Bilateral filtering remote sensing image denoising method based on EMD distance

Abstract

The invention relates to a bilateral filtering remote sensing image denoising method based on an EMD distance, and the method comprises the steps: firstly, carrying out the clustering analysis of an image polluted by noise, extracting a key spectrum cluster and a center thereof, and generating a fuzzy membership matrix; thirdly, calculating the distance between spectral clustering centers, and constructing a spectral clustering difference matrix; for each pixel of the target image and a neighborhood thereof, calculating an Euclidean distance between the pixels, and substituting the Euclidean distance into a Gaussian function to generate a distance weight; then, extracting fuzzy membership degree vectors corresponding to the pixels from the fuzzy membership degree matrix, and calculating an EMD distance by combining the spectral clustering difference matrix as a spectral difference weight between the pixels; and finally, synthesizing the spectral difference weight and the distance weight, calculating an EMD bilateral filtering coefficient of each pixel, and carrying out EMD bilateral filtering on the image according to the EMD bilateral filtering coefficient, so as to effectively remove Gaussian noise and obtain a clear high-resolution satellite image.

Description

Technical Field

[0001] The present invention relates to the technical field of image filtering processing, and particularly to a method for denoising remote sensing images by bilateral filtering based on the EMD distance. Background Art

[0002] With the rapid development of technology, remote sensing technology has become an important tool for information acquisition, monitoring and analysis, and decision-making support in modern society. As an important product of remote sensing technology, remote sensing satellite images have shown broad application prospects in multiple fields with their characteristics of high resolution, large-scale coverage, and rapid update.

[0003] Remote sensing image noise, as unnecessary or redundant interference information existing in image data, seriously affects the quality of remote sensing images. Therefore, before the processing and application of remote sensing images, denoising processing must be carried out. Noise can be theoretically defined as "unpredictable random error that can only be identified by probability statistical methods". Therefore, it is appropriate to regard image noise as a multi-dimensional random process, and thus the method of describing noise can completely borrow the description of random processes, that is, describe noise with its probability distribution function and probability density distribution function. Gaussian noise, as a kind of noise widely existing in nature, often pollutes remote sensing images and causes interference to their processing and application. Therefore, denoising the remote sensing images contaminated by Gaussian noise is a very important step in remote sensing image processing and application.

[0004] Currently, the denoising algorithms for Gaussian noise mainly adopt isotropic filtering models including methods such as mean filtering and Gaussian filtering. Among them, the Gaussian filter is relatively the most stable. However, the Gaussian filter usually focuses on the distance weights of neighboring points and does not take into account the spectral differences between neighboring points and the central point, resulting in partial edge blurring in the filtered image and causing errors in straight-line positioning. The bilateral filter, as a filter that simultaneously considers distance weights and spectral difference weights, can better retain the edge information of remote sensing images while eliminating Gaussian noise, and has received extensive application and attention. However, if neighboring pixels or the central pixel are contaminated by relatively severe Gaussian noise, the bilateral filtering algorithm cannot guarantee the accuracy of its spectral difference weights, affecting the denoising effect, and thus cannot effectively filter out some Gaussian noise in the image. Therefore, how to ensure the accuracy of the spectral difference weight calculation of the bilateral filtering algorithm, further improve its robustness to Gaussian noise, and strengthen its denoising ability against Gaussian noise has become an urgent problem to be solved. Summary of the Invention

[0005] To solve the problems existing in the above-mentioned prior art, the present invention aims at high-resolution satellite images affected by Gaussian noise, decomposes the remote sensing image mixed pixels through fuzzy C-means clustering, captures the key spectral clusters, and calculates the spectral differences of neighboring pixels based on the Earth mover’s distance (EMD). Finally, based on the neighborhood spectral differences and neighborhood distances, the filter neighborhood weights are generated. The key spectral clusters captured by fuzzy C-means clustering reflect the overall spectral information of the image and are not easily affected by Gaussian noise. The EMD distance is an optimal distance based on the Simplex algorithm and has a certain robustness to noise. Therefore, the neighborhood spectral differences calculated by the cooperation of FCM and EMD distance are not easily affected by Gaussian noise, so that the filter coefficients calculated by the EMD bilateral filtering algorithm are relatively accurate and have strong denoising ability for Gaussian noise.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A bilateral filtering remote sensing image denoising method based on EMD distance, comprising:

[0008] Obtain a high-resolution satellite image contaminated by Gaussian noise, perform clustering analysis on the high-resolution satellite image, obtain the key spectral clusters in the image, extract the clustering centers of each key spectral cluster, generate the corresponding fuzzy membership matrix, and further calculate the distances between all spectral clustering centers to generate a spectral clustering difference matrix;

[0009] Obtain each pixel of the target image and its corresponding neighborhood, calculate the Euclidean distance between the pixel and the neighborhood pixels, bring the Euclidean distance center into the Gaussian function to generate a distance weight;

[0010] Taking the pixel as the central pixel, extract the first fuzzy membership vector and the second fuzzy membership vector corresponding to the pixel and the neighborhood pixels from the fuzzy membership matrix, and combine with the spectral clustering difference matrix to calculate the EMD distance between the first fuzzy membership vector and the second fuzzy membership vector, and use the EMD distance as the spectral difference between the pixel and the neighborhood pixels to calculate the spectral difference weight;

[0011] According to the spectral difference weight and the distance weight, calculate the EMD bilateral filtering coefficient of each pixel, and based on the EMD bilateral filtering coefficient of each pixel, perform EMD bilateral filtering on the high-resolution satellite image to obtain a high-resolution satellite image with Gaussian noise removed.

[0012] Optionally, the Gaussian function is:

[0013]

[0014] Among them, σ is the standard deviation of the Gaussian distribution, and x is the independent variable of the Gaussian distribution.

[0015] Optionally, generating the distance weight includes:

[0016]

[0017] Among them, σ s is the standard deviation of the distance weight Gaussian function, and ||(i,j)-(u,v)|| is the Euclidean distance between the pixel and the neighboring pixels in the neighborhood.

[0018] Optionally, calculating the spectral difference weight includes:

[0019] Obtaining the feature of the pixel and the feature of the neighboring pixels, obtaining a spectral clustering matching matrix according to the feature of the pixel and the feature of the neighboring pixels, and setting the objective and constraint conditions of the spectral clustering matching matrix; the objective is: minimizing the matching cost of transforming the feature of the pixel to the feature of the neighboring pixels;

[0020] Calculating the spectral difference weight according to the spectral clustering matching matrix.

[0021] Optionally, obtaining the feature of the pixel and the feature of the neighboring pixels includes:

[0022] f ij = {(c 1 , u 1 (p ij )),…,(c k , u k (p ij )),…,(c n , u n (p ij ))}

[0023] f uv = {(c 1 , u 1 (p uv )),…,(c k , u k (p uv )),…,(c n , u n (p uv ))}

[0024] Among them, f ij is the feature of the pixel p ij , f uv is the feature of the neighboring pixel p uv , c k is the signal center of the spectral cluster s k , uk (p ij ) and u k (p uv ) are the fuzzy membership degrees of pixel p ij and p uv with respect to spectral clustering s k . n is the number of spectral clusterings of the image

[0025] Optionally, setting the objective of the spectral clustering matching matrix includes:

[0026]

[0027] where EMD is the EMD distance function, f ij is the feature of pixel p ij , f uv is the feature of neighboring pixel p uv , DM is the spectral clustering difference matrix, w kl is the matching weight between spectral clustering centers c k and c l , d kl is the spectral difference between spectral clustering centers c k and c l . n is the number of spectral clusterings of the image

[0028] Optionally, setting the constraint conditions of the spectral clustering matching matrix includes:

[0029] d kl ≥0 1≤k≤n, 1≤l≤n

[0030]

[0031] where k and l are spectral clustering indices

[0032] Optionally, calculating the spectral difference weight includes:

[0033]

[0034] where EMD(f ij , f uv , DM) is the spectral difference between the pixel and the neighboring pixel, and σ emd is the standard deviation of the spectral difference weight Gaussian function

[0035] Optionally, calculating the EMD bilateral filtering coefficient for each pixel includes:

[0036]

[0037] where is the normalization coefficient, and σ sis the standard deviation of the distance-weighted Gaussian function, σ emd is the standard deviation of the spectral-difference-weighted Gaussian function is the distance-weighted Gaussian function, I(p ij ) is the spectral value of pixel p ij is the spectral-difference-weighted Gaussian function

[0038] The beneficial effects of the present invention are as follows:

[0039] The bilateral filtering algorithm based on the EMD distance provided by the present invention captures key spectral clusters in high-resolution remote sensing images by using the FCM algorithm. Based on the generated fuzzy membership matrix and spectral clustering difference matrix, the spectral difference weight is calculated using the EMD distance. Based on the spectral difference weight and distance weight, the EMD bilateral filtering coefficient is generated to filter the image contaminated by Gaussian noise and remove the Gaussian noise. The present invention uses FCM to capture key spectral clusters in high-resolution remote sensing images and reduces the interference of Gaussian noise on the calculation of spectral differences in the image; the present invention calculates the spectral difference using the EMD distance based on the fuzzy membership matrix and spectral clustering difference matrix. Since the EMD distance is calculated by the linear programming Simplex optimal algorithm, it has a certain robustness to the interference of Gaussian noise. In summary, due to the certain robustness of FCM and EMD to Gaussian noise, the synergistic effect of the two enables the EMD bilateral filtering algorithm to better remove Gaussian noise in high-resolution remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 is a flowchart of a bilateral filtering remote sensing image denoising method based on the EMD distance according to an embodiment of the present invention

[0042] Figure 2 is a schematic diagram of denoising for Image 1 according to an embodiment of the present invention; (a) is the original image of Image 1, (b) is the noisy image of Image 1, and (c) is the denoised image of Image 1

[0043] Figure 3 is a schematic diagram of denoising for Image 2 according to an embodiment of the present invention; (a) is the original image of Image 2, (b) is the noisy image of Image 2, and (c) is the denoised image of Image 2

[0044] Figure 4 ​Schematic diagram of triple denoising of an image according to an embodiment of the present invention; (a) is the original image of Image 3, (b) is the noisy image of Image 3, and (c) is the denoised image of Image 3;

[0045] Figure 5 Schematic diagram of quadruple denoising of an image according to an embodiment of the present invention; (a) is the original image of Image 4, (b) is the noisy image of Image 4, and (c) is the denoised image of Image 4;

[0046] Figure 6 Peak signal-to-noise ratio graph according to an embodiment of the present invention;

[0047] Figure 7 Graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image according to an embodiment of the present invention; wherein, (a) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v5; (b) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v10; (c) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v15; (d) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v20; (e) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v25; (f) is the graph showing the relationship between the model window size and the signal-to-noise ratio of the denoised image with variance v30. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0050] As Figure 1 shown, this embodiment discloses a bilateral filtering remote sensing image denoising method based on EMD distance, including: obtaining a high-resolution satellite image contaminated by Gaussian noise, performing clustering analysis on the high-resolution satellite image, obtaining key spectral clusters in the image, extracting the cluster centers of each key spectral cluster, and generating a corresponding fuzzy membership matrix, further calculating the distances between all spectral cluster centers to generate a spectral cluster difference matrix; obtaining each pixel p ij of the target image and the corresponding neighborhood N ij , calculating the pixel p ij and the neighborhood pixel p uvThe Euclidean distance is calculated. The center of the Euclidean distance is substituted into the Gaussian function to generate the distance weight. For pixel p ij as the central pixel, the fuzzy membership degree vector f ij corresponding to pixel p uv and the fuzzy membership degree vector f ij are extracted from the fuzzy membership degree matrix. Combining with the spectral clustering difference matrix, the EMD distance between the fuzzy membership degree vector f uv and the fuzzy membership degree vector f ij is calculated. The EMD distance is used as the spectral difference between pixel p uv and pixel p ij and pixel p uv . The spectral difference weight is calculated. According to the spectral difference weight and the distance weight, the EMD bilateral filtering coefficient of each pixel is calculated. Based on the EMD bilateral filtering coefficient of each pixel, the high-resolution satellite image is subjected to EMD bilateral filtering to obtain a high-resolution satellite image with Gaussian noise removed.

[0051] Specifically: The technical solution provided by the present invention is: A method for denoising remote sensing images based on EMD distance, including the following steps:

[0052] Step 1: Obtain a high-resolution satellite image contaminated by Gaussian noise.

[0053] Step 2: Use the fuzzy C-means clustering algorithm to cluster the high-resolution remote sensing image, capture the key spectral clusters in the image, generate the corresponding fuzzy membership degree matrix, and save the clustering center of each spectral cluster.

[0054] Step 3: Calculate the distances between all spectral clustering centers to generate DM.

[0055] Further, the Gaussian function is:

[0056]

[0057] where σ is the standard deviation of the Gaussian distribution, and x is the independent variable of the Gaussian distribution.

[0058] Further, generating the distance weight includes:

[0059]

[0060] where σ s is the standard deviation of the distance weight Gaussian function, and ||(i,j)-(u,v)|| is the Euclidean distance between pixel p ij and the pixel p uv in the neighborhood.

[0061] Specifically: Step 4: For each pixel p ijand the corresponding neighborhood N ij , calculate the central pixel p ij and the pixel p in the neighborhood uv of the Euclidean distance ||(i,j)-(u,v)||, and substitute the calculated distance ||(i,j)-(u,v)|| into the Gaussian function G σ (x) to generate the distance weight. The Gaussian function and the distance weight are as follows:

[0062]

[0063] where σ s is the standard deviation of the distance weight Gaussian function.

[0064] Furthermore, calculating the spectral difference weight includes: obtaining the features of the pixel p ij and the features of the pixel p uv . According to the features of the pixel p ij and the features of the pixel p uv , obtain the spectral clustering matching matrix, and set the objectives and constraints of the spectral clustering matching matrix; the objective is: to minimize the matching cost of transforming the features of the pixel p ij into the features of the pixel p uv . Calculate the spectral difference weight according to the spectral clustering matching matrix.

[0065] Furthermore, obtaining the features of the pixel p ij and the features of the pixel p uv include:

[0066] f ij ={(c 1 , u 1 (p ij )),…,(c k , u k (p ij )),…,(c n , u n (p ij ))}

[0067] f uv ={(c 1 , u 1 (p uv )),…,(c k , u k (p uv )),…,(c n , u n (p uv ))}

[0068] where f ij is the feature of the pixel p ij , fuv For the neighborhood pixel p uv 's feature, c k For the spectral clustering s k 's signal center, u k (p ij ) and u k (p uv ) are the fuzzy membership degrees of the pixel p ij and p uv with respect to the spectral clustering s k . n is the number of spectral clusters of the image.

[0069] Furthermore, the objectives of setting the spectral clustering matching matrix include:

[0070]

[0071] Among them, f ij is the feature of the pixel p ij , f uv is the feature of the neighborhood pixel p uv , DM is the spectral clustering difference matrix, w kl is the matching weight between the spectral clustering centers c k and c l , d kl is the spectral difference between the spectral clustering centers c k and c l . n is the number of spectral clusters of the image.

[0072] Furthermore, the constraint conditions for setting the spectral clustering matching matrix include:

[0073] d kl ≥0 1≤k≤n, 1≤l≤n

[0074]

[0075] Among them, k and l are spectral clustering indices.

[0076] Furthermore, calculating the spectral difference weight includes:

[0077]

[0078] Among them, EMD(f ij , f uv , DM) is the spectral difference between the pixel p ij and p uv , σ emd is the standard deviation of the Gaussian function of the spectral difference weight.

[0079] Specifically: Step Five: Using the pixel p ij in the satellite image as the central pixel, the pixel puv For p ij in the neighborhood N ij of the neighborhood pixels. The pixel p is taken from the fuzzy membership matrix ij and p uv corresponding fuzzy membership vector f ij and f uv . Based on the fuzzy membership vectors f ij and f uv , and the spectral difference matrix DM, calculate the EMD distance EMD(f ij , f uv , DM) between them. Take the EMD distance as the spectral difference between p ij and p uv , and calculate the corresponding spectral difference weight. The spectral difference calculation can be formalized as the following linear programming problem: ij and p uv The spectral difference, calculate the corresponding spectral difference weight. The spectral difference calculation can be formalized as the following linear programming problem:

[0080] f ij ={(c 1 , u 1 (p ij )),…,(c k , u k (p ij )),…,(c n , u n (p ij ))} represents the feature of pixel p ij ; f uv ={(c 1 , u 1 (p uv )),…,(c l , u l (p uv )),…,(c n , u n (p uv ))} represents the feature of the neighborhood pixel p uv ; c k and c l are the signal centers of the spectral clusters s k and s l ; u k (p ij ) and u l (p uv ) are the fuzzy memberships of pixel p ij and p uv with respect to the spectral clusters s k and s l ; n is the number of spectral clusters of the image, and the signal classes of the two-temporal pixel features are not the same; DM = [d kl is the spectral cluster difference matrix, dkl is the spectral clustering center c k and c l of the spectral difference.

[0081] Find the spectral clustering matching matrix W = [w kl , where w kl is the matching weight between the spectral clustering centers c k and c l to minimize the matching cost from the feature f ij to the feature f uv as shown in the following formula:

[0082]

[0083] The formula should satisfy the following constraints:

[0084] d kl ≥0 1≤k≤n, 1≤l≤n

[0085]

[0086] The calculation formula of the spectral difference weight is as follows:

[0087]

[0088] where EMD(f ij , f uv , DM) is the spectral difference between pixels p ij and p uv ; σ emd is the standard deviation of the Gaussian function of the spectral difference weight.

[0089] Furthermore, calculating the EMD bilateral filtering coefficient for each pixel includes:

[0090]

[0091] where is the normalization coefficient, σ s is the standard deviation of the Gaussian function of the distance weight, σ emd is the standard deviation of the Gaussian function of the spectral difference weight, is the Gaussian function of the distance weight, I(p ij ) is the spectral value of pixel p ij , is the Gaussian function of the spectral difference weight.

[0092] Specifically: Step Six: Calculate the EMD bilateral filtering coefficient based on the spectral difference weight and the distance weight, and the formula is as follows:

[0093]

[0094] wherein, is a normalization coefficient.

[0095] Step Seven: For each pixel in the remote sensing image, repeat Steps Four to Six, and perform EMD bilateral filtering on the image based on the EMD bilateral filtering coefficient to obtain an image with Gaussian noise removed.

[0096] such as Figure 2 , such as Figure 2 (a) is the original image of Image One, Figure 2 (b) is the noisy image of Image One, Figure 2 (c) is the denoised image of Image One; such as Figure 3 , Figure 3 (a) is the original image of Image Two, Figure 3 (b) is the noisy image of Image Two, Figure 3 (c) is the denoised image of Image Two; such as Figure 4 , Figure 4 (a) is the original image of Image Three, Figure 4 (b) is the noisy image of Image Three, Figure 4 (c) is the denoised image of Image Three; such as Figure 5 , Figure 5 (a) is the original image of Image Four, Figure 5 (b) is the noisy image of Image Four, Figure 5 (c) is the denoised image of Image Four.

[0097] such as Figure 6 shown, the peak signal-to-noise ratio graph of the present invention.

[0098] such as Figure 7 shown, Figure 7 (a) is the graph of the relationship between the model window size of variance v5 and the signal-to-noise ratio of the denoised image; Figure 7 (b) is the graph of the relationship between the model window size of variance v10 and the signal-to-noise ratio of the denoised image; Figure 7 (c) is the graph of the relationship between the model window size of variance v15 and the signal-to-noise ratio of the denoised image; Figure 7 (d) is the graph of the relationship between the model window size of variance v20 and the signal-to-noise ratio of the denoised image; Figure 7 (e) is the graph of the relationship between the model window size of variance v25 and the signal-to-noise ratio of the denoised image; Figure 7 (f) is the graph of the relationship between the model window size of variance v30 and the signal-to-noise ratio of the denoised image.

[0099] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A bilateral filtering remote sensing image denoising method based on EMD distance, characterized in that: include: Obtaining a high-resolution satellite image contaminated by Gaussian noise, performing cluster analysis on the high-resolution satellite image, obtaining key spectral clusters in the image, extracting the cluster center of each key spectral cluster, and generating a corresponding fuzzy membership matrix, further calculating the distances between all spectral cluster centers, and generating a spectral cluster difference matrix; Obtain each pixel of the target image and its corresponding neighborhood, calculate the Euclidean distance between the pixel and the neighborhood pixels, bring the Euclidean distance center into the Gaussian function, and generate a distance weight; Taking the pixel as the central pixel, extracting the first fuzzy membership vector and the second fuzzy membership vector corresponding to the pixel and the neighborhood pixels from the fuzzy membership matrix, and calculating the EMD distance between the first fuzzy membership vector and the second fuzzy membership vector in combination with the spectral clustering difference matrix, taking the EMD distance as the spectral difference between the pixel and the neighborhood pixels, and calculating the spectral difference weight; The EMD bilateral filtering coefficient of each pixel is calculated according to the spectral difference weight and the distance weight, and the high-resolution satellite image is subjected to EMD bilateral filtering based on the EMD bilateral filtering coefficient of each pixel to obtain a high-resolution satellite image with Gaussian noise removed.

2. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 1, characterized in that: The Gaussian function is: Where σ is the standard deviation of the Gaussian distribution, and x is the independent variable of the Gaussian distribution.

3. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 1, characterized in that: Generating the distance weight includes: Among them, σ s is the standard deviation of the distance weighted Gaussian function, and ||(i,j)-(u,v)|| is the Euclidean distance between a pixel and its neighboring pixels in the neighborhood.

4. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 1, characterized in that: Calculating the spectral difference weight includes: Acquire the features of the pixel and the features of the neighborhood pixels, acquire a spectral clustering matching matrix according to the features of the pixel and the features of the neighborhood pixels, and set the goal and restriction conditions of the spectral clustering matching matrix; the goal is to minimize the matching cost of converting the features of the pixel to the features of the neighborhood pixels; The spectral difference weight is calculated according to the spectral cluster matching matrix.

5. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 4, characterized in that: Acquiring the features of the pixel and the features of the neighborhood pixels includes: f ij ={(c1,u1(p ij )),…,(c k ,u k (p ij )),…,(c n ,u n (p ij ))} f uv ={(c1,u1(p uv )),…,(c k ,u k (p uv )),…,(c n ,u n (p uv ))} Among them, f ij is pixel p ij The characteristics of uv is the neighborhood pixel p uv The characteristics of c k is the spectral clustering s k The signal center, u k (p ij ) and u k (p uv ) is pixel p ij and p uv About Spectral Clustering k The fuzzy membership of n is the number of spectral clusters of the image.

6. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 4, characterized in that: The objectives of setting the spectral clustering matching matrix include: Among them, EMD is the EMD distance function, f ij is pixel p ij The characteristics of uv is the neighborhood pixel p uv The characteristics of , DM is the spectral clustering difference matrix, w kl is the spectral cluster center c k and c l The matching weight between kl is the spectral cluster center c k and c l The spectral difference is , and n is the number of image spectral clusters.

7. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 4, characterized in that: The restriction conditions for setting the spectral cluster matching matrix include: d kl ≥0 1≤k≤n,1≤l≤n Among them, k and l are spectral clustering indices.

8. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 1, characterized in that: Calculating the spectral difference weight includes: Among them, EMD(f ij ,f uv ,DM) is the spectral difference between the pixel and its neighboring pixels, σ emd is the standard deviation of the spectral difference weighted Gaussian function.

9. The method for denoising remote sensing images by using bilateral filtering based on EMD distance according to claim 1, characterized in that: Calculating the EMD bilateral filter coefficients for each pixel involves: in, is the normalization coefficient, σ s is the standard deviation of the distance weighted Gaussian function, σ emd is the standard deviation of the spectral difference weighted Gaussian function, is the distance weighted Gaussian function, I(p ij ) is pixel p ij The spectral value of is the spectral difference weighted Gaussian function.

Images