The Study Of ECG Classification And Recognition Based On Wavelet Transform And Neural Network

As population aging in China is intensifying and the distribution of existing medical resource is not balance, medical and health problems become more and more important in our country. Cardiovascular disease is one of diseases which heavily harm modern people’s health, and arrhythmia disease is a major kind of cardiovascular disease. Researchs on classification and recognition of ECG are the precondition and the key of the diagnosis on arrhythmia. Study of ECG classification recognition in this paper is the study of digital automatic classification and recognition based on wavelet transform and neural network. The study is helpful to promote the development of mobile ECG monitoring system and then balances the distribution of medical resources.There are many kinds of ECG signal, the ECG signal classified and recognized in this paper include the normal ECG signal(N), the left bundle branch block signal(LBBB), the right bundle branch block signal(RBBB), the premature ventricular contraction signal(PVC), the atrial premature beat signal(APC). The main works in this paper are as follows:1. This paper uses the wavelet transform to filter out the noises which are produced by the process of acquiring ECG signals using the detection device. This paper uses forced filter method to remove the baseline wander(BW) noise and uses threshold filter method to filter out the power-line interference(PLI) noise and electromyogram(EMG) noise.2. This paper uses the filtered ECG signal to recognize R wave, Q wave, S wave, P wave and T wave in order. This paper analyses the shortcomings in the traditional recognization method, and improve the method to eliminate these shortcomings. For example, in the traditional R wave recognization method, the selection of threshold has a significant impact on R wave recognition effect, and the number of the maximum value does not match the number of the minimum value, and then the improved method in this paper can eliminate these problems.3. This paper uses waveform’s amplitudes and waveform interval times as the feature value of neural network’s training samples. Use the training samples to train the traditional BP neural network and radial basis function neural network, classify the ECG signal records using the trained neural network, and set the correct classification rate of the ECG records as the standard of the neural network. This paper improves the performance of neural network by modifying parameters of these neural networks, improves the classification accuracy rate by adjusting the structure of training samples. This paper contrasts the traditional BP neural network and radial basis function neural network, the traditional BP neural network has some defects and the radial basis function neural network can make up such disadvantages. For example, the traditional BP neural network has many parameters, low training speed and the training results have high randomness and other shortcomings, but the radial basis function neural network has fewer parameters, high training speed and the training results have certain regularity.This paper improves the method in waveform recognition and select radial basis function neural network as the neural network to classify the ECG signal. Finally, this paper gets high classification correct rate.