基于分类器组合的心电信号身份识别算法研究

发布时间：2019-03-14 17:38

【摘要】：目前用于身份识别的方法有很多,常用的有人脸识别,指纹识别等等。随着现在技术的不断更新,人脸可以被照片所代替,指纹也可以复制,但每个人的心电图信号是独一无二的,不可复制的,目前心电图信号主要用于临床医学心脏方面疾病的诊断,近些年来,提出基于心电信号进行身份识别的研究的学者不断增多,其目的就是为了实现能更好更精确的对人类进行身份识别,生物识别技术迅速发展的原因正是因为人类自身的特别的行为特征或者生理进行的身份鉴别,其可靠性和不可替代性非常高。本文研究的是基于分类器组合的心电信号身份识别算法,使用无基准点特征提取的方法来提取QRS波形。HOAC-DCT特征提取、DWT特征提取和PCA特征提取以及分类器组合算法相结合的方法,提出的方法可以对身份识别的准确率进行提高。首先,本文预处理ECG信号,原始的心电信号中常常伴随着肌频干扰、工频干扰、基线漂移等噪声成分,ECG信号受采集设备的等因素的影响,必须对ECG信号进行滤波,本文使用四阶巴特沃斯带通滤波器。再使用无基准点特征提取的方法来提取心电信号的QRS波形,即HOAC算法来提取ECG信号正规化的QRS波形,去除基准点的影响。其次,对提取的正规化QRS波形再次使用HOAC、DWT和PCA提取用于身份识别的特征,由于HOAC的特征维数高,通过DCT算法降低了特征维度。每种方法分别提取一个特征,找出三个特征以后,使用最近邻分类器分类识别,最后使用乘法、最大、最小、中值、大多数投票等规则对分类结果进行组合,找出对ECG信号识别最好的组合算法,实现对ECG信号的准确率更高的身份识别。最后针对PTB和MIT-BIH数据库验证所提出算法的性能,并利用MATLAB得出身份识别准确率结果,验证所提出算法的优越性。实验结果表明,在组合规则中,乘法和中值规则组合的分类器的分类能力最好,比使用单个特征提取的分类能力更强,分类错误率也最低,验证了本文提出的基于分类器组合的ECG信号身份识别算法研究分类效果更好,且实现起来简单,可以为基于ECG身份识别的系统提供良好的技术支撑。
[Abstract]:At present, there are many methods for identification, such as face recognition, fingerprint identification and so on. As the technology continues to update, faces can be replaced by photographs and fingerprints can be copied, but everyone's ECG signals are unique and cannot be copied. At present, ECG signals are mainly used in the diagnosis of cardiac diseases in clinical medicine. In recent years, more and more scholars have put forward the research of identification based on ECG signals. The purpose is to achieve better and more accurate identification of human beings. The reason for the rapid development of biometric technology is precisely because of human own special behavioral characteristics or physiological identity identification. Its reliability and irreplaceable are very high. In this paper, the ECG signal identification algorithm based on classifier combination is studied. The QRS waveform is extracted by the method of extracting the feature without reference point. The HOAC-DCT feature is extracted, which is based on the combination of classifiers. Combining DWT feature extraction with PCA feature extraction and classifier combination algorithm, the proposed method can improve the accuracy of identity recognition. Firstly, the ECG signal is preprocessed in this paper. The original ECG signal is often accompanied by the noise components such as muscle frequency interference, power frequency interference, baseline drift and so on. The ECG signal is affected by the acquisition equipment and other factors, so it is necessary to filter the ECG signal. In this paper, a fourth-order Butterworth bandpass filter is used. Then the QRS waveform of ECG signal is extracted by the feature extraction method without reference point, that is, the HOAC algorithm is used to extract the normalized QRS waveform of the ECG signal and remove the influence of the reference point. Secondly, the extracted normalized QRS waveform is extracted by HOAC,DWT and PCA again. Because the feature dimension of HOAC is high, the feature dimension is reduced by DCT algorithm. Each method extracts one feature separately, after finding out three features, it uses nearest neighbor classifier to classify and recognize, and finally uses multiplication, maximum, minimum, median, majority voting rules to combine the classification results. Find out the best combination algorithm to identify the ECG signal, and realize the more accurate identification of the ECG signal. Finally, the performance of the proposed algorithm is verified by PTB and MIT-BIH database, and the accuracy result of identification is obtained by using MATLAB to verify the superiority of the proposed algorithm. The experimental results show that the combination of multiplication and median rule has the best classification ability, which is stronger than that of single feature extraction, and the classification error rate is the lowest. The proposed ECG signal identification algorithm based on classifier combination is proved to be more effective and easy to implement. It can provide good technical support for the system based on ECG identification.
【学位授予单位】：延边大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】