基于改进的Relief F多生理信号情绪识别算法  

Multi-physiological signal emotion recognition algorithm based on improved Relief F

作　　者：张晓丹[1] 杜金祥 李涛[1] 佘翼翀 赵瑞 柯熙政[3] 康俊玮 王舒仪 ZHANG Xiaodan;DU Jinxiang;LI Tao;SHE Yichong;ZHAO Rui;KE Xizheng;KANG Junwei;WANG Shuyi(School of Electronic and Information, Xi’an Polytechnic University, Xi’an 710048, China;College of Life Science and Technology, Xidian University, Xi’an 710071, China;School of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710054, China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048 [2]西安电子科技大学生命科学技术学院,陕西西安710071 [3]西安理工大学自动化与信息工程学院,陕西西安710054

出　　处：《西安工程大学学报》2022年第2期40-48,共9页Journal of Xi’an Polytechnic University

基　　金：国家自然科学基金青年科学基金(81901827);西安市科技局创新引导基金(201805030YD8CG14(9));国家级大学生创新创业训练项目(202110709052)。

摘　　要：针对单一生理信号特征信息不足以及个体特异性与全局阈值不匹配导致的情绪识别正确率低的问题,提出了一种改进的Relief F匹配多生理信号特征选择算法。通过小波包分解多生理信号并重构与情绪相关的6个波段,以及经验模态分解提取基于小波系数和重构信号本征模函数分量的8类特征;使用Relief F算法先获得优选特征组,再构建优化特征组权重获得全局最优匹配特征组,以及与其对应的匹配通道;并采用概率神经网络结合全局最优匹配特征组训练情绪分类模型。结果表明:该方法能够较好地对愉悦、愤怒、放松、悲伤4类情绪进行分类,其平均识别正确率分别为90.89%、85.39%、82.81%、87.56%,对比单一生理信号平均提升了1.76%,验证了此方法的有效性。In view of the problem of low accuracy of emotion recognition caused by insufficient feature information of a single physiological signal and the mismatch between individual specificity and the global threshold,an improved Relief F matching multi-physiological signal feature selection algorithm was proposed.Firstly,wavelet packet was used to decompose multiple physiological signals and reconstruct them into six emotional-related bands,and extract 8 types of features based on wavelet coefficients and reconstructed signal IMF components through empirical mode decomposition.Secondly,the Relief F algorithm was used to obtain the preferred feature group first,and then the optimized feature group weight formula was constructed to obtain the global optimal matching feature group and its corresponding matching channel.Finally,the PNN method was used to train the sentiment classification model with the data of the global optimal matching feature set and channel.The results show that the proposed method can classify happiness,anger,relaxation and sadness well.The average recognition accuracy rates are 90.89%,85.39%,82.81%and 87.56%,respectively,which is an average increase of 1.76%compared to a single physiological signal.The effectiveness of the proposed method was verified.

关 键 词：多生理信号 情绪识别 小波包分解 经验模态分解 概率神经网络 

分 类 号：TN911.7[电子电信—通信与信息系统]