·132· 智能系统学报 第14卷 通过对CoST数据集的7805个手势进行分析，对 France,2013:1679-1688 数据集进行数据预处理，剔除部分错误的手势， [6]GUEST S,DESSIRIER J M,MEHRABYAN A,et al.The 使数据集更加完善。使用MATLAB等软件对数 development and validation of sensory and emotional 据进行整理，并通过相关论文引证，提出一系列 scales of touch perception[J].Attention,perception,psy- 情感识别的特征。使用多种分类器进行分类比 chophysics,2011,73(2)y:531-550. 对，整体来看14种手势中SVM和随机森林的情 [7]KIM N Y.SHIN Y.KIM E Y.Emotion-based textile in- dexing using neural networks[C]//12th International Con- 感识别效果不相伯仲。不同分类器下的stroke手 ference on Human-Computer Interaction.Beijing,China, 势的情感识别效果均为最高，但ELM的stroke手 2007:349-357. 势的情感识别效果要比基于RBF核函数的SVM [8]HUANG Xinyin,SOBUE S,KANDA T,et al.Linking 分类器的效果好(SVM=70.95%,ELM=72.07%)。 KANSAI and image features by multi-layer neural net- 并且ELM的识别时间要明显短于SVM(SVM= works[C]//11th International Conference on Knowledge- 0.33s,ELM=0.04s)。本文针对CoST数据集进行 Based and Intelligent Information and Engineering Sys- 了一系列研究实验，得到了3点结论：stroke手势 tems.Vietri sul Mare,Italy,2007:318-325 具有最好的情感识别效果，且分类精度较高：ELM [9]JUNG MM,CANG X L,POEL M,et al.Touch 作为触觉情感识别的分类器具有较好的表现，识 challenge'15:recognizing social touch gestures[C]//Pro- 别精度高且识别速度快；有的手势（比如pinch、 ceedings of the 2015 ACM on International Conference on press)本身对应着某种情感，会影响情感识别的 Multimodal Interaction.Seattle,Washington,USA,2015: 结果。因此，在设计触觉情感识别传感器时，可 387-390 以诱导用户做出情感识别率最高的手势，从而间 [10]ZHOU Nan,DU Jun.Recognition of social touch ges- tures using 3D convolutional neural networks[C]// 接地提高分类精度。这为以后的触觉情感识别系 Chinese Conference on Pattern Recognition.Chengdu, 统的设计提供了思路，也为建立新的情感识别数 China,2016:164173 据集奠定了基础。 [11]HUGHES D,LAMMIE J,CORRELL N.A Robotic skin 参考文献： for collision avoidance and affective touch recognition[J]. IEEE robotics and automation letters,2018,3(3): [1]马蕊，刘华平，孙富春，等.基于触觉序列的物体分类[U 1386-1393. 智能系统学报，2015,10(3)：362-368. [12]MARAMIS C,STEFANOPOULOS L,CHOUVARDA I. MA Rui,LIU Huaping.SUN Fuchun,et al.Object classi- et al.Emotion recognition from haptic touch on android fication based on the tactile sequence[J].CAAI transac- device screens[M]//MAGLAVERAS N,CHOUVARDA tions on intelligent systems,2015,10(3):362-368 I,DE CARVALHO P.Precision Medicine Powered by [2]郝敏，刘光远，温万惠.基于进化策略的生理信号情感识 pHealth and Connected Health.Singapore:Springer, 别J.智能系统学报，2009,4(4)：352-356 2018:205-209 HAO Min,LIU Guangyuan,WEN Wanhui.Recognition of [13]GAO Yuan,BIANCHI-BERTHOUZE N,MENG Hongy- emotion in physiological signals using evolutionary ing.What does touch tell us about emotions in touch- strategies.CAAl transactions on intelligent systems, screen-based gameplay?[J].ACM transactions on com- 2009,4(4):352-356 puter-human interaction,2012,19(4):31 [3]MORRISON I,LOKEN L S,OLAUSSON H.The skin as [14]林连冬，李思奇，陈春雨，等.触觉传感器非线性补偿仿 a social organ[J].Experimental brain research,2010. 生算法.哈尔滨工程大学学报，2017,38(2)：288-292. 2043):305-314 LIN Liandong,LI Siqi,CHEN Chunyu,et al.Bionic al- [4]DEBROT A.SCHOEBI D.PERREZ M.et al.Touch as an gorithm for nonlinear compensation of tactile sensors[] interpersonal emotion regulation process in couples'daily Journal of Harbin engineering university,2017,38(2) lives:the mediating role of psychological intimacy[J].Per- 288-292 sonality and social psychology bulletin,2013,39(10): [15]HUANG Guangbin,WANG Dianhui,LAN Yuan.Ex- 1373-1385. treme learning machines:a survey[J].International journ- [5]PARK Y W,BAEK K M,NAM T J.The roles of touch al of machine learning and cybernetics,2011,2(2): during phone conversations:long-distance couples'use of 107-122. POKE in their homes[Cl//Proceedings of the SIGCHI Con- [16]JUNG MM,POPPE R,POEL M,et al.Touching the ference on Human Factors in Computing Systems.Paris, void--introducing CoST:corpus of social touch[C]//Pro-通过对 CoST 数据集的 7 805 个手势进行分析，对 数据集进行数据预处理，剔除部分错误的手势， 使数据集更加完善。使用 MATLAB 等软件对数 据进行整理，并通过相关论文引证，提出一系列 情感识别的特征。使用多种分类器进行分类比 对，整体来看 14 种手势中 SVM 和随机森林的情 感识别效果不相伯仲。不同分类器下的 stroke 手 势的情感识别效果均为最高，但 ELM 的 stroke 手 势的情感识别效果要比基于 RBF 核函数的 SVM 分类器的效果好 (SVM=70.95%，ELM=72.07%)。 并且 ELM 的识别时间要明显短于 SVM(SVM= 0.33 s，ELM=0.04 s)。本文针对 CoST 数据集进行 了一系列研究实验，得到了 3 点结论：stroke 手势 具有最好的情感识别效果，且分类精度较高；ELM 作为触觉情感识别的分类器具有较好的表现，识 别精度高且识别速度快；有的手势 (比如 pinch、 press) 本身对应着某种情感，会影响情感识别的 结果。因此，在设计触觉情感识别传感器时，可 以诱导用户做出情感识别率最高的手势，从而间 接地提高分类精度。这为以后的触觉情感识别系 统的设计提供了思路，也为建立新的情感识别数 据集奠定了基础。 参考文献： 马蕊, 刘华平, 孙富春, 等. 基于触觉序列的物体分类[J]. 智能系统学报, 2015, 10(3): 362–368. MA Rui, LIU Huaping, SUN Fuchun, et al. Object classi￾fication based on the tactile sequence[J]. CAAI transac￾tions on intelligent systems, 2015, 10(3): 362–368. [1] 郝敏, 刘光远, 温万惠. 基于进化策略的生理信号情感识 别[J]. 智能系统学报, 2009, 4(4): 352–356. HAO Min, LIU Guangyuan, WEN Wanhui. Recognition of emotion in physiological signals using evolutionary strategies[J]. CAAI transactions on intelligent systems, 2009, 4(4): 352–356. [2] MORRISON I, LÖKEN L S, OLAUSSON H. The skin as a social organ[J]. Experimental brain research, 2010, 204(3): 305–314. [3] DEBROT A, SCHOEBI D, PERREZ M, et al. Touch as an interpersonal emotion regulation process in couples’daily lives: the mediating role of psychological intimacy[J]. Per￾sonality and social psychology bulletin, 2013, 39(10): 1373–1385. [4] PARK Y W, BAEK K M, NAM T J. The roles of touch during phone conversations: long-distance couples’use of POKE in their homes[C]//Proceedings of the SIGCHI Con￾ference on Human Factors in Computing Systems. Paris, [5] France, 2013: 1679–1688. GUEST S, DESSIRIER J M, MEHRABYAN A, et al. The development and validation of sensory and emotional scales of touch perception[J]. Attention, perception, & psy￾chophysics, 2011, 73(2): 531–550. [6] KIM N Y, SHIN Y, KIM E Y. Emotion-based textile in￾dexing using neural networks[C]//12th International Con￾ference on Human-Computer Interaction. Beijing, China, 2007: 349–357. [7] HUANG Xinyin, SOBUE S, KANDA T, et al. Linking KANSAI and image features by multi-layer neural net￾works[C]//11th International Conference on Knowledge￾Based and Intelligent Information and Engineering Sys￾tems. Vietri sul Mare, Italy, 2007: 318–325. [8] JUNG M M, CANG X L, POEL M, et al. Touch challenge’15: recognizing social touch gestures[C]//Pro￾ceedings of the 2015 ACM on International Conference on Multimodal Interaction. Seattle, Washington, USA, 2015: 387–390. [9] ZHOU Nan, DU Jun. Recognition of social touch ges￾tures using 3D convolutional neural networks[C]// Chinese Conference on Pattern Recognition. Chengdu, China, 2016: 164–173. [10] HUGHES D, LAMMIE J, CORRELL N. A Robotic skin for collision avoidance and affective touch recognition[J]. IEEE robotics and automation letters, 2018, 3(3): 1386–1393. [11] MARAMIS C, STEFANOPOULOS L, CHOUVARDA I, et al. Emotion recognition from haptic touch on android device screens[M]//MAGLAVERAS N, CHOUVARDA I, DE CARVALHO P. Precision Medicine Powered by pHealth and Connected Health. Singapore: Springer, 2018: 205–209. [12] GAO Yuan, BIANCHI-BERTHOUZE N, MENG Hongy￾ing. What does touch tell us about emotions in touch￾screen-based gameplay?[J]. ACM transactions on com￾puter-human interaction, 2012, 19(4): 31. [13] 林连冬, 李思奇, 陈春雨, 等. 触觉传感器非线性补偿仿 生算法[J]. 哈尔滨工程大学学报, 2017, 38(2): 288–292. LIN Liandong, LI Siqi, CHEN Chunyu, et al. Bionic al￾gorithm for nonlinear compensation of tactile sensors[J]. Journal of Harbin engineering university, 2017, 38(2): 288–292. [14] HUANG Guangbin, WANG Dianhui, LAN Yuan. Ex￾treme learning machines: a survey[J]. International journ￾al of machine learning and cybernetics, 2011, 2(2): 107–122. [15] JUNG M M, POPPE R, POEL M, et al. Touching the void--introducing CoST: corpus of social touch[C]//Pro- [16] ·132· 智 能 系 统 学 报 第 14 卷