Research On Denoising Model Based On Deep Learning And Fractional Diffusion Equation

In recent years,many excellent models have emerged in the field of image denoising,including the denoising neural network model due to the rapid development of deep learning technology,and the fractional diffusion equation denoising model due to the successful application of the traditional fractional differential equation in the modeling of various n atural phenomena.At present,the denoising effect of all kinds of partial differential equation models can 't exceed the deep learning method.At the same time,the problem of "artifact" in the process of denoising based on the deep learning method has not been solved as the partial differential equation model.Therefore,this dissertation attempts to fuse the deep learning model and the fractional equation model to get a series of better denoising models,the main research contents are as follows:In order to solve the problem of Gaussian noise removal,the deep learning method has been developed perfect enough.In this dissertation,the fractional derivative of clean image is learned by using DnCNN network,which is used as the prior of the model.Then,according to the specific denoising tasks,such as removing the noise of texture images,the fractional diffusion equation is constructed by using the property of fractional derivative.Finally,the prior is introduced into the coefficients of fractional diff usion equation to get the Gaussian noise removal model based on depth learning and fractional diffusion equation.Aiming at the problem of removing gamma noise,compared with deep learning method,the denoising result of partial differential equation model is more in line with people's expectations.The problem of deep learning method is that in the case of large image noise level,there will be a very serious over smooth phenomenon,leading to image distortion.In this dissertation,we first construct a denoising network suitable for multiplicative noise,and then construct a fractional diffusion equation suitable for multiplicative noise removal by using the properties of fractional derivative.Finally,we introduce the results of the denoising network into the gray detection operator or edge detection operator of fractional diffusion equation as a priori.Now,the multiplicative noise removal model based on deep learning and fractional diffusion equation is obtained.In the numerical experiment,in order to improve the speed of the fractional diffusion equation model,this dissertation attempts to use the fast explicit algorithm,which has periodic stability,and the step size in a period can 't meet the stability conditions.In addition,the experimental pa rt compares the algorithm with the traditional diffusion model,TNRD model and DnCNN model,and verifies the superiority and stability of the model.