陶嘉棟 尹鐘
摘 要:為了提高腦力負荷分類準確率,提出一種將Bagging和極限學(xué)習(xí)機相結(jié)合的集成算法。用極限學(xué)習(xí)機(ELM)作為底層弱分類器,通過多數(shù)投票方式?jīng)Q定最終類別的標簽,從而構(gòu)建最終強分類器。實驗結(jié)果表明,在腦力負荷識別研究問題上,該集成算法的分類準確率在4個被試數(shù)據(jù)集上分別達到了96.17%、96.02%、92.50%和93.50%。相較于傳統(tǒng)的ELM算法,分類準確率在4個被試數(shù)據(jù)集上分別提升了1.59%、1.34%、2.86%和1.80%。并且新算法在精確率、靈敏度和特異度等評估標準上均高于傳統(tǒng)ELM分類器。
關(guān)鍵詞:Bagging;極限學(xué)習(xí)機;集成算法;腦力負荷
DOI:10. 11907/rjdk. 191593
中圖分類號:TP301 ? 文獻標識碼:A??????????????? 文章編號:1672-7800(2020)003-0027-04
Mental Workload Recognition Model Based on Bagging
and Extreme Learning Machine
TAO Jia-dong, YIN Zhong
(School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology,
Shanghai 200093, China)
Abstract: In order to improve the accuracy of mental workload classification, an integrated algorithm combining Bagging and extreme learning machine is proposed. The extreme learning machine (ELM) is used as the weak classifier at the bottom level and the label of the final class is determined by majority voting to construct the final strong classifier. The experimental results show that the classification accuracy of the proposed algorithm was 96.17%, 96.02%, 92.50% and 93.50% on the four data sets, respectively. Compared with the traditional ELM algorithm, the classification accuracy rate was increased by 1.59%, 1.34%, 2.86% and 1.80% on the four data sets, respectively. Moreover, the new algorithm is superior to the traditional ELM classifier in accuracy, sensitivity and specificity.
Key Words: Bagging; extreme learning machine; integrated algorithm; mental workload
0 引言
在當前人機協(xié)同系統(tǒng)中,利用機器學(xué)習(xí)中的不同算法模型對腦電信號進行分類尤為必要。識別和評估人的腦力負荷水平可以有效避免人的腦力負荷帶來的風(fēng)險。腦力負荷一般可以被定義為在短暫的任務(wù)需求下剩余工作記憶的認知資源或能力[1]。近年來,許多研究人員利用神經(jīng)信號分析技術(shù)對腦電等生理生物標志物進行神經(jīng)信號分析,這為實時和定量評價腦力負荷變化提供了依據(jù)[2-3]。邱靜等[4]提出了一種名為原功率調(diào)制的新型特征提取方式,結(jié)合傳統(tǒng)信號分類方法對腦電信號進行了分類;劉寶善等[5]建立殲擊機飛行員腦力負荷評價模型評定飛機飛行品質(zhì);衛(wèi)宗敏等[6]在飛行模擬任務(wù)條件下研究了三級腦力負荷測量與評價;劉維平等[7]提出了基于任務(wù)的乘員腦力負荷典型主觀評價方法;ZHAO等[8]提出了一種利用支持向量機(Support Vector Machine,SVM)對跨任務(wù)腦力負荷水平進行分類的生理異常檢測系統(tǒng);ZHANG等[9]提出了一種自回歸建模方法,研究了心率變異性與腦力負荷的關(guān)系,從而提高了對腦力負荷水平的評估精度。
最近研究表明,極限學(xué)習(xí)機(Extreme Learning Machine,ELM)是一個很好的腦力負荷分類器。ELM代表一套機器學(xué)習(xí)技術(shù),證明隱藏層神經(jīng)元無需調(diào)整[10]。腦電信號已廣泛用于警戒估計,以避免風(fēng)險[11]。ELM算法可以實現(xiàn)對腦電特征學(xué)習(xí)的訓(xùn)練和處理[12-13]。盡管ELM算法有以上優(yōu)勢,但依然存在許多問題。比如ELM模型的輸出權(quán)重是固定的,導(dǎo)致模型無法適用于所有腦電數(shù)據(jù)。目前,許多研究人員提出了改善原始ELM模型的方法。比如,賈偉等[14]利用粒子群優(yōu)化算法優(yōu)化原始ELM,從而改善其分類效果;閆河等[15]提出卷積神經(jīng)網(wǎng)絡(luò)與ELM相結(jié)合的新分類方法;魏思政等[16]提出了一種基于深度信念網(wǎng)絡(luò)和ELM相結(jié)合的新檢測方法,先用深度信念網(wǎng)絡(luò)提取特征,再用ELM進行分類,最終取得良好分類效果;尹剛等[17]在ELM中引入結(jié)構(gòu)風(fēng)險最小化理論,用小波函數(shù)替代原有的隱層激勵函數(shù),與在線學(xué)習(xí)方法相結(jié)合,改善了原始ELM的分類性能。
[1]SANJRAM P K. Attention and intended action in multitasking: an understanding of workload[J]. Displays, 2013, 34(4): 283-291.
[2]TING C, MAHFOUF,NASSEF A M,et al. Real-time adaptive automation system based on identification of operator functional state in simulated process control operations[J]. IEEE Transactions on Systems, Man and Cybernetics-Part A: System and Humans, 2010, 40(2): 251-262.
[3]MAJUMDAR K. Human scalp EEG processing: various soft computing approaches[J]. Applied Soft Computing,2011,11(8):4433-4447.
[4]邱靜,高龍,盧軍. 實時EEG腦力負荷預(yù)測研究[J]. 人類工效學(xué), 2015,21(3): 10-13.
[5]劉寶善,廖建橋,王玉紅,等. 殲擊機飛行員腦力負荷評價模型的研究[J]. 中華航空航天醫(yī)學(xué)雜志,2000,11(1): 24-24.
[6]衛(wèi)宗敏,完顏笑如,莊達民. 飛機座艙顯示界面腦力負荷測量與評價[J]. 北京航空航天大學(xué)學(xué)報,2014, 40(1): 86-91.
[7]劉維平,趙凱旋,聶俊峰,等. 基于任務(wù)的乘員腦力負荷典型主觀評價方法[J]. 科學(xué)技術(shù)與工程,2018, 18(14): 131-135.
[8]ZHAO G Z,LIU Y J,SHI Y C.Real-time assessment of the cross-task mental workload using physiological measures during anomaly detection[J]. IEEE Transactions on Human-Machine Systems,2018,48(2): 149-160.
[9]ZHANG J H,WANG X Y,MAHFOUF M, et al. Use of heart rate variability analysis for quantitatively?assessing?operators?mental?workload[J]. IEEE Conferences, 2008, 1(1): 668-672.
[10]HUANG G B. An insight into extreme learning machines: random neurons, random features and kernels[J]. Cognitive Computation, 2014, 6(3): 376-90.
[11]SHI L C,LU B L. EEG-based vigilance estimation using extreme learning machines[J]. Neurocomputing, 2013, 102(15): 135-143.
[12]YUAN Q,ZHOU W D.Epileptic EEG classification based on extreme learning machine and nonlinear features[J]. Epilesy Research, 2011, 96(1-2): 29-38.
[13]HUANG G B,WANG D. Advances in extreme learning machines[J]. Neurocomputing, 2011, 74(16): 2411-2412.
[14]賈偉,華慶一,張敏軍,等.改進極限學(xué)習(xí)機的移動界面模式半監(jiān)督分類[J]. 計算機工程與應(yīng)用, 2018, 54(2): 11-19.
[15]閆河,王鵬,董鶯艷,等.基于深度CNN和極限學(xué)習(xí)機相結(jié)合的實時文檔分類[J].計算機應(yīng)用與軟件, 2019, 36(3): 174-179.
[16]魏思政,劉厚泉,趙志凱.基于DBN-ELM的入侵檢測研究[J].計算機工程, 2018, 44(9): 153-158.
[17]尹剛,張英堂,李志寧,等. 改進在線貫序極限學(xué)習(xí)機在模式識別中的應(yīng)用[J]. 計算機工程, 2012,38(8): 164-166.
[18]HUANG G B, ZHU Y Q,SIEW C K. Extreme learning machine: a new learning scheme of feedforward neural networks[C]. Budapest:2004 IEEE International Joint Conference on Nenral Networks, 2004.
[19]HUANG G B,ZHU Y Q,SIEW C K. Extreme Learning machine:Theory and applications[J]. Neurocomputing,2006,70(1-3):489-501.
[20]姚明海,趙連朋,劉維學(xué). 基于特征選擇的Bagging分類算法研究[J]. 計算機技術(shù)與發(fā)展, 2014, 24(4): 103-106.
[21]ZHANG J,YIN Z,WANG R. Recognition of mental workload levels under complex human-machine collaboration by using physiological features and adaptive support vector machines[J]. IEEE Transactions on Human-Machine Systems, 2015, 45(2): 200-214.
[22]ZHANG Z, ZHAO M, CHOW T W S. Constrained large margin local projection algorithms and extensions for multimodal dimensionality reduction[J].? Pattern Recognition, 2012, 45(12): 4466-4493.
(責(zé)任編輯:孫 娟)
收稿日期:2019-05-07
基金項目:國家自然科學(xué)基金青年項目(61703277);上海青年科技英才揚帆計劃項目(17YF1427000)
作者簡介:陶嘉棟(1996-),男,上海理工大學(xué)光電信息與計算機工程學(xué)院碩士研究生,研究方向為基于腦電信號的腦力負荷分類;尹鐘(1988-),男,博士,上海理工大學(xué)光電信息與計算機工程學(xué)院副教授,研究方向為生理信號智能處理算法與模式識別、認知任務(wù)負荷檢測。本文通訊作者:尹鐘。