Research On ECG Feature Extraction Algorithm Based On Wavelet Theory

The ECG used in the clinical medicine contained many important features like the Heart Rate,waveform trend and wave group cycle.These underlying informations can assist doctors to adjust therapeutic schedule appropriately in the process of operation scheme.There is a special kind of signal in ECG: FECG,because there are MECG and noise interference,the extraction of clear FECG became a difficult point;the corresponding feature extraction such as FHR also increased the difficulty.In this thesis,an improved neural network algorithm for FECG extraction from MECG was proposed,and the improved feature extraction algorithm based on wavelet decomposition was used for R wave detection and FHR calculation of FECG.The main research content was divided into the following parts:Firstly,aiming at the particularity and difference of ECG contains noise,in this thesis;the de-noising pretreatment was carried out.The factory frequency noise,baseline drift and electrical noise adopted different de-noising methods: comb filter restrain power frequency noise,median filter inhibition of baseline drift and low pass filter to remove electrical noise.The experiment proved that separate the disturbance signal processing can get better de-noising effect.Secondly,aiming at the shortcomings of the traditional BPNN algorithm with the gradient descent method such as long calculation time and easy falling into the local minimum,this thesis proposes to combine the LM algorithm with BPNN algorithm to extract FECG from MECG.The improved algorithm was applied to the simulation experiment and clinical experiment,and the results showed that the mean square error and signal-to-noise ratio were obviously better than the traditional BPNN algorithm.Finally,an improved R wave detection algorithm for wavelet analysis adaptive threshold is proposed.R wave in the work of extraction to the FECG mistakenly identified and residual due to the error of FHR calculation.This thesis proposes the adaptive threshold detection algorithm based on wavelet analysis modulus maximum to detect R wave,and apply it in the simulated data,the results prove that the method of R wave positioning accuracy is above 99%,the instantaneous heart rate error is around 0.37%,showing good detection characteristic.The FECG feature extraction algorithm proposed in this thesis has good applicability in both analog and clinical signals,which makes further research on the neural network in medical signal processing and lays a good foundation for the feature extraction of the remaining FECG waves.