Research On Intelligent Signal Processing Restoration Based On Total Variation Regularization

ECG signal detection process is essentially a pattern recognition process, mainly related to feature extraction and it's design. Because of physiological changes of patients and the impact of external environmental interference, records often contain a variety of noise and signal interference, ECG signal waveforms are also changes. Therefore, the ECG automatic analysis is difficult. How to establish a mathematical model to describe the signal characteristics is of vital importance. There is no general theoretical method on present study.Based on nonlinear approximation ability of traditional neural networks, Wavelet neural networks(WNN) can be trained to approximate any nonlinear function through choosing proper wavelet basis to build neural network structure. In this paper, we present a detection algorithm based on WNN considering it's advantages. As the ECG signal is one-dimensional time-varying electrical signal, characterized by uncertainty, the traditional signal processing methods can't obtain satisfactory results. The use of WNN model for detecting ECG is more appropriate to identify the characteristics of ECG and it is useful in solving the complexities and uncertainty issues encountered in ECG signal recognition.In this paper, firstly we make a brief introduction to the theory of neural network and wavelet analysis, then give a detailed analysis of the WNN and design a WNN algorithm to detect the ECG signal. In order to verify the network capacity, we use MI-TBHI ECG data to process and found the processing speed can meet the requirements, but there are still areas for improvement on the network structure and learning algorithm, and the ability of detection to irregular ECG signal is not enough.Test results show that WNN with signal feature extraction on original signal by wavelet basis show the characteristic of local time-frequency and speed up the convergence of wavelet neural network. Also, Because the wavelet basis merger the invariance of the scale and transformation, making the generalization ability of neural networks have greatly improved, thereby reducing the number of training samples; meanwhile it has a good shielding effect to random noise, increasing non-stationary and nonlinear signal detection ability.