基于Fisher判别和GKF-RELM算法的多特征步态模式识别

doi:10.12263/DZXB.20200033

摘要/Abstract

摘要：

为提高下肢表面肌电信号步态识别的识别精度和计算效率，采用一种基于高斯核函数优化正则化超限学习机（GKF-RELM）算法，对肌电信号提取时域、频域和非线性动力学三类特征并分别计算步态识别率，运用Fisher判别函数分析所提特征的可分性，得到多类特征的融合特征作为输入数据对分类器进行训练，再用训练好的分类器进行步态识别，从识别率和计算时间两方面，分别与支持向量机（SVM）和深度神经网络（DNN）方法进行了对比分析.结果表明，基于Fisher判别可分性指标确定的多类特征组合，能得到最优识别效果，并在提高分类精度的同时，优化了计算效率.此外，GKF-RELM方法的识别率也优于传统的ELM方法.

关键词: 步态识别, 多特征融合, GKF-RELM算法, 深度神经网络, Fisher判别分析

Abstract:

To improve the recognition accuracy and computational efficiency of gait recognition for lower extremity surface electromyography (sEMG), the Gaussian kernel function-regularized extreme learning machine (GKF-RELM) algorithm is presented. The features of time domain, frequency domain and non-linear dynamics via sEMG signals are extracted and the corresponding gait recognition rates are calculated, respectively. Fisher discriminant function is utilized to analyze the separability of the proposed features, and the fusion features of multi-class features are obtained as the input data to train the classifiers, and the trained classifier is used for gait recognition. The recognition rate and calculation time are compared with support vector machine (SVM) and deep neural network (DNN). The results show that the combination of multi-class features based on Fisher discriminant separability index can obtain the optimal recognition effects, and improve the classification accuracy, as well as optimize the calculation efficiency. In addition, the recognition rate of GKF-RELM method is preferable to that of traditional ELM method.

Key words: gait recognition, multi-feature fusion, GKF-RELM algorithm, deep neural network (DNN), Fisher discriminant analysis

中图分类号:

TP391

黎毅达, 高发荣, 姚婷, 蔡利杰. 基于Fisher判别和GKF-RELM算法的多特征步态模式识别[J]. 电子学报, 2021, 49(10): 1993-2001.

LI Yi-da, GAO Fa-rong, YAO Ting, CAI Li-jie. Multi-Feature Gait Pattern Recognition Based on Fisher Discriminant and GKF-RELM Algorithm[J]. Acta Electronica Sinica, 2021, 49(10): 1993-2001.

图/表 10

参考文献 34

1	MeiC, GaoF R, LiY. A determination method for gait event based on acceleration sensors[J]. Sensors, 2019, 19(24): 5499.
2	石鹏, 张启忠, 张华平, 等. 基于多元经验模态分解的多元多尺度熵静态平衡能力评估[J]. 电子学报, 2020, 48(4): 670－674.
	ShiP, ZhangQ Z, ZhangH P, et al. Static balance capability assessment of multivariate multiscale entropy based on multivariate empirical mode decomposition[J]. Acta Electronica Sinica, 2020, 48(4): 670－674. (in Chinese)
3	RyuJ, LeeB H, KimD H. sEMG signal-based lower limb human motion detection using a top and slope feature extraction algorithm[J]. IEEE Signal Processing Letters, 2017, 24(7): 929－932.
4	ConnorP, RossA. Biometric recognition by gait: A survey of modalities and features[J]. Computer Vision and Image Understanding, 2018, 167: 1－27.
5	ZhangX F, SunS Q, LiC, et al. Impact of load variation on the accuracy of gait recognition from surface EMG signals[J]. Applied Sciences, 2018, 8(9): 1462－1476.
6	PhinyomarkA, PhukpattaranontP, LimsakulC. Feature reduction and selection for EMG signal classification[J]. Expert Systems With Applications, 2012, 39(8): 7420－7431.
7	牟永阁, 彭承琳, 郑小林, 等. 肌肉动态收缩期间表面肌电信号的时频分析[J]. 生物物理学报, 2004, 20(4): 323－328.
	MuY G, PengC L, ZhengX L, et al. Time-frequency analysis of surface myoelectric signals during dynamic contractions[J]. Acta Biophysica Sinica, 2004, 20(4): 323－328. (in Chinese)
8	TooJ, AbdullahA R, ZawawiT N, et al. Classification of EMG signal based on time domain and frequency domain features[J]. International Journal of Human and Technology Interaction, 2017, 1(1): 25－29.
9	谢平, 魏秀利, 杜义浩, 等. 基于自排序熵的表面肌电信号特征提取方法[J]. 模式识别与人工智能, 2014, 27(6): 496－501.
	XieP, WeiX L, DuY H, et al. Feature extraction method of sEMG based on auto permutation entropy[J]. Pattern Recognition and Artificial Intelligence, 2014, 27(6): 496－501. (in Chinese)
10	成娟, 陈勋, 彭虎. 基于样本熵的肌电信号起始点检测研究[J]. 电子学报, 2016, 44(2): 479－484.
	ChengJ, ChenX, PengH. An onset detection method for action surface electromyography based on sample entropy[J]. Acta Electronica Sinica, 2016, 44(2): 479－484. (in Chinese)
11	罗志增, 王人成. 基于表面肌电信号的前臂手部多运动模式识别[J]. 仪器仪表学报, 2006, 27(9): 996－999.
	LuoZ Z, WangR C. Hand movement pattern recognition based on surface electromyography[J]. Chinese Journal of Scientific Instrument, 2006, 27(9): 996－999. (in Chinese)
12	BoydJ E, LittleJ J. Biometric gait recognition[A]. Advanced Studies in Biometrics[M]. Berlin, Heidelberg, GER: Springer Berlin Heidelberg, 2005. 19－42.
13	丁帅, 王亮. 基于块稀疏贝叶斯学习的肌电信号特征提取[J]. 仪器仪表学报, 2014, 35(12): 2731－2738.
	DingS, WangL. Feature extraction of surface electromyography based on block sparse Bayesian learning[J]. Chinese Journal of Scientific Instrument, 2014, 35(12): 2731－2738. (in Chinese)
14	蔡竞, 王万良, 郑建炜, 等. 基于Fisher判别分析的增量式非负矩阵分解算法[J]. 模式识别与人工智能, 2018, 31(6): 505－515.
	CaiJ, WangW L, ZhengJ W, et al. Incremental non-negative matrix factorization based on fisher discriminant analysis[J]. Pattern Recognition and Artificial Intelligence, 2018, 31(6): 505－515. (in Chinese)
15	XiX G, TangM Y, MiranS M, et al. Evaluation of feature extraction and recognition for activity monitoring and fall detection based on wearable sEMG sensors[J]. Sensors, 2017, 17(6): 1229－1249.
16	伏云发, 徐保磊, 李永程, 等. 基于运动相关皮层电位握力运动模式识别研究[J]. 自动化学报, 2014, 40(6): 1045－1057.
	FuY F, XuB L, LiY C, et al. Recognition of actual grip force movement modes based on movement-related cortical potentials[J]. Acta Automatica Sinica, 2014, 40(6): 1045－1057. (in Chinese)
17	何国辉, 甘俊英, 李春芝, 等. 人脸与虹膜特征层融合模型的研究[J]. 电子学报, 2007, 35(7): 1365－1371.
	HeG H, GanJ Y, LiC Z, et al. A model study for face and iris feature fusion and recognition[J]. Acta Electronica Sinica, 2007, 35(7): 1365－1371. (in Chinese)
18	贲晛烨, 徐森, 王科俊. 行人步态的特征表达及识别综述[J]. 模式识别与人工智能, 2012, 25(1): 71－81.
	BenX Y, XuS, WangK J. Review on pedestrian gait feature expression and recognition[J]. Pattern Recognition and Artificial Intelligence, 2012, 25(1): 71－81. (in Chinese)
19	席旭刚, 汤敏彦, 张自豪, 等. 融合表面肌电和加速度信号的下肢运动模式识别研究[J]. 电子学报, 2017, 45(11): 2735－2741.
	XiX G, TangM Y, ZhangZ H, et al. Lower limb motion recognition based on the fusion of sEMG and acceleration signal[J]. Acta Electronica Sinica, 2017, 45(11): 2735－2741. (in Chinese)
20	王佳佳, 高发荣, 孙曜, 等. 基于sEMG的行走步态非均匀特性及其对识别的影响[J]. 传感技术学报, 2016, 29(3): 384－389.
	WangJ J, GaoF R, SunY, et al. Non-uniform characteristics and its recognition effects for walking gait based on sEMG[J]. Chinese Journal of Sensors and Actuators, 2016, 29(3): 384－389. (in Chinese)
21	LeCunY, BengioY, HintonG. Deep learning[J]. Nature, 2015, 521(7553): 436－444.
22	GharehbaghiA, LindénM. A deep machine learning method for classifying cyclic time series of biological signals using time-growing neural network[J]. IEEE Transactions on Neural Networks and Learning Systems, 2018, 29(9): 4102－4115.
23	DuanN, LiuL Z, YuX J, et al. Classification of multichannel surface-electromyography signals based on convolutional neural networks[J]. Journal of Industrial Information Integration, 2019, 15: 201－206.
24	SuriK, GuptaR. Transfer learning for sEMG-based hand gesture classification using deep learning in a master- slave architecture[A]. The 3rd International Conference on Contemporary Computing and Informatics (IC3I)[C]. Delhi, India: IEEE, 2018. 178－183.
25	MukhopadhyayA K, SamuiS. An experimental study on upper limb position invariant EMG signal classification based on deep neural network[J]. Biomedical Signal Processing and Control, 2020, 55: 101669.
26	HuangG B, ZhuQ Y, SiewC K. Extreme learning machine: Theory and applications[J]. Neurocomputing, 2006, 70(1/2/3): 489－501.
27	KuangY X, WuQ, ShaoJ K, et al. Extreme learning machine classification method for lower limb movement recognition[J]. Cluster Computing, 2017, 20(4): 3051－3059.
28	曹祥红, 刘磊, 杨鹏, 等. 利用多源信息和极限学习机的人体运动意图识别[J]. 传感技术学报, 2017, 30(8): 1171－1177.
	CaoX H, LiuL, YangP, et al. The research of locomotion-mode recognition based on multi-source information and extreme learning machine[J]. Chinese Journal of Sensors and Actuators, 2017, 30(8): 1171－1177. (in Chinese)
29	YangY M, WangY N, YuanX F. Bidirectional extreme learning machine for regression problem and its learning effectiveness[J]. IEEE Transactions on Neural Networks and Learning Systems, 2012, 23(9): 1498－1505.
30	HuangG B, SiewC K. Extreme learning machine: RBF network case[A]. ICARCV 2004 8th Control, Automation, Robotics and Vision Conference[C]. Kunming, China: IEEE, 2004. 1029－1036.
31	HuangG B, ZhouH M, DingX J, et al. Extreme learning machine for regression and multiclass classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part B, 2012, 42(2): 513－529.
32	DuttaS, SinghM, KumarA. Automated classification of non-motor mental task in electroencephalogram based brain-computer interface using multivariate autoregressive model in the intrinsic mode function domain[J]. Biomedical Signal Processing and Control, 2018, 43: 174－182.
33	KrizhevskyA, SutskeverI, HintonG E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84－90.
34	SzegedyC, LiuW, JiaY Q, et al. Going deeper with convolutions[A]. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)[C]. Boston, MA, USA: IEEE, 2015. 1－9.

组别		支撑前期	支撑中期	支撑后期	摆动前期	摆动后期	平均识别率
Z₁	ELM	97.22	91.67	85.60	86.11	79.63	88.05
Z₁	GKF-RELM	94.44	93.06	88.02	85.19	87.96	89.73
Z₂	ELM	97.25	90.97	82.40	88.89	77.78	87.46
Z₂	GKF-RELM	97.22	94.44	90.40	96.30	87.04	93.08
Z₃	ELM	97.22	87.50	90.40	87.96	80.56	88.73
Z₃	GKF-RELM	97.22	92.36	89.60	91.67	93.52	92.87
Z₄	ELM	93.06	95.31	86.40	98.15	84.03	91.39
Z₄	GKF-RELM	94.44	92.36	95.20	97.22	93.52	94.55
Z₅	ELM	94.97	95.51	95.50	96.24	96.59	95.76
Z₅	GKF-RELM	97.22	99.31	96.80	99.07	96.30	97.74

组别		支撑前期	支撑中期	支撑后期	摆动前期	摆动后期	平均识别率
Z₁	ELM	97.22	91.67	85.60	86.11	79.63	88.05
Z₁	GKF-RELM	94.44	93.06	88.02	85.19	87.96	89.73
Z₂	ELM	97.25	90.97	82.40	88.89	77.78	87.46
Z₂	GKF-RELM	97.22	94.44	90.40	96.30	87.04	93.08
Z₃	ELM	97.22	87.50	90.40	87.96	80.56	88.73
Z₃	GKF-RELM	97.22	92.36	89.60	91.67	93.52	92.87
Z₄	ELM	93.06	95.31	86.40	98.15	84.03	91.39
Z₄	GKF-RELM	94.44	92.36	95.20	97.22	93.52	94.55
Z₅	ELM	94.97	95.51	95.50	96.24	96.59	95.76
Z₅	GKF-RELM	97.22	99.31	96.80	99.07	96.30	97.74

组别		支撑前期	支撑中期	支撑后期	摆动前期	摆动后期	平均识别率
MPF	ELM	95.14	96.53	94.40	95.20	94.44	95.14
MPF	GKF-RELM	97.22	95.14	96.80	99.08	96.30	96.91
MF	ELM	97.22	83.33	84.00	77.78	65.74	81.61
MF	GKF-RELM	94.44	84.03	93.60	89.81	79.63	88.30
MPF+MF	ELM	94.44	96.53	97.60	96.80	96.30	96.33
MPF+MF	GKF-RELM	97.22	99.30	96.80	99.07	98.15	98.11

组别		支撑前期	支撑中期	支撑后期	摆动前期	摆动后期	平均识别率
MPF	ELM	95.14	96.53	94.40	95.20	94.44	95.14
MPF	GKF-RELM	97.22	95.14	96.80	99.08	96.30	96.91
MF	ELM	97.22	83.33	84.00	77.78	65.74	81.61
MF	GKF-RELM	94.44	84.03	93.60	89.81	79.63	88.30
MPF+MF	ELM	94.44	96.53	97.60	96.80	96.30	96.33
MPF+MF	GKF-RELM	97.22	99.30	96.80	99.07	98.15	98.11

组别		支撑前期	支撑中期	支撑后期	摆动前期	摆动后期	平均识别率
SE	ELM	86.82	83.20	79.24	85.14	83.96	83.67
SE	GKF-RELM	87.45	87.64	86.85	90.72	89.45	88.42
FE	ELM	91.67	75.69	90.40	98.15	77.78	86.74
FE	GKF-RELM	94.44	81.94	93.60	95.37	89.81	91.03
PE	ELM	85.47	80.56	80.25	88.14	83.13	83.51
PE	GKF-RELM	88.60	81.84	82.61	90.51	86.15	85.94
SE+FE	ELM	94.44	85.42	96.00	93.52	89.81	91.84
SE+FE	GKF-RELM	95.83	89.06	95.81	95.14	93.06	93.78
SE+PE	ELM	86.23	89.22	85.42	82.64	90.47	86.80
SE+PE	GKF-RELM	92.61	90.28	93.48	92.87	94.14	92.68
FE+PE	ELM	91.67	77.78	84.80	93.52	87.96	87.15
FE+PE	GKF-RELM	91.67	86.46	87.43	97.22	92.36	91.03
SE+FE+PE	ELM	94.44	83.33	97.60	93.52	92.59	92.30
SE+FE+PE	GKF-RELM	97.92	83.85	97.01	95.83	93.75	93.67