The Research Of Ecg Signal Detection And Pattern Classification Methods

ECG is a low frequency, weak biological signal. It objectively reflects the working state of the heart and implies physiological and pathological information of the heart. It has important reference value for the diagnosis of heart disease. Because of the small ECG amplitude and low frequency, signal detection progress is very easily disturbed by the external environment. Some of the interference signals’ frequency is so high and amplitude is so large that always cover up the normal ECG signals, which will lead ECG waveform to be unable to identify. In addition, ECG signal wave form of heart disease varies with patient’s condition. Heart disease can be diagnosed only through detecting and analyzing the characteristics of the wave form of ECG signal. Currently, the diagnoses of cardiac arrhythmias heart disease mainly rely on the doctor’s medical knowledge and clinical experience. Because ECG waveform anomalies are not consecutive and ECG data are very large, if a doctor does a lot of ECG waveform recognition, he is easy to produce fatigue, resulting in misjudgment and delaying the patient’s condition. Therefore, how to filter out all kinds of interferences in the ECG signal, extract the characteristic information of ECG signal and classify different kinds of ECG data is the focus in the medical research. The thesis researches the following four aspects:(1)In order to study the generation mechanism and characteristics of ECG signal, the thesis designs preamplifier circuit and right leg drive circuit to collect ECG signal. Aiming to collect the noises in the process of test, the thesis designs corresponding filter group and amplifies the ECG signal after being filtered. Then ECG signal is transformed, saved, displayed and communicated with the PC through the A/D convert circuit, key circuit, serial communication circuit, LCD display circuit and data storage circuit.(2)Because the hardware circuit can not filter out the noise, so we analyze the noise characteristics.Through the wavelet threshold de-noising digital filter, a fixed step LMS adaptive de-noising digital filter, variable step size LMS adaptive de-noising digital filter and RLS adaptive de-noising digital filter, we filter out the interference signal again. In order to simulate the interference signal, we joined baseline drift, EMG interference, frequency interference in the MIT-BIH arrhythmia database No.101ECG data. Through the simulation experiment and the performance parameters, the RLS adaptive digital filter de-noising filter is better than the other three filters.(3)In order to extract ECG feature information comfortably, this thesis proposed a quadratic spline mother wavelet function based on ECG QRS complex detection algorithm. It used the Quadratic spline wavelet function to decompose the ECG signal into4scales. Then we obtain the wavelet coefficients from the first scale to the forth scale. Under the third scale, looking for zero crossings of the wavelet coefficients, and then determining the position of the R wave. In order to improve the R-wave detection rate, we adjusted the threshold to remove erroneous points and compensate undetected points. Under the first scale, looking for the local modulus maxima between R-wave zero crossings, and then determining determine the Q wave, S wave and the start point, end point of QRS complex wave. Verified by MIT-BIH arrhythmia database and compared with other QRS complex detection algorithm, in this thesis, the QRS complex detection algorithms with high accuracy.(4)By designing performance of different classifiers to classify different types of ECG. Because the ECG data samples are too many, so we use PCA, LDA, and PCA-LDA to reduce ECG data samples. The result shows that LDA dimensionality reduction is better than the other two methods. Then we introduce three classifiers: SVM, LS-SVM and ELM. Then optimize the parameters of SVM and LS-SVM through methods of cross-validation, GA, PSO. Finally, through example to evaluate the performance of three classifiers, it shows that SVM has the highest classification accuracy, the time of ELM training and testing is the shortest.