Research On Heart Beat Classification And Recognition Based On RBF Neural Network

ECG signal is the direct reaction of the electrical activity of the heart. The analysis and diagnosis of the electrocardiogram play an important role in the clinical diagnosis, so it is a research hotspot for the scholars. Automatic analysis of ECG signals including preprocessing, feature extraction and classification and there are various algorithms for each part. This paper chooses Wavelet Transform and Radial-Basis Function neural network to analyze and diagnose the heart beats.First, we give a brief introduction about the development history, research status and difficulties of the ECG automatic analysis system, then we introduce the basic knowledge of electrocardiogram and the performance of several common abnormal rhythm. Because ECG signal is a kind of weak signal, it is easy to be disturbed by the noises, such as baseline wander, power frequency interference and muscle power interference and so on. According to the characteristics of these noises, we choose the suitable filters to restrain them. In feature extraction, we make use of the annotation time of the heart beats to locate the peak of R wave, thus we can extract the QRS waves, calculate RR intervals and the difference between RR intervals and the average RR interval. At last, we use Biorthogonal Wavelet Transform to complete the feature parameters extraction and data dimension reduction.The automatic classification of heart beats is the difficulty of the ECG signal diagnosis system, this paper adopts RBF neural network to realize the classification and recognition of heart beats. The signals in MIT-BIH were separated into two groups, the first24signals were used as the training samples and the rest24signals as identification samples. The experimental results show that several common beats' recognition rate above85%and premature ventricular contraction is achieved99%, which shows that the neural network has a good performance in the classification of the heart beats.