心电信号的检测与模式分类方法的研究

发布时间：2018-08-05 18:19

【摘要】：心电信号是一种低频、微弱的生物电信号,它客观地反映了心脏的工作状态,其蕴涵着心脏的生理、病理信息对心脏疾病的诊断具有重要的参考价值。由于心电信号的幅值较小,频率较低,对信号进行检测时易受外界环境的干扰。有些干扰信号频率高,幅值大,往往会掩盖正常的心电信号,使心电波形无法识别。此外,心脏病患者的心电波形因病情而异,只有通过对心电信号的特征波形加以检测和分析,才可以诊断相应的心脏疾病。目前,心率失常疾病的诊断主要依靠医生的心脏医学知识和临床工作经验,由于心电数据量较大且异常波形并不连续出现,如果从事大量心电波形识别工作,易产生疲劳从而造成错判和误判而耽误患者的病情。因此,如何滤除心电信号中的各种干扰,对心电信号的特征信息加以提取及对各种不同的心电数据进行分类是心电医学界研究的重点。本文主要从以下四个方面进行研究： (1)针对心电信号的产生机理和特点,通过设计前置放大电路,右腿驱动电路对心电信号加以采集。针对采集过程中的噪声,设计相应的滤波器组,并对滤波后心电信号的进行放大。通过A/D转换电路、按键电路、串口通信电路、LCD显示电路和数据存储电路对心电信号进行转换存储、显示并与PC机进行通信。 (2)对于硬件电路中所不能滤除的噪声,对其特性进行分析,并设计了小波阈值去噪数字滤波器、固定步长LMS自适应去噪数字滤波器、可变步长LMS自适应去噪数字滤波器及RLS自适应去噪数字滤波器对干扰信号再次滤除。通过在MIT-BIH心率失常数据库中第101号心电数据加入三种干扰信号基线漂移,肌电干扰,工频干扰进行仿真实验,再对四种滤波器从去噪后图形和去噪性能参数对比可知,RLS自适应去噪数字滤波器的滤波效果明显优于其他三种滤波器。 (3)为了方便心电信号特征信息的提取,提出了一种基于二次样条母小波函数的心电信号QRS复合波检测算法。采用二次样条小波函数对心电信号作4尺度分解,分别获取各个尺度下的小波系数,在尺度3下,通过一定的阈值搜索小波系数模极大值对之间的过零点,确定R波位置。通过调整阈值以删除误检点,补偿漏检点,从而提高对R波的检测率。再从尺度1上R波过零点前后寻找局部模极大值对,分别确定QRS复合波的Q波,S波及QRS复合波起始位置和终止位置。通过MIT-BIH心率失常数据库中的心电数据对本文算法进行验证,并与其他QRS复合波检测算法相对比。结果表明本文算法对QRS复合波检测具有较高的准确率。 (4)设计了多种分类器对不同类别的心电信号加以分类。由于心电样本数据过于冗多,因此采用主成分分析法(PCA),线性判别法(LDA)以及主成分分析与线性判别融合法(PCA-LDA)对数据进行降维。实验证明线性判别法降维效果明显优于其他两种方法。接着设计了支持向量机(SVM),最小二乘支持向量机(LS-SVM),极限学习机(ELM)三种分类器,并分别以交叉验证法、遗传算法(GA)及粒子群算法(PSO)对支持向量机,最小二乘支持向量机的控制参数进行优化。最后通过实例对三种分类器的性能进行评价,结果表明：支持向量机分类精度最高,而极限学习机训练和测试时间最短。
[Abstract]:ECG signal is a low frequency, weak bioelectrical signal, which objectively reflects the working state of the heart. It contains the physiology of the heart. The pathological information has important reference value for the diagnosis of heart disease. Because the amplitude of the ECG signal is small and the frequency is low, it is easy to be disturbed by the external environment when the signal is detected. Some interference is disturbed. The signal frequency is high and the amplitude is large. It often covers the normal ECG signal and makes the heart wave shape unable to identify. In addition, the heart wave shape of heart disease patients varies according to the condition. Only through the detection and analysis of the characteristic waveform of the ECG signal, the corresponding heart disease can be diagnosed. At present, the diagnosis of the arrhythmia disease is mainly depended on the doctor. The heart medical knowledge and clinical experience, due to the large amount of ECG data and abnormal waveform is not continuous, if a large number of ECG waveform identification work, easy to cause fatigue and misjudge and delay the patient's condition. Therefore, how to filter all kinds of interference in the ECG signal and add the characteristic information to the ECG signal To extract and classify various ECG data is the focus of ECG medical research. This paper mainly studies from the following four aspects:
(1) in view of the mechanism and characteristics of the generation of ECG signal, the right leg drive circuit is designed to collect the ECG signal by designing the preamplifier circuit, and the corresponding filter bank is designed for the noise in the acquisition process, and the ECG signal after the filter is amplified. The A/D circuit, the key circuit, the serial communication circuit, the LCD display circuit and the display circuit are used. The data storage circuit converts and stores ECG signals, displays and communicates with PC.
(2) for the noise that can not be filtered in the hardware circuit, the characteristics are analyzed, and the wavelet threshold denoising digital filter is designed, the LMS adaptive de-noising digital filter is fixed, the variable step length LMS adaptive denoising digital filter and the RLS adaptive denoising digital filter are used to filter the interference signal again. The heart rate is passed in the heart rate. The number 101st ECG data in the abnormal database is added to the baseline drift of three kinds of interference signals, EMI and frequency interference, and then the comparison of the figure and the denoising performance parameters of the four filters from the denoising shows that the filtering effect of the RLS adaptive denoising digital filter is obviously better than the other three kinds of filter.
(3) in order to facilitate the extraction of the feature information of the ECG signal, a QRS composite wave detection algorithm based on the two spline mother wavelet function is proposed. The wavelet coefficients of each scale are obtained by using the two spline wavelet function to decompose ECG signals in 4 scales, and the wavelet coefficients are searched by a certain threshold in the scale 3. The R wave position is determined by the zero crossing point between the maximum value. The detection rate of the R wave is improved by adjusting the threshold to delete the false detection point and compensating the leakage point. Then the local modulus maximum value is found before and after the R wave over zero on the scale 1, and the Q wave of the QRS complex wave, the S wave and the starting position and the termination position of the QRS compound wave are respectively determined. The heart rate is lost through MIT-BIH. The ECG data in the normal database are verified by this algorithm and compared with other QRS complex wave detection algorithms. The results show that the proposed algorithm has a high accuracy for the QRS composite wave detection.
(4) a variety of classifiers are designed to classify different types of ECG signals. Because the data of ECG samples are too redundant, the main component analysis (PCA), linear discriminant (LDA), principal component analysis and linear discriminant fusion (PCA-LDA) are used to reduce the data. The experimental results show that the linear discriminant method is better than the other two. Then we design three classifiers for support vector machine (SVM), least squares support vector machine (LS-SVM) and limit learning machine (ELM), and optimize the control parameters of support vector machine and least squares support machine by cross validation, genetic algorithm (GA) and particle swarm optimization (PSO). Finally, three classifiers are selected by an example. Performance evaluation shows that the support vector machine classification accuracy is the highest, while the extreme learning machine training and testing time is the shortest.
【学位授予单位】：浙江师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.23

【参考文献】