心音信号MFCC特征向量提取方法的优化
|
摘要
为了提高利用梅尔频率倒谱系数(Mel-frequency cepstral coefficients, MFCC)特征向量进行心音信号分类的准确率,本文提出以一种基于独立成分分析(independent component analysis, ICA)及权值优化的MFCC特征向量优化方法。首先,通过消除趋势项、降噪、提取心动周期与基础心音分割等步骤对心音信号预处理;接着,对提取的基础心音信号做Mel频谱变换及倒谱分析提取MFCC特征向量,其中用ICA替代离散余弦变换去除分量间高阶量的相关性,同时采用相关系数为权值优化整体混合矩阵;最后,采用F比衡量特征向量贡献率,并以其为权值优化各维特征向量。通过提取MFCC特征向量采用支持向量机(support vector machine, SVM)的分类器识别第一心音及第二心音,并与人工标注心音状态集进行对比。实验结果表明,基于ICA及权值优化的MFCC特征向量在SVM分类器中识别率得到了有效的提升,且优化算法具备一定抗噪性能。
In order to improve the accuracy of heart sound signal classification using the Mel frequency cepstral coefficient(MFCC)feature vector, we have proposed an MFCC feature vector optimization method based on independent component analysis(ICA)and weight optimization. First, the heart sound signal is preprocessed by removing the trend terms, noise reduction, extracting the cardiac cycle and fundamental heart sounds segmentation, etc. Then, the MFCC feature vector is extracted by performing Mel spectrum transform and cepstrum analysis on each extracted fundamental heart sounds. In the process, the discrete cosine transform is replaced by ICA to remove the correlation of high-order quantities between components, and the correlation coefficient is used as the weight to optimize the overall mixing matrix. Finally, the F-ratio is used as the feature vector contribution rate, and each dimension feature vector is optimized with the weight. The first heart sound and the second heart sound are identified by the support vector machine(SVM)classifier by extracting the MFCC feature vector, and compared with the artificially labeled heart sound states set. The experimental results show that the recognition rate of MFCC feature vector based on ICA and weight optimization is effectively improved in SVM classifier, and the optimization algorithm has certain anti-noise performance.
In order to improve the accuracy of heart sound signal classification using the Mel frequency cepstral coefficient(MFCC)feature vector, we have proposed an MFCC feature vector optimization method based on independent component analysis(ICA)and weight optimization. First, the heart sound signal is preprocessed by removing the trend terms, noise reduction, extracting the cardiac cycle and fundamental heart sounds segmentation, etc. Then, the MFCC feature vector is extracted by performing Mel spectrum transform and cepstrum analysis on each extracted fundamental heart sounds. In the process, the discrete cosine transform is replaced by ICA to remove the correlation of high-order quantities between components, and the correlation coefficient is used as the weight to optimize the overall mixing matrix. Finally, the F-ratio is used as the feature vector contribution rate, and each dimension feature vector is optimized with the weight. The first heart sound and the second heart sound are identified by the support vector machine(SVM)classifier by extracting the MFCC feature vector, and compared with the artificially labeled heart sound states set. The experimental results show that the recognition rate of MFCC feature vector based on ICA and weight optimization is effectively improved in SVM classifier, and the optimization algorithm has certain anti-noise performance.
引文
[1] 刘喻,唐雪辉,陈洪波,等.基于PCA-SOM的异常心音分类识别方法的研究[J].航天医学与医学工程,2018,31(1):57- 61.Liu Yu,Tang Xuehui,Chen Hongbo,et al.Clustering Recognition Method for Abnormal Heart Sounds Based on PCA-SOM[J].Space Medicine & Medical Engineering,2018,31(1):57- 61.(in Chinese)
[2] 成谢锋,李允怡,高珮熙,等.一种心音信号的源成分获取方法[J].南京邮电大学学报:自然科学版,2018,38(1):54-59.Cheng Xiefeng,Li Yunyi,Gao Peixi,et al.New method for obtaining components of heart sound signal source[J].Journal of Nanjing University of Posts and Telecommun-ications:Natural Science Edition,2018,38(1):54-59.(in Chinese)
[3] 巩燕,胡杰,高彬,等.心血管疾病即时检测技术的研究进展[J].中国科学:技术科学,2016,46(11):1116-1134.Gong Y,Hu J,Gao B,et al.Advances in monitoring of cardiovascular diseases at the point of care[J].Sci Sin Tech,2016,46(11):1116-1134.(in Chinese)
[4] Berkaya S K,Uysal A K,Gunal E S,et al.A survey on ECG analysis[J].Biomedical Signal Processing & Control,2018,43:216-235.
[5] 侯雷静,应冬文,国雁萌,等.基于心音分类的识别器分析[C]//中国声学学会2017年全国声学学术会议.中国声学学会,2017.Hou Leijing,Ying Dongwen,Guo Yanmeng,et al.Identifier analysis based on heart sound classification[C]//2017 National Acoustic Conference of the Chinese Acoustic Society.Chinese Acoustic Society,2017.(in Chinese)
[6] Puneet Kumar J,Anil Kumar T.Heart monitoring systems-a review[J].Computers in Biology & Medicine,2014,54(C):1-13.
[7] Rubin J,Rui A,Ganguli A,et al.Classifying heart sound recordings using deep convolutional neural networks and Mel:Frequency Cepstral Coefficients[C]//2016 Computing in Cardiology Conference.IEEE,2016.
[8] Zhang L,Tang R,Jiang J,et al.Research on pretreatment and envelope extraction algorithm of heart sound signal[J].Journal of Biomedical Engineering,2014,31(4):734-741.
[9] Lin Y,ChongYang W,ZhiCheng W.Research on the selection of wavelet bases for wavelet-based signal trend elimination[C]//International Conference on Wavelet Active Media Technology & Information Processing.IEEE,2013.
[10] Jain P K,Tiwari A K.An adaptive thresholding method for the wavelet based denoising of phonocardiogram signal[J].Biomedical Signal Processing & Control,2017,38:388-399.
[11] Whitaker B M,Suresha P B,Liu C,et al.Combining sparse coding and time-domain features for heart sound classification[J].Physiological Measurement,2017.
[12] Salman A H,Ahmadi N,Mengko R,et al.Automatic Segmentation and Detection of Heart Sound Components S1,S2,S3 and S4[C]//International Conference on Instrumentation.IEEE,2016.
[13] 马勇,鲍长春.基于高层信息特征的重叠语音检测[J].清华大学学报:自然科学版,2017,57(1):79- 83.Ma Yong,Bao Changchun.Overlapping speech detection using high-level information features[J].J Tsinghua Univ:Sci & Technol,2017,57(1):79- 83.(in Chinese)
[14] 徐淑正,孙忆南,皇甫丽英,等.基于MFCC和时频图等多种特征的综合鸟声识别分类器设计[J].实验室研究与探索,2018,37(9):81- 86,91.Xu Shuzheng,Sun Yinan,Huangfu Liying,et al.Design of Synthesized Bird Sounds Classifier Based on Multi Feature Extraction Classifiers and Time-frequency Chat[J].Research and Exploration in Laboratory,2018,37(9):81- 86,91.(in Chinese)
[15] Dixit A,Vidwans A,Sharma P.Improved MFCC and LPC algorithm for bundelkhandi isolated digit speech recognition[C]//2016 International Conference on Electrical,Electronics,and Optimization Techniques(ICEEOT).IEEE,2016.
[16] Jancovic,Liu,Zou.The Effectiveness of ICA-based Representation:Application to Speech Feature Extraction for Noise Robust Speaker Recognition[C]//European Signal Processing Conference.IEEE,2006.
[17] 张健,徐杰,包秀国,等.应用于语种识别的加权音素对数似然比特征[J].清华大学学报:自然科学版,2017(10):1038-1041.Zhang Jian,Xu Jie,Bao Xiuguo,et al.Weighted phone log-likelihood ratio feature for spoken language recognition[J].J Tsinghua Univ:Sci & Technol,2017(10):1038-1041.(in Chinese)
[18] 林秋华,郑永瑞,殷福亮.基于参考独立分量分析的语音增强方法[J].大连理工大学学报,2006,46(6):915-919.Lin Qiuhua,Zheng Yongrui,Yin Fuliang.Speech enhancement method based on ICA-R[J].Journal of Dalian University of Technology,2006,46(6):915-919.(in Chinese)
[19] Zhang G,Harichandran R S,Ramuhalli P.Application of Noise Cancelling and Damage Detection Algorithms in NDE of Concrete Bridge Decks Using Impact Signals[J].Journal of Nondestructive Evaluation,2011,30(4):259-272.
[20] 胡炼.基于NPC和改进的MFCC鲁棒语音特征提取研究[D].湖南:湖南大学,2011.Hu Lian.The Research of Robust Speech Features Extraction Based on NPC and Improved MFCC[D].Hunan:Hunan University,2011.(in Chinese)
[2] 成谢锋,李允怡,高珮熙,等.一种心音信号的源成分获取方法[J].南京邮电大学学报:自然科学版,2018,38(1):54-59.Cheng Xiefeng,Li Yunyi,Gao Peixi,et al.New method for obtaining components of heart sound signal source[J].Journal of Nanjing University of Posts and Telecommun-ications:Natural Science Edition,2018,38(1):54-59.(in Chinese)
[3] 巩燕,胡杰,高彬,等.心血管疾病即时检测技术的研究进展[J].中国科学:技术科学,2016,46(11):1116-1134.Gong Y,Hu J,Gao B,et al.Advances in monitoring of cardiovascular diseases at the point of care[J].Sci Sin Tech,2016,46(11):1116-1134.(in Chinese)
[4] Berkaya S K,Uysal A K,Gunal E S,et al.A survey on ECG analysis[J].Biomedical Signal Processing & Control,2018,43:216-235.
[5] 侯雷静,应冬文,国雁萌,等.基于心音分类的识别器分析[C]//中国声学学会2017年全国声学学术会议.中国声学学会,2017.Hou Leijing,Ying Dongwen,Guo Yanmeng,et al.Identifier analysis based on heart sound classification[C]//2017 National Acoustic Conference of the Chinese Acoustic Society.Chinese Acoustic Society,2017.(in Chinese)
[6] Puneet Kumar J,Anil Kumar T.Heart monitoring systems-a review[J].Computers in Biology & Medicine,2014,54(C):1-13.
[7] Rubin J,Rui A,Ganguli A,et al.Classifying heart sound recordings using deep convolutional neural networks and Mel:Frequency Cepstral Coefficients[C]//2016 Computing in Cardiology Conference.IEEE,2016.
[8] Zhang L,Tang R,Jiang J,et al.Research on pretreatment and envelope extraction algorithm of heart sound signal[J].Journal of Biomedical Engineering,2014,31(4):734-741.
[9] Lin Y,ChongYang W,ZhiCheng W.Research on the selection of wavelet bases for wavelet-based signal trend elimination[C]//International Conference on Wavelet Active Media Technology & Information Processing.IEEE,2013.
[10] Jain P K,Tiwari A K.An adaptive thresholding method for the wavelet based denoising of phonocardiogram signal[J].Biomedical Signal Processing & Control,2017,38:388-399.
[11] Whitaker B M,Suresha P B,Liu C,et al.Combining sparse coding and time-domain features for heart sound classification[J].Physiological Measurement,2017.
[12] Salman A H,Ahmadi N,Mengko R,et al.Automatic Segmentation and Detection of Heart Sound Components S1,S2,S3 and S4[C]//International Conference on Instrumentation.IEEE,2016.
[13] 马勇,鲍长春.基于高层信息特征的重叠语音检测[J].清华大学学报:自然科学版,2017,57(1):79- 83.Ma Yong,Bao Changchun.Overlapping speech detection using high-level information features[J].J Tsinghua Univ:Sci & Technol,2017,57(1):79- 83.(in Chinese)
[14] 徐淑正,孙忆南,皇甫丽英,等.基于MFCC和时频图等多种特征的综合鸟声识别分类器设计[J].实验室研究与探索,2018,37(9):81- 86,91.Xu Shuzheng,Sun Yinan,Huangfu Liying,et al.Design of Synthesized Bird Sounds Classifier Based on Multi Feature Extraction Classifiers and Time-frequency Chat[J].Research and Exploration in Laboratory,2018,37(9):81- 86,91.(in Chinese)
[15] Dixit A,Vidwans A,Sharma P.Improved MFCC and LPC algorithm for bundelkhandi isolated digit speech recognition[C]//2016 International Conference on Electrical,Electronics,and Optimization Techniques(ICEEOT).IEEE,2016.
[16] Jancovic,Liu,Zou.The Effectiveness of ICA-based Representation:Application to Speech Feature Extraction for Noise Robust Speaker Recognition[C]//European Signal Processing Conference.IEEE,2006.
[17] 张健,徐杰,包秀国,等.应用于语种识别的加权音素对数似然比特征[J].清华大学学报:自然科学版,2017(10):1038-1041.Zhang Jian,Xu Jie,Bao Xiuguo,et al.Weighted phone log-likelihood ratio feature for spoken language recognition[J].J Tsinghua Univ:Sci & Technol,2017(10):1038-1041.(in Chinese)
[18] 林秋华,郑永瑞,殷福亮.基于参考独立分量分析的语音增强方法[J].大连理工大学学报,2006,46(6):915-919.Lin Qiuhua,Zheng Yongrui,Yin Fuliang.Speech enhancement method based on ICA-R[J].Journal of Dalian University of Technology,2006,46(6):915-919.(in Chinese)
[19] Zhang G,Harichandran R S,Ramuhalli P.Application of Noise Cancelling and Damage Detection Algorithms in NDE of Concrete Bridge Decks Using Impact Signals[J].Journal of Nondestructive Evaluation,2011,30(4):259-272.
[20] 胡炼.基于NPC和改进的MFCC鲁棒语音特征提取研究[D].湖南:湖南大学,2011.Hu Lian.The Research of Robust Speech Features Extraction Based on NPC and Improved MFCC[D].Hunan:Hunan University,2011.(in Chinese)