Research On Wavelet Neural Network Denoising Based On Sampling Principle

In the era of digital information,image is an important carrier of information transmission and acquisition.It is widely used in industrial production and daily life.However,images are often subject to noise interference,resulting in poor image quality.In this paper,a wavelet neural network denoising algorithm based on sampling principle is proposed.On one hand,it takes advantage of the good time-frequency characteristics of the wavelet transform and overcomes the shortcomings of the traditional Fourier transform.On the other hand,the strong self-learning and self-adaptive ability of the neural network is also played,so the wavelet neural network has a strong ability.The ability of approximation and fault to lerance.This paper focuses on the more common noises in two kinds of images: stationary noise(such as the most common Gauss white noise)and non-stationary noise.The wavelet transform denoising technology,median filter technology and wavelet neural network denoising technology are discussed and studied,and the experimental verification on MATLAB is carried out.The main work of this paper can be summarized as the following parts:Firstly,we study the wavelet transform theory.Based on the good time frequency characteristics of the wavelet function and the scale function,we can apply the wavelet transform to the neural network and construct the scale function of the six order spline wavelet,which is used to replace the traditional neural network excitation function.Secondly,the principle of median filtering and wavelet denoising is introduced,and the reasons for these two traditional filtering algorithms to deal with the deficiency of the noisy image and the two algorithms that will destroy the image edge information are analyzed.The simulation experiments of two traditional filtering methods are carried out,and the advantages and disadvantages of the two methods are summarized and compared with the new wavelet neural network denoising algorithm proposed in this paper.Thirdly,we first select the appropriate neural network model,then we propose a new algorithm of wavelet neural network denoising based on sampling principle,and theoretically analyze it in one dimension,determine the parameters of the wavelet neural network,and prove that the training wavelet neural network is equivalent to the signal wavelet.A training method of wavelet neural network is proposed.By putting forward the mathematical model of image,we further study the feasibility of wavelet neural network for image denoising,prove the arbitrariness of sampling period selection,and determine the weight value of the input layer,the number of hidden layer nodes and the weight of the output layer based on the image model.At last,we test and simulate the images with stationary noise(such as Gauss white noise)and non-stationary noise on MATLAB and compare the other denoising methods.Finally,we draw a conclusion that the new wavelet neural network denoising algorithm proposed in this paper can filter the noise and protect the image details compared to the traditional median filter.Wave and wavelet denoising have great advantages.