The Research Of Ecg Signal Preprocessing And Feature Extraction Algorithms

Since ecg signal was found, it has been one of the important basis for the clinical diagnosis of heart disease. ECG signal is a typical low frequency weak biological electrical signal under strong noise background, with low SNR, non-stationary, random, and the noise characteristics of strong time and frequency coupling. With modern medicine for automatic analysis of ECG increasingly urgent need, the computer processing of the ECG signal analysis processing has become increasingly popular in research. Automated processing ECG can be divided into three areas, namely the pre-treatment ECG, ECG feature extraction and classification of ECG. The three research content interlocking, preprocessing is the base of feature extraction, and feature extraction accuracy can play a decisive role for classification and recognition. In this paper, the main work lies in the ECG signal preprocessing and feature extraction of ECG signal, for the two aspects of the work is divided into three parts:First study the fractional Fourier transform of the optimal order times problem determination. Fractional Fourier transform variable order for analysis of ecg signals have a vital role. In this paper, use the invasive weed optimization algorithm in fractional Fourier transform to carry out the order optimization. Use invasive weed optimization to search peak in fractional Fourier domain, then determine the peak to find the optimal order. Simulation results show that in speed and precision, invasive weed optimization algorithms have achieved good results for the pre-treatment ECG foundation.Then according to the characteristics of the ECG signal noise, respectively improved filtering algorithm and preprocess ECG signal. Baseline drift noise frequency is relatively low, and there is no overlap with ECG waveform frequency, the trend is relatively slow, so use the EMD to filter noise in ECG signal. Although power frequency interference noise frequency overlap with ECG signal frequency, but its frequency is fixed, so using the algorithm of the LMS adaptive notch filter to process the signal. In order to overcome the delay problem of the algorithm, in this paper, extend to the front of the original signal. Simulation experiment proves that the improved algorithm can effectively alleviate the oscillation of front-end signal waveform. Myoelectricity interference noise’s frequency distribution is widespread, its frequency has strong aliasing with characteristics of ECG signal waveform, so denoise ECG based fractional Fourier transform combination with spectral subtraction. Under the optimal order, ECG signals through the fractional Fourier transform, there will be a significant amplitude gathered themselves together, and electrical noise does not have such feature. Simulation experiment show that, compared with pure spectral subtraction algorithm, based on fractional Fourier transform combination with spectrum subtraction algorithm improves the signal-to-noise ratio.Finally integrated signal of time domain and frequency domain information, use HHT transform combined with WT transformation to extract the feature points in the QRS complex. Use the EMD decomposition to get the scope of QRS complex. By the Hilbert transformation makes the position of QRS complex features has been enhanced. On this basis use wavelet transform to signals. Here, the generating function is gaussian first derivative.Combined with time domain and frequency domain information to locate the peak value of QRS complex. Simulation results show that, The method in the QRS wave peaks location of ECG signal achieved good effect.