Kappa	0.534 0	0.584 9	0.252 7	0.246 6	0.336 1	0.721 2	0.711 9
S5	Acc	0.360 2	0.526 3	0.656 2	0.613 6	0.687 2	0.708 7	0.734 5
S5	Kappa	0.038 9	0.288 9	0.424 7	0.360 4	0.530 8	0.565 5	0.602 0
S6	Acc	0.533 9	0.469 6	0.565 6	0.589 0	0.705 3	0.680 1	0.708 3
S6	Kappa	0.300 9	0.204 4	0.290 5	0.322 7	0.557 6	0.522 5	0.562 7
S7	Acc	0.535 9	0.435 4	0.509 8	0.562 9	0.448 3	0.610 3	0.616 2
S7	Kappa	0.302 7	0.156 1	0.210 6	0.283 8	0.173 3	0.429 6	0.429 6
S8	Acc	0.401 8	0.515 0	0.505 8	0.499 7	0.540 5	0.546 9	0.564 5
S8	Kappa	0.101 4	0.271 3	0.199 5	0.189 6	0.310 6	0.320 8	0.347 2
S9	Acc	0.466 2	0.626 8	0.516 9	0.552 7	0.760 2	0.743 1	0.776 4
S9	Kappa	0.195 6	0.438 5	0.213 6	0.269 6	0.639 7	0.617 6	0.664 6
S10	Acc	0.348 1	0.326 4	0.444 0	0.449 5	0.416 3	0.438 0	0.451 9
S10	Kappa	0.025 1	0.029 0	0.171 1	0.131 0	0.126 9	0.167 7	0.177 9
Mean	Acc	0.543 1	0.563 7	0.557 7	0.560 9	0.622 4	0.682 8	0.699 4
Mean	Kappa	0.314 2	0.349 7	0.286 1	0.283 1	0.434 4	0.527 7	0.549 2

图4 模型不同变体的实验结果

Fig.4 Experimental results of different model variants

参考文献 24

[1]	BIRBAUMER N, GHANAYIM N, HINTERBERGER T, et al. A spelling device for the paralysed[J]. Nature, 1999, 398(6725): 297-298. DOI URL
[2]	SANTHANAM G, RYU S I, YU B M, et al. A high-performance brain-computer interface[J]. Nature, 2006, 442(7099): 195-198. DOI URL
[3]	PARSONS L M, PARSONS L M, FOX P T, et al. Use of implicit motor imagery for visual shape discrimination as revealed by PET[J]. Nature, 1995, 375(6526): 54-58. DOI URL
[4]	JIA Z Y, LIN Y F, WANG J, et al. MMCNN: a multi-branch multi-scale convolutional neural network for motor imagery classification[C]// LNCS 12459: Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases, Ghent, Sep 14-18, 2020. Cham: Springer, 2021: 736-751.
[5]	陈萌, 李幼军, 刘岩. 脑电信号与个人情绪状态关联性分析研究[J]. 计算机科学与探索, 2017, 11(5): 794-801. DOI
	CHEN M, LI Y J, LIU Y. Analysis of correlation between EEG signal and personal emotional state[J]. Journal of Frontiers of Computer Science and Technology, 2017, 11(5): 794-801. DOI
[6]	JIA Z, LIN Y, WANG J, et al. GraphSleepNet: adaptive spatial-temporal graph convolutional networks for sleep stage classification[C]// Proceedings of the 29th International Joint Conference on Artificial Intelligence. San Francisco: Margan Kaufmann, 2020: 1324-1330.
[7]	LI Y, ZHANG X R, ZHANG B, et al. A channel-projection mixed-scale convolutional neural network for motor imagery EEG decoding[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019, 27(6): 1170-1180. DOI URL
[8]	PFURTSCHELLER G, NEUPER C. Motor imagery and direct brain-computer communication[J]. Proceedings of the IEEE, 2001, 89(7): 1123-1134. DOI URL
[9]	ANG K K, CHIN Z Y, WANG C, et al. Filter bank common spatial pattern algorithm on BCI competition IV datasets 2a and 2b[J]. Frontiers in Neuroscience, 2012, 6: 39.
[10]	GANDHI T, PANIGRAHI B K, ANAND S. A comparative study of wavelet families for EEG signal classification[J]. Neurocomputing, 2011, 74(17): 3051-3057. DOI URL
[11]	ZABIDI A, MANSOR W, LEE Y K, et al. Short-time Fourier transform analysis of EEG signal generated during imagined writing[C]// Proceedings of the 2012 International Conference on System Engineering and Technology, Bandung, Sep 11-12, 2012. Piscataway: IEEE, 2012: 1-4.
[12]	SCHIRRMEISTER R T, SPRINGENBERG J T, FIEDERER L D J, et al. Deep learning with convolutional neural networks for EEG decoding and visualization[J]. Human Brain Mapping, 2017, 38(11): 5391-5420. DOI URL
[13]	ZHAO X, ZHANG H, ZHU G, et al. A multi-branch 3D convolutional neural network for EEG-based motor imagery classification[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019, 27(10): 2164-2177. DOI URL
[14]	WU H, LI F, LI Y, et al. A parallel multiscale filter bank convolutional neural networks for motor imagery EEG classi-fication[J]. Frontiers in Neuroscience, 2019, 13: 1275. DOI URL
[15]	LI X, QIAN B, WEI J, et al. Classify EEG and reveal latent graph structure with spatio-temporal graph convolutional neural network[C]// Proceedings of the 2019 IEEE International Conference on Data Mining, Beijing, Nov 8-11, 2019. Piscataway: IEEE, 2019: 389-398.
[16]	LUN X, JIA S, HOU Y, et al. GCNs-Net: a graph convolutional neural network approach for decoding time-resolved EEG motor imagery signals[J]. arXiv:2006.08924, 2020.
[17]	SUN B, ZHANG H, WU Z, et al. Adaptive spatiotemporal graph convolutional networks for motor imagery classification[J]. IEEE Signal Processing Letters, 2021, 28: 219-223. DOI URL
[18]	GILMER J, SCHOENHOLZ S S, RILEY P F, et al. Neural message passing for quantum chemistry[J]. arXiv:1704.01212, 2017.
[19]	DEFFERRARD M, BRESSON X, VANDERGHEYNST P. Convolutional neural networks on graphs with fast localized spectral filtering[C]// Proceedings of the Annual Conference on Neural Information Processing Systems 2016, Barcelona, Dec 5-10, 2016. Red Hook: Curran Associates, 2016: 3844-3852.
[20]	POLAT K, GÜNEŞ S. Classification of epileptiform EEG using a hybrid system based on decision tree classifier and fast Fourier transform[J]. Applied Mathematics and Compu-tation, 2007, 187(2): 1017-1026.
[21]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// Proceedings of the Annual Conference on Neural Information Processing Systems 2017, Long Beach, Dec 4-9, 2017. Red Hook: Curran Associates, 2017: 5998-6008.
[22]	WANG F, ZHONG S H, PENG J F, et al. Data augmentation for EEG-based emotion recognition with deep convolutional neural networks[C]// LNCS 1070: Proceedings of the 24th International Conference on Multimedia Modeling, Bangkok, Feb 2-5, 2018. Cham: Springer, 2018: 82-93.
[23]	LAWHERN V J, SOLON A J, WAYTOWICH N R, et al. EEGNet: a compact convolutional network for EEG-based brain-computer interfaces[J]. Journal of Neural Engineering, 2018, 15(5): 056013. DOI URL
[24]	ZHANG R, ZONG Q, DOU L, et al. A novel hybrid deep learning scheme for four-class motor imagery classification[J]. Journal of Neural Engineering, 2019, 16(6): 066004. DOI URL

融合注意力的多维特征图卷积运动想象分类

Attention-Based Multi-dimensional Feature Graph Convolutional Network for Motor Imagery Classification

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 24

相关文章 15

编辑推荐

Metrics

[1]	吕晓琦, 纪科, 陈贞翔, 孙润元, 马坤, 邬俊, 李浥东. 结合注意力与循环神经网络的专家推荐算法[J]. 计算机科学与探索, 2022, 16(9): 2068-2077.
[2]	杨知桥, 张莹, 王新杰, 张东波, 王玉. 改进U型网络在视网膜病变检测中的应用研究[J]. 计算机科学与探索, 2022, 16(8): 1877-1884.
[3]	夏鸿斌, 肖奕飞, 刘渊. 融合自注意力机制的长文本生成对抗网络模型[J]. 计算机科学与探索, 2022, 16(7): 1603-1610.
[4]	李运寰, 闻继伟, 彭力. 高帧率的轻量级孪生网络目标跟踪[J]. 计算机科学与探索, 2022, 16(6): 1405-1416.
[5]	张雁操, 赵宇海, 史岚. 融合图注意力的多特征链接预测算法[J]. 计算机科学与探索, 2022, 16(5): 1096-1106.
[6]	赵鹏飞, 谢林柏, 彭力. 融合注意力机制的深层次小目标检测算法[J]. 计算机科学与探索, 2022, 16(4): 927-937.
[7]	陆仲达, 张春达, 张佳奇, 王子菲, 许军华. 双分支网络的苹果叶部病害识别[J]. 计算机科学与探索, 2022, 16(4): 917-926.
[8]	黄思远, 赵宇海, 梁燚铭. 融合图嵌入和注意力机制的代码搜索[J]. 计算机科学与探索, 2022, 16(4): 844-854.
[9]	王燕妮, 余丽仙. 注意力与多尺度有效融合的SSD目标检测算法[J]. 计算机科学与探索, 2022, 16(2): 438-447.
[10]	李想, 杨兴耀, 于炯, 钱育蓉, 郑捷. 基于知识图谱卷积网络的双端推荐算法[J]. 计算机科学与探索, 2022, 16(1): 176-184.
[11]	蒋光峰, 胡鹏程, 叶桦, 仰燕兰. 基于重构误差的同构图分类模型[J]. 计算机科学与探索, 2022, 16(1): 185-193.
[12]	蔡明昕, 孙晶, 王斌. 多角度语义轨迹相似度计算模型[J]. 计算机科学与探索, 2021, 15(9): 1632-1640.
[13]	陈俊芬, 张明, 赵佳成, 谢博鋆, 李艳. 结合降噪和自注意力的深度聚类算法[J]. 计算机科学与探索, 2021, 15(9): 1717-1727.
[14]	张晗, 贾甜远, 骆方, 张生, 邬霞. 面向网络文本的BERT心理特质预测研究[J]. 计算机科学与探索, 2021, 15(8): 1459-1468.
[15]	陈剑南, 杜军平, 薛哲, 寇菲菲. 基于多重注意力的金融事件大数据精准画像[J]. 计算机科学与探索, 2021, 15(7): 1237-1244.