A method for reconstructing data affected by cloud and snow
By using spatial distance and spectral similarity to screen and fill similar cells when processing MODIS data, the data loss problem under the influence of cloud and snow is solved, and efficient and accurate data reconstruction is achieved, especially in seasons in Nepal and other regions. Reconstruction of long-term data under the influence of sexual snow.
Patent Information
- Application Number
- CN202210629380.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-02
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When processing data affected by cloud and snow, there are problems of continuous data loss and large-scale data loss, especially in low latitude and high altitude areas such as Nepal, which cannot effectively reconstruct long-term series data caused by seasonal snow. Missing, with reduced computational efficiency and inaccurate search for similar pixels.
By obtaining the initial image of the target area, pre-processing is performed to obtain the pending image, removing the data affected by cloud and snow, selecting similar cells in the initial window, and normalizing the spatial distance and spectral similarity as weights. Filter out the target cell and fill it into the empty value area to reconstruct the MCD43A4 dataset.
It improves the accuracy and efficiency of data reconstruction, can effectively eliminate data loss, and reconstruct a stable and complete data set, especially in long-term data reconstruction under the influence of seasonal snow.
Smart Images

Figure CN114998468B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of earth observation, and in particular to a method for reconstructing data affected by clouds and snow. Background Art
[0002] MODIS (Moderate Resolution Imaging Spectroradiometer) is carried on two satellites, Terra and Aqua. The combination of the two satellites can provide global surface coverage observation data every 1-2 days, including 36 discrete spectral bands, and can obtain surface data with spatial resolutions of 250m, 500m and 1km. The MODIS sensor was successfully launched in 1999 and began to transmit data in 2000. Its products are characterized by long-term, large-scale, continuous and stable, and are particularly suitable for monitoring and research with wide spatial distribution and long time duration. The series of products produced have been verified by various rigorous evaluation work, with high reliability and accuracy, and can provide a large amount of available data for land, ocean, cryosphere and solar power system.
[0003] MODIS's land product MOD09GA can provide daily surface reflectance data of 500m for channels 1-7 after atmospheric correction. It has high temporal and spatial resolution and data integrity, can represent the real surface reflectance value, and is the original underlying data for producing other products. The MCDA3A4 Nadir BRDF Corrected Surface Reflectance Product (NBAR) comes from MOD09 and MYD09 data. It is processed into the surface reflectance observed in the nadir direction after bidirectional reflectance distribution function correction, eliminating the influence of the observation angle direction effect, and can obtain more stable and consistent surface reflectance data. Based on the MCD43A4 surface reflectance data, it has high accuracy in vegetation monitoring, surface cover classification, and atmospheric monitoring research. However, the MCDA3A4 data is inverted through multi-day, multi-angle, cloud-free, and atmospheric correction processing to obtain surface reflectance data. During the processing process, the existence of clouds, cloud shadows, and snow caused data loss to a certain extent, which hindered the practical application of this product.
[0004] In order to solve the data loss caused by clouds, cloud shadows, snow cover or sensor failure, many gap filling algorithms have been developed. The existing gap filling methods can be divided into three categories according to the auxiliary information used:
[0005] Methods based on spatial information, based on temporal information, and based on a combination of temporal and spatial information.
[0006] The data filling methods based on spatial information mainly include the geostatistical method of Kriging interpolation, which uses the surrounding valid pixel values to interpolate missing data;
[0007] Data interpolation methods based on time information include time replacement, time filtering, and time learning models, which use the valid pixel values of the same position of the missing pixels at different times to reconstruct the missing data;
[0008] A hybrid method based on time and space combines two or more of the above methods to capture more information and obtain more accurate results. One of the data interpolation methods based on NSPI iteration is a time and space based gap filling method developed to fill the ETM+ missing strip problem. It is relatively simple and has high computational efficiency. It makes full use of the time and space information in the time series and obtains better missing data reconstruction results in some applications.
[0009] Existing technologies have achieved good results in data reconstruction, but the following problems still exist:
[0010] 1. Continuous data loss. The data loss is caused by the presence of clouds and seasonal snow, especially in low-latitude and high-altitude areas such as Nepal. Existing algorithms can reconstruct the data of short-term data loss caused by the presence of clouds based on time series, but cannot effectively reconstruct the data of long-term data loss caused by the presence of seasonal snow.
[0011] 2. Large-scale data missing. Existing technologies have solved the problem of missing Landsat ETM+ stripes, but for large-scale data missing, the window size increases as the missing pixel ratio increases, resulting in a decrease in computational efficiency and inaccurate search for similar pixels. Summary of the invention
[0012] The purpose of the present invention is to solve the above problems existing in the prior art.
[0013] To achieve the above-mentioned purpose, the present invention provides a method for reconstructing data affected by snow and clouds, comprising: obtaining an initial image acquired by an image acquisition device of a target area at a target time, the initial image comprising: a plurality of initial images of the target area arranged in a time series; performing a first preprocessing on the initial image to obtain an image to be processed, the image to be processed comprising: a second null value area of unreconstructed data; selecting an initial window with the second null value area as the center on each of the plurality of images to be processed; selecting a second similar pixel similar to the first data affected by snow and clouds from the second data not affected by snow and clouds within the initial window according to a predetermined threshold; using spatial distance and spectral similarity as weights, and processing the weights to obtain normalized weights, screening all the second similar pixels by the normalized weights to obtain the second target pixel, and filling the second target pixel into the second null value area to obtain the target image with the MCD43A4 data set.
[0014] A method for reconstructing data affected by cloud and snow according to an embodiment of the present invention improves the accuracy and efficiency of data reconstruction, and reconstructs a stable and complete data set for data missing caused by elimination. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 This is a flow chart of a method for reconstructing data affected by cloud and snow according to an embodiment of the present invention;
[0016] Figure 2 A flowchart of an example of a method for reconstructing data affected by cloud and snow according to an embodiment of the present invention;
[0017] Figure 3 This is the existing data map for Nepal;
[0018] Figure 4 is a data graph after data reconstruction according to an embodiment of the present invention;
[0019] Figure 5 This is a data reconstruction evaluation table according to an embodiment of the present invention. DETAILED DESCRIPTION
[0020] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.
[0021] Figure 1 A method for reconstructing data affected by cloud and snow according to an embodiment of the present invention comprises:
[0022] S101: Acquire an initial image of a target area captured by an image acquisition device at a target time, the initial image comprising: initial images of a plurality of target areas arranged in time series;
[0023] After the initial image of the target area at the target time is acquired, an initial image in which the initial image is arranged in time sequence can be acquired simultaneously. Each initial image includes a plurality of pixels, and each pixel has image data of the target area.
[0024] The initial image is acquired by an image acquisition device. In some embodiments of the present invention, the image acquisition device may be a medium-resolution imaging spectrometer.
[0025] S102: performing a first preprocessing on the initial image to obtain an image to be processed, wherein the image to be processed includes: a second null value region without reconstructed data;
[0026] After obtaining the initial image, a data set formed by multiple initial images with a time series can be obtained. After a first preprocessing is performed on the multiple initial images in the data set to obtain an image to be processed, subsequent data reconstruction can be facilitated. The image to be processed has a second null value area, and the second null value area is the area where data reconstruction was not performed in the first preprocessing.
[0027] The first preprocessing method in step S102 includes the following sub-steps:
[0028] S1021: removing first data affected by clouds and snow in each initial image to obtain a first null value region, evaluating the quality of each initial image, and marking all initial images according to the evaluation results;
[0029] Specifically, each of the initial images includes multiple pixels. As mentioned above, since the target area may be affected by clouds and seasonal snow, when the satellite collects data images of the ground, some pixels in the acquired image information are affected by clouds and seasonal snow, resulting in poor shooting quality.
[0030] Therefore, the pixel information with poor quality such as unclear image is defined as the first data affected by clouds and snow, and the first data affected by clouds and snow is eliminated, and a no-data area is formed in the area of the eliminated first data affected by clouds and snow, which is the first null value area.
[0031] Furthermore, the first data affected by clouds and snow in all the initial images are removed by MCD43A4 to form a first null value area.
[0032] That is to say, the evaluation of the first data affected by clouds and snow is carried out by processing all the initial images through MCD43A4. In the processed initial image data, all the pixel information can obtain the pixels with poor pixel quality caused by the impact, which is defined as the first data affected by clouds and snow. Therefore, the pixels defined as the first data affected by clouds and snow can be eliminated. In each initial image, the position after eliminating the first data affected by clouds and snow, that is, the position without data information, is defined as the first null value area.
[0033] Furthermore, the quality of each initial image is evaluated, and all initial images are marked according to the evaluation results; the quality of the initial images can be evaluated based on the MCD43A4 QA quality evaluation band.
[0034] By performing quality evaluation on all initial images, the quality of each initial image can be effectively obtained, that is, whether the quality of the initial image is good or bad.
[0035] In some embodiments of the present invention, the quality evaluation of all initial images can be performed by preparing mask data of each initial image. After each initial image is quality evaluated and marked, it is convenient to check whether the image quality is good or bad. It is understandable that if the quality of the initial image is good, it can be used as the basis for subsequent search of similar pixels. If the quality of the initial image is poor, it can be directly skipped as the basis for searching similar pixels, thereby improving the search efficiency of similar pixels.
[0036] The initial image may be marked in a binary data manner, for example, a poor image may be marked as 0 and a good image may be marked as 1, and the image marked as 0 may be directly skipped during subsequent searches for similar pixels.
[0037] After step S1021, the image information in all the initial images may also be analyzed and classified;
[0038] Specifically, by analyzing the image information in the initial image, the data in the image can be obtained, and the image information in the same category can be classified into one category. When searching for similar pixels in pixel information of the same category, the search can be directly performed in the image information of the same category, which can also improve the efficiency of the search and the efficiency of the overall data reconstruction.
[0039] Furthermore, the image information in all the initial images is classified based on the typical K-means unsupervised classification, and the image information includes: embankments, roads, vegetation, etc.
[0040] That is to say, the image information may include information such as embankments, roads and vegetation. By classifying pixels displayed as roads into one category and information displayed as vegetation into another category, when a pixel recording road information in an initial image is missing, similar pixels can be directly searched in the pixels of the road information category. The search efficiency of similar pixels can also be improved. Moreover, by searching similar pixels in the same category, the accuracy and stability of data reconstruction are guaranteed.
[0041] S1022: searching all first similar pixels similar to the first data affected by cloud and snow removed from the first null value region from the second data according to a predetermined method, and processing the weights according to the spatial distance and the spectral similarity, and screening all first similar pixels by normalizing the weights;
[0042] Specifically, in this step, pixels with poor quality in the initial image are recorded as the first data affected by clouds and snow, and pixels with better quality are recorded as the second data. After the first data affected by clouds and snow are eliminated, a first null value area will be displayed. In order to search for similar pixels of the first data affected by clouds and snow, that is, first similar pixels, the second data can be searched.
[0043] The predetermined method may be based on a time series, in which the time of the first data affected by the cloud and snow is from far to near.
[0044] As can be seen from the foregoing, after obtaining the initial image, the initial images arranged in time series can be obtained. After the first data affected by clouds and snow that needs to be removed appears in a certain initial image, the search for similar pixels needs to be searched in time sequence.
[0045] For example, if the first data affected by cloud and snow appears in the initial image of the 10th second, the search for the first similar pixel can be done in the second data in the initial image of the 11th second, or in the second data in the initial image of the 12th second. However, the search needs to be done in chronological order, that is, the search needs to be done in the image of the 11th second, and then in the image of the 12th second.
[0046] In other words, the search starts from the image closest to the target date. It can be understood that the target date can be the initial image of the 10th second mentioned above.
[0047] Furthermore, in order to obtain similar pixels with higher accuracy, the search process can be repeated to find at least 20 similar pixels. Of course, in order to improve the accuracy, at least 30 similar pixels can be searched according to actual needs. It is also possible to traverse all the pixels of the initial image until the iteration ends.
[0048] The spectral similarity RMSD is obtained by the following formula:
[0049]
[0050] The spatial distance D is obtained by the following formula:
[0051]
[0052] The weight normalization is done through the following formula:
[0053] CD j =RMSD J ×D J ;
[0054]
[0055] Where D represents the spatial distance, t1 represents the date of the first similar pixel, L(x,y,t1,b) represents the reflectivity value of the first data affected by cloud and snow, and L(x y ,y i,t1,b) represents the reflectance value of the first similar pixel, W represents the normalized weight, and N represents the number of the first similar pixels. The formula for spectral similarity can be determined based on the formula for the root mean square error.
[0056] The smaller the root mean square error (RMSD) obtained by the above formula, the higher the accuracy.
[0057] The first similar pixel finally screened by the normalized weights has higher quality. The more similar it is to the first data affected by clouds and snow, the more stable the data after final reconstruction will be.
[0058] S1023: simulating the screened first similar pixel and the eliminated first data to obtain a corresponding first target pixel;
[0059] Specifically, the first similar pixel obtained after screening is simulated with the first data affected by cloud and snow that has been removed, and the first target pixel obtained has good quality. After the data is reconstructed, the data has high accuracy and good stability.
[0060] Furthermore, in the reconstruction method of the data affected by cloud and snow in the embodiment of the present invention, the simulation of the first data is reconstructed by two parts of data, one part predicts the missing pixels based on the spatial information using similar pixels in the target pixel window of the target date (t1), as shown in the following formula:
[0061]
[0062] The second part is based on time information. The first data removed at time t2 is equal to the pixel value at time t1 plus the value changed from t2 to t1, as shown in the following formula:
[0063]
[0064] Finally, spatial continuity and temporal continuity are used as the weights of the two initial predictions to perform a weighted combination to calculate the final simulation result, as shown in the following formula:
[0065] L(x,y,t2,b)=T1×L1(x,y,t t ,b)+T2×L2(x,y,t t ,b).
[0066] S1024: Filling at least a portion of the first null value region of the initial image with the first target pixel to obtain an image to be processed, and redefining the unfilled region in the first null value region as a second null value region;
[0067] Specifically, in this step, the first target pixel obtained can fill the first empty value area. Since the number of first data affected by cloud and snow eliminated in all initial images can be relatively large, the number of first similar pixels obtained through simulation can be less than the number of first data affected by cloud and snow. Therefore, the first empty value area cannot be completely filled, and the filled initial image is the image to be processed. Similarly, this is also a problem that the prior art cannot completely reconstruct data with high precision for long-term data loss. The long-term data loss caused by seasonal snowfall cannot be well filled. For this part of the missing data, the complete original underlying data MOD09GA of the target date is used to simulate the missing data of this part, and the MOD09GA data of the same study area corresponding to the target date is collected.
[0068] Therefore, in the present invention, the unfilled first empty value region is redefined as the second empty value region.
[0069] S103: selecting an initial window with the second null value area as the center on each of the multiple images to be processed;
[0070] Specifically, in the image to be processed, the unfilled area is the second null value area, so an initial window is selected in the image to be processed with the second null value area as the center, with the purpose of limiting the second null value area to the center position of the initial window.
[0071] Furthermore, the size of the initial window is centered on the position of the first data affected by the cloud and snow, and the space of ±w*±w is selected as the initial window. The size of the initial window can be freely defined, such as 10*10 or 20*20. Those skilled in the art can understand that the size setting of the initial window can be set according to actual needs.
[0072] Furthermore, at least 20 similar pixels are searched in the initial window. If there are less than 20 similar pixels, the position of the first data affected by the cloud and snow is taken as the center to expand one pixel in all directions, that is, w=w+1, until the minimum number of similar pixels is met or the search range exceeds half of the entire image, then the search is stopped.
[0073] S104: selecting, according to a predetermined threshold, second similar pixels similar to the first data affected by clouds and snow from the second data not affected by clouds and snow in the initial window;
[0074] Specifically, through the predetermined threshold, the predetermined threshold can be used as a boundary, that is, the pixels exceeding the predetermined threshold can be defined as non-similar pixels, and the pixels below the predetermined threshold can be defined as second similar pixels.
[0075] Furthermore, each pixel in the initial window and the first data affected by the cloud and snow removed are calculated by the following formula to obtain a calculation result;
[0076] If the calculated result is less than the predetermined threshold, the pixel is taken as the second similar pixel. The formula is as follows:
[0077]
[0078] Among them, RMSD represents the root mean square error, σ(b) represents the standard deviation of the entire image, m represents the number of categories estimated for the entire image, and n is the number of bands.
[0079] S105: The spatial distance and the spectral similarity are used as weights and the weights are processed to obtain normalized weights, all second similar pixels are screened by the normalized weights to obtain second target pixels, and the second target pixels are filled into the second empty value area to obtain a target image with the MCD43A4 data set.
[0080] Specifically, in this step, the second similar pixels need to be screened to obtain the second target pixels.
[0081] Furthermore, it is necessary to use the spatial distance and spectral similarity as weights, and finally normalize the weights, and screen all the second similar pixels through the normalized weights.
[0082] The screened second similar pixels are reconstructed to obtain the second target pixels. After the second target pixels are obtained, the second target pixels are filled into the positions where the image data is missing in the corresponding second null value regions, thereby obtaining the target image. In the embodiment of the present invention, the image data is obtained through the MCD43A4 image acquisition device, therefore, the obtained target image is also based on the MCD43A4 data set.
[0083] The spectral similarity RMSD is obtained by the following formula:
[0084]
[0085] The spatial distance D is obtained by the following formula:
[0086]
[0087] The weight normalization is done through the following formula:
[0088] CD j =RMSD J ×D J ;
[0089]
[0090] The second similar pixel is reconstructed by the following formula:
[0091]
[0092] Where D represents the spatial distance, t1 represents the date of the first similar pixel, L(x,y,t1,b) represents the reflectivity value of the first data affected by cloud and snow, and L(x y ,y i ,t1,b) represents the reflectance value of the first similar pixel, W represents the normalized weight, and N represents the number of the first similar pixels.
[0093] Combine the following Figure 2 , illustrated with practical examples.
[0094] like Figure 2 As shown, the reconstruction method for MCD43A4 data of the present invention is mainly divided into two steps.
[0095] First, the first part preliminarily reconstructs the temporary data loss caused by cloud influence based on the NSPI iteration method, namely, steps S101 to S106.
[0096] Obtain the initial image data of MCD43A4, obtain the time series of the initial image of the initial image data, and make a mask. Further unsupervised classification is performed, that is, image information such as roads, embankments or vegetation in the initial image is classified to facilitate subsequent similar pixel searches.
[0097] The search for similar pixels is done by searching for the dates close to the first data affected by cloud and snow that have been removed, and then searching for multiple similar pixels through iteration.
[0098] A first target pixel is obtained by simulating and constructing multiple similar pixels, and the first empty value area is filled by the first target pixel. If it is completely filled, the data reconstruction is completed. If it is not completely filled, the unfilled first empty value area is defined as the second empty value area.
[0099] Next is the second part, namely step S107-step S109. For the missing data that cannot be reconstructed in the first step, a linear weighted model is used to simulate the missing pixels: assuming that similar pixels have the same spectral characteristics, based on the original underlying data MOD09, the size of the adaptive window is calculated with the missing pixel as the center, and the threshold method is used to search for similar pixels. The spatial distance and spectral difference between the target pixel and the similar pixel are extracted as weights, and the missing pixel value is simulated by linear weighting.
[0100] That is, on the initial image, an initial window is set with the second null value area as the center, and the second similar pixel is searched within the range of the initial window. The second target pixel is obtained by screening and reconstructing the second similar pixel, and the second target pixel fills the second null value area.
[0101] The reconstruction method of data affected by cloud and snow according to the embodiment of the present invention solves the significant problems existing in the existing data reconstruction methods. It can reconstruct the missing data in large areas and long time series, and reconstruct the surface cover under cloud cover for the missing data caused by the combined influence of cloud and seasonal snow in MCD43A4 data, while retaining the surface type of long-term seasonal snow. It can build a complete and long-term stable MCD43A4 data set.
[0102] Combine the following Figure 3-Figure 5 The effect of the method for reconstructing data affected by clouds and snow according to an embodiment of the present invention is described.
[0103] like Figure 3 As shown in the figure, the MCD43A4 data of Nepal in the dry season of 2016 has missing data due to the influence of clouds and snow. The parts with lighter colors and diagonal lines relative to other areas are the areas with missing data.
[0104] like Figure 4 As shown in the figure, it is the result of data reconstruction. From the results shown in the figure, it can be seen that the part with missing data has a good reconstruction effect, and the snow part has also been well restored. The method proposed in this paper is quantitatively compared with the NSPI iterative method. Four accuracy indicators, namely relative global dimensionless error (ERGAS), structural similarity index (SSIM), root mean square error (RMSE) and correlation coefficient (CC), are selected for quantitative comparison. Figure 5 As shown, it is found that the improved method for reconstructing data affected by cloud and snow in the embodiment of the present invention performs well in all four indicators. In summary, the improved method we proposed can achieve good reconstruction results in qualitative and quantitative evaluation.
[0105] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for reconstructing data affected by cloud and snow, characterized in that ,include: Acquire an initial image of a target area captured by an image acquisition device at a target time, wherein the initial image includes: a plurality of initial images of the target area arranged in time series; Performing a first preprocessing on the initial image to obtain an image to be processed, wherein the image to be processed includes: a second null value region without reconstructed data; Selecting an initial window with the second null value area as the center on each of the plurality of images to be processed; Selecting, according to a predetermined threshold, second similar pixels similar to the first data affected by clouds and snow from the second data not affected by clouds and snow within the initial window; By taking the spatial distance and the spectral similarity as weights and processing the weights to obtain normalized weights, all the second similar pixels are screened by the normalized weights to obtain second target pixels, and the second target pixels are filled into the second null value region to obtain a target image having the MCD43A4 data set; The first preprocessing comprises: Eliminate the first data affected by clouds and snow in each of the initial images to obtain a first null value region, evaluate the quality of each of the initial images, and mark all the initial images according to the evaluation results; Searching all first similar pixels similar to the first data removed from the first null value region from the second data according to a predetermined method, and taking spatial distance and spectral similarity as weights, and processing the weights, and screening all the first similar pixels by the normalized weights; Simulating the first similar pixel after screening with the first data removed to obtain a corresponding first target pixel; The first target pixel is used to fill at least a part of the first null value region in the initial image to obtain an image to be processed, and the unfilled area in the first null value region is redefined as a second null value region.
2. The method according to claim 1, characterized in that: The step of removing the first data affected by clouds and snow in each of the initial images to obtain a first null value region comprises: The first data affected by clouds and snow in all the initial images are removed by the MCD43A4 to form the first null value area.
3. The method according to claim 1, characterized in that: The evaluating the quality of each of the initial images comprises: The quality of the initial image is evaluated based on the MCD43A4QA quality evaluation band.
4. The method according to claim 1, characterized in that: The reservation method includes: Based on the time series, the time of the first data affected by cloud and snow is arranged from far to near.
5. The method according to claim 1, characterized in that The spectrum similarity RMSD is obtained by the following formula: The spatial distance D is obtained by the following formula: The weight processing is done by the following formula: CD j =RMSD J ×D J ; Where D represents the spatial distance, t1 represents the date of the first similar pixel, L(x,y,t1,b) represents the reflectivity value of the first data affected by cloud and snow, and L(x y ,y i ,t1,b) represents the reflectance value of the first similar pixel, W represents the normalized weight, and N represents the number of the first similar pixels.
6. The method according to claim 5, characterized in that The first similar pixel after screening is simulated with the first data affected by cloud and snow removed to obtain the corresponding first target pixel, through the following formula: L(x,y,t2,b)=T1×L1(x,y,t t ,b)+T2×L2(x,y,t t ,b)。 7. The method according to claim 1, characterized in that The size of the initial window is centered on the position of the first data affected by snow and cloud, and a space of ±w*±w is selected as the initial window.
8. The method according to claim 7, characterized in that The selecting, according to a predetermined threshold, a second similar pixel similar to the first data affected by cloud and snow from the second data in the initial window comprises: Calculate each pixel in the initial window and the first data affected by the removed cloud and snow by the following formula to obtain a calculation result; If the calculation result is less than the predetermined threshold, the pixel is used as the second similar pixel. The formula is as follows: Among them, RMSD represents the root mean square error, σ(b) represents the standard deviation of the entire image, m represents the number of categories estimated for the entire image, and n is the number of bands.
9. The method according to claim 8, characterized in that The spectrum similarity RMSD is obtained by the following formula: The spatial distance D is obtained by the following formula: The weight processing is done by the following formula: CD j =RMSD J ×D J ; The second similar pixel is reconstructed by the following formula: Where D represents the spatial distance, t1 represents the date of the first similar pixel, L(x, y, t1, b) represents the reflectivity value of the first data affected by cloud and snow, and L(x y ,y i ,t1,b) represents the reflectance value of the first similar pixel, W represents the normalized weight, and N represents the number of the first similar pixels.
