Research On Filtering Algorithm Of Salt And Pepper Noise In Remote Sensing Images

Most of human’s access to outside information comes from images,and images,as information carriers,will be affected by various noises during acquisition and transmission,which will affect the further use of images.Therefore,it is very necessary to denoise the image.By researching the traditional denoising algorithm and the denoising algorithm based on deep learning,this paper aims at the problem that the existing image denoising algorithm improves the denoising performance and causes the loss of image edge information.filtering(AMF)algorithm proposes two improved algorithms with different focuses and designs a convolutional neural network structure suitable for denoising salt and pepper noise,and designs simulation experiments to verify that the three algorithm models are effective in natural images and The denoising performance of image denoising operation and edge preservation effect.The main work of this article is as follows:1)Based on the AMF algorithm,this paper proposes an edge-preserving adaptive median filter algorithm(ERAMF).The ERAMF algorithm designed an edge extractor to extract the edge pixels of the picture and generate an edge picture.At the same time,the AMF algorithm was used to denoise the picture.At the end,the denoised picture of the AMF algorithm was merged with the edge picture to obtain a clean image.This approach further improves the algorithm’s retention of edge information.The simulation results show that the ERAMF algorithm is superior to the AMF algorithm in denoising and edge-preserving effects on natural and remote sensing images,and the ERAMF algorithm is more improved than the AMF algorithm in edge-preserving effects.2)Based on the AMF algorithm,this paper proposes a threshold cutoff adaptive median filtering algorithm(TAMF).On the basis of the AMF algorithm,the TAMF algorithm proposes to use the visual phenomenon that the brightness of the noise pixels of salt and pepper noise differs greatly from the pixels of the local background,and constructs a threshold operator to distinguish the noisy pixels.The operation is that for each row and column in the window,the median value is taken,and all the median values are combined into a new vector,then the maximum and minimum values in the vector are used as the vector interval,and finally the pixel value of the center of the window is determined Whether it is within the interval to determine whether the pixel is a noise pixel.This judgment method improves the algorithm’s denoising performance for high-density salt and pepper noise.In the simulation experiment,the TAMF algorithm is used to denoise natural and remotely sensed images contaminated with high-density salt and pepper noise.Its denoising performance is improved compared with the AMF algorithm,and its edge retention effect is better than the AMF algorithm.3)In order to better protect the edge detail information in the remote sensing image,this paper uses the VGG model as the basic architecture to construct the main network framework,adding a hollow parallel convolution feature extraction module to assist the main network to extract information and remove all pooling Layer,replace the convolution kernel with a hollow convolution kernel to play the role of the pooling layer,and finally design two convolutional layers for information fusion,and replace the fully connected layer with a convolutional layer to construct a fully convolutional neural network as The basic improved network and named it Empty Parallel Convolutional Neural Network(EPCNN).This design scheme makes the model suitable for denoising salt and pepper noise,and reduces the loss of edge information in the denoising operation.Through the simulation experiment of natural images and remote sensing images,it can be found that EPCNN can effectively remove the pepper and salt noise,and has a good effect of maintaining the edge information.The denoising performance and edge retention effect of EPCNN is better than Dn CNN.