一种改进脑电特征提取算法及其在情感识别中的应用
发布时间：2018-11-20 12:56
【摘要】：音乐诱发下的情感状态评估结果可为辅助音乐治疗提供理论支持与帮助。情感状态评估的关键是情感脑电的特征提取,故本文针对情感脑电特征提取算法的性能优化问题开展研究。采用Koelstra等提出的分析人类情绪状态的多模态标准数据库DEAP,提取8种正负情绪代表各个脑区的14个通道脑电数据,基于小波分解重构δ、θ、α、β四种节律波;在分析比较小波特征(小波系数能量和小波熵)、近似熵和Hurst指数三种脑电特征情感识别效果的基础上,提出一种基于主成分分析(PCA)融合小波特征、近似熵和Hurst指数的脑电特征提取算法。本算法保留累积贡献率大于85%的主成分,并选择特征根差异较大的特征参数,基于支持向量机实现情感状态评估。结果表明,使用单一小波特征(小波系数能量和小波熵)、近似熵和Hurst指数特征量,情感识别的正确率均值分别是73.15%、50.00%和45.54%,而改进算法识别准确率均值在85%左右。基于改进算法情感识别的分类准确率比传统方法至少能提升12%,可为情感脑电特征提取以及辅助音乐治疗提供帮助。
[Abstract]:The results of music-induced emotional state evaluation can provide theoretical support and help for music therapy. The key to emotional state evaluation is the feature extraction of emotional EEG, so this paper focuses on the performance optimization of EEEG feature extraction algorithm. The multimodal standard database DEAP, proposed by Koelstra et al is used to extract 14 channel EEG data from 8 positive and negative emotions representing various brain regions, and to reconstruct 未, 胃, 伪 and 尾 rhythms based on wavelet decomposition. On the basis of analyzing and comparing the emotional recognition effects of wavelet features (wavelet coefficient energy and wavelet entropy), approximate entropy and Hurst exponent, a (PCA) fusion wavelet feature based on principal component analysis (PCA) is proposed. EEG feature extraction algorithm based on approximate entropy and Hurst exponent. In this algorithm, the principal components whose cumulative contribution rate is more than 85% are retained, and the feature parameters with large differences in feature roots are selected to realize emotional state evaluation based on support vector machine. The results show that using single wavelet feature (wavelet coefficient energy and wavelet entropy), approximate entropy and Hurst exponent feature, the average accuracy of emotion recognition is 73.1550.00% and 45.54g, respectively. The average recognition accuracy of the improved algorithm is about 85%. The classification accuracy of emotion recognition based on the improved algorithm is at least 12% higher than that of the traditional method, which can help to extract emotional EEG features and assist music therapy.
【作者单位】： 燕山大学电气工程学院生物医学工程研究所;河北省测试计量技术及仪器重点实验室;北京工业大学生命科学与生物工程学院;
【基金】：国家自然科学基金项目(51677162) 河北省自然科学基金项目(F2014203244) 中国博士后科学基金项目(2014M550582)
【分类号】：R318;TN911.7
，
本文编号：2344975
[Abstract]:The results of music-induced emotional state evaluation can provide theoretical support and help for music therapy. The key to emotional state evaluation is the feature extraction of emotional EEG, so this paper focuses on the performance optimization of EEEG feature extraction algorithm. The multimodal standard database DEAP, proposed by Koelstra et al is used to extract 14 channel EEG data from 8 positive and negative emotions representing various brain regions, and to reconstruct 未, 胃, 伪 and 尾 rhythms based on wavelet decomposition. On the basis of analyzing and comparing the emotional recognition effects of wavelet features (wavelet coefficient energy and wavelet entropy), approximate entropy and Hurst exponent, a (PCA) fusion wavelet feature based on principal component analysis (PCA) is proposed. EEG feature extraction algorithm based on approximate entropy and Hurst exponent. In this algorithm, the principal components whose cumulative contribution rate is more than 85% are retained, and the feature parameters with large differences in feature roots are selected to realize emotional state evaluation based on support vector machine. The results show that using single wavelet feature (wavelet coefficient energy and wavelet entropy), approximate entropy and Hurst exponent feature, the average accuracy of emotion recognition is 73.1550.00% and 45.54g, respectively. The average recognition accuracy of the improved algorithm is about 85%. The classification accuracy of emotion recognition based on the improved algorithm is at least 12% higher than that of the traditional method, which can help to extract emotional EEG features and assist music therapy.
【作者单位】： 燕山大学电气工程学院生物医学工程研究所;河北省测试计量技术及仪器重点实验室;北京工业大学生命科学与生物工程学院;
【基金】：国家自然科学基金项目(51677162) 河北省自然科学基金项目(F2014203244) 中国博士后科学基金项目(2014M550582)
【分类号】：R318;TN911.7
，
本文编号：2344975
本文链接：https://www.wllwen.com/yixuelunwen/swyx/2344975.html
