Research On SAR Image Denoising Algorithm Based On Convolutional Neural Network

Synthetic Aperture Radar(SAR)is commonly carried in environmental monitoring,military and surveying and other areas.It has the advantages of high penetration,high resolution,all-weather and all-weather.The SAR imaging system must have amplitude information and phase information to improve its resolution,and the SAR image is formed by superimposing the coherent signals of the scattered echoes of the ground objects,which is easily polluted by noise.With the innovation of science and technology and the rapid development of computer applications,there are stricter requirements for InSAR technology and denoising technology.Many scholars have proposed various SAR image denoising principles,such as Lee filtering method,Frost filtering method,and SAR image denoising method based on wavelet transform.But so far there is still no consistent denoising method to remove the effects of noise.In recent years,with the rise of computer technology,various types of neural network models have gradually emerged and each has mature areas of application.In this paper,based on several mainstream SAR image filtering and denoising methods,the convolutional neural network algorithm is used to study the SAR image denoising effect.The main research contents are as follows:1.Introduce the principle of InSAR,and elaborate the calculation method of the height difference of the ground by calculating the phase difference that can be formed by the interference of the remote sensing image.Taking the actual mining area as an example,the SARscape module of ENVI software is used to interpret and generate the deformation map of the mining area in various periods,which provides the data basis for image denoising in this paper.2.First,the image is tested for the selection of the filter window size for the Lee filter,and the filter window sizes of the 3*3,5*5,7*7,9*9,and 11*11 filter windows are respectively tested,and each filter is obtained 5*5 The window size filter has the best effect.The Lee filter,Kuan filter,Frost filter,Gamma Map filter and convolutional neural network were used to analyze the denoising accuracy of the image.From the results,it can be seen that the Lee filter and Frost filter have the best denoising effect.Secondly,it briefly analyzes the development history and algorithm principle of convolutional neural network,and analyzes and implements the training layer and prediction layer functions in the CNN algorithm.Compared with the previous two filtering methods,the denoising effect is compared,and it is concluded that the CNN algorithm model has a prominent effect on image denoising.3.Integrate ArcGIS technology to analyze the change of regional surface settlement value of SAR image data of a working face of Huaibei Mining Group,and compare and analyze the deformation data of remote sensing image and level observation data before and after the filtering process.Evaluate the denoising ability of each filter.The root mean square error of CNN filtering is all smaller than the other filtering algorithms and SAR image extraction pixel values selected in the paper.It shows that the fit of CNN algorithm model and level value is higher.It is found from the test results that the CNN algorithm has a good denoising and filtering effect in the image of the shape variable of the mining area in the area where the surface shape variable of the mining area changes greatly.Figure[28]Table[7]Reference[73]...