CN112945384B - A Data Preprocessing Method for Multi-Angle Polarization Satellites - Google Patents

Abstract

The invention discloses a multi-angle data recombination and matching method for L1 radiation/polarization observation and observation geometry suitable for the Gaofen-5 satellite multi-angle polarization imager (DPC/GF-5) and similar on-board sensors, comprising the following steps: Orbit image data block technology; reference pixel selection method; adjacent pixel observation angle matching method based on the combination of observation zenith angle and observation azimuth angle; pixel-by-pixel multi-angle data sequencing and reorganization technology. The multi-angle recombination and matching method for radiation/polarization observation data proposed by the invention can solve the problems of data jumping and banding caused by cropping and splicing of domestic multi-angle polarization satellite images, and overcome the problems caused by the inversion of atmospheric and surface parameters. It can effectively improve the quantitative inversion accuracy.

Description

A Data Preprocessing Method for Multi-Angle Polarization Satellites

technical field

The invention relates to a satellite image multi-angle data recombination and matching method in the fields of remote sensing science and image processing, and is suitable for the multi-angle radiation/polarization data recombination and matching method of multi-angle polarization satellites, especially domestic high-resolution satellites.

Background technique

Satellite remote sensing earth imaging observation technology is currently one of the most effective means to achieve rapid detection of the earth's land, ocean, atmosphere and other layers. It has continuous spatial coverage, large-scale information detection, instantaneous imaging, and high spatial resolution. The detection advantages such as rate can meet the needs of quantitative and qualitative detection of key parameters of the earth in different fields such as climate, environment, resources, ecology, etc., so it has become a research hotspot and an important development direction in recent decades. In remote sensing quantitative applications, the imaging quality and image preprocessing effect of remote sensing images are important factors that determine the inversion accuracy of remote sensing parameters. For example, radiometric correction accuracy, spectral correction accuracy, geometric registration accuracy, atmospheric correction effect, etc. It is an important indicator that has a non-negligible impact on the quantitative inversion results of remote sensing.

Atmospheric particulate matter refers to solid or liquid particulate suspended matter in the earth's atmosphere, including naturally occurring particles such as sand dust, sea salt, volcanic ash, and biological aerosols, as well as carbon aerosols, secondary particulate matter, etc. The diameter range spans up to 5 orders of magnitude, and it has become an important part of the complex giant system of the earth. Atmospheric particulate matter content and its optical, physical, chemical, radiation and other characteristics have an important impact on air quality, climate change, earth observation and other national social development levels and scientific research fields, so it is one of the key goals of satellite remote sensing detection. .

Atmospheric particulate matter satellite remote sensing detection has been developed for decades, and has undergone the establishment and improvement of many technologies such as single-channel detection, multi-channel detection, intensity detection, polarization detection, and multi-angle detection. Representative atmospheric particulate matter detection satellite sensors include: the MODIS sensor on the Aqua satellite has a visible light to short-wave infrared detection channel, and a relatively mature aerosol dark target retrieval algorithm and deep blue retrieval algorithm have been developed; the MISR sensor mounted on the Terra satellite It can realize the observation of the target pixel from multiple angles, and the aerosol inversion algorithm adapted to its multi-angle detection has been developed accordingly; the CALIPSO satellite is equipped with the active lidar CALIOP, and the information of the vertical direction of the aerosol is obtained by receiving the echo signal. ; Geostationary satellites detect specific areas of the earth by staring, and can obtain high-phase regional aerosol monitoring results. Representative sensors include GOCI/COMS in South Korea, AHI/HIMAWARI-8 in Japan, and PMS/GF in my country's high score series. -4.

A common limitation of the above satellite sensors is that the observation dimension is relatively small, so only the aerosol optical depth parameter can be obtained, and the detection ability of other key parameters such as absorption, particle size and other physical properties of atmospheric particles has not yet been developed. Proven technical approach. The POLDER-3 sensor mounted on the French PARASOL satellite has multi-angle, multi-band, intensity, and polarization detection capabilities. It is considered to be the most powerful, comprehensive and effective spaceborne detection method for aerosol detection. extensive attention.

In May 2018, my country launched the Gaofen-5 satellite in the high-resolution earth observation system. It is equipped with a visible light-shortwave infrared hyperspectral camera, a full-spectrum spectral imager, two earth observation sensors, and a multi-angle. Polarization Imager, Atmospheric Environment Infrared Very High Spectral Resolution Detector, Atmospheric Main Greenhouse Gas Monitor, Atmospheric Trace Gas Differential Absorption Spectrometer Four advanced atmospheric observation payloads are the flagship satellites for environmental monitoring in my country's high-resolution satellite series. Among them, the Directional Polarization Camera (DPC) is the first satellite in my country with multi-spectral (covering visible light to near-infrared), multi-angle (9-12 angle imaging), and polarization (detecting 3 polarization components). The carrier wide field of view imager (width 1850 kilometers) has a dedicated detection channel for detecting atmospheric composition information such as aerosols, clouds, and water vapor, and takes into account the detection capabilities of terrestrial and marine environments. DPC realizes the detection of polarization and intensity radiation information at a maximum of 12 angles of the target pixel through continuous high-speed imaging along the track, and obtains polarization and intensity radiation on different spectral channels through the rotation and coupling of the filter and the polarizer wheel. Signal. The DPC/GF-5 launched by China adopts a similar advanced detection mechanism to the French POLDER-3 sensor. It has high-precision detection capabilities in multiple dimensions of spectrum, angle, intensity, and polarization, and has a spatial resolution of 3.3 km, which is higher than that of POLDER-3. The resolution (6*7 km) is 1 times higher, which can meet the high spatial resolution requirements of atmospheric products for important applications such as urban-scale refined atmospheric detection and regional pollution transmission channel monitoring.

However, when the current data is used for quantitative inversion of atmospheric parameters, the effect is not ideal.

SUMMARY OF THE INVENTION

The study found that DPC is similar to POLDER-3/PARASOL. During the production of standardized data products (L1 level), the radiation and polarization observation data of multiple observation angles were cut, adjusted and spliced according to the order of earth observation. In order to ensure that more complete observation information is concentrated in the observation angles that are ranked first (the data observation angles of the same scene imaging are the same), the continuity and readability of the data are improved. But this also brings a problem, that is, a corresponding remote sensing image is not actually a scene image obtained by the sensor at a certain angle, but the observation data from multiple angles is cropped according to certain rules. and the result after splicing. This method will lead to misalignment at the cropped edge in the image, which is mainly manifested in the corresponding observation zenith angle, observation azimuth angle, and observational quantities with angular variation characteristics such as intensity and polarized radiation. beating. When it comes to the spatial domain processing of images, this form of data organization is not conducive to quantitative inversion, especially for the inversion of aerosol parameters that are sensitive to observation geometry, which may lead to spatially discontinuous abnormal bands in the inversion results. Therefore, it is necessary to reprocess the L1-level data products of DPC and similar sensors to restore the single-scene data imaged by the sensor "simultaneously" during shooting, so as to improve the accuracy of quantitative inversion and processing of satellite remote sensing data.

At present, there is no research or report on the method and technology of multi-angle data reorganization and matching for such payloads. There is an urgent need to develop a multi-angle remote sensing data recombination and matching method for the application requirements of domestic multi-angle polarization satellites, which provides an important data basis for improving the quantitative inversion accuracy of atmospheric parameters.

The object of the present invention is to provide a data preprocessing method for multi-angle polarization satellites that improves the accuracy of quantitative inversion of atmospheric parameters; the specific technical scheme is:

A data preprocessing method for a multi-angle polarization satellite, comprising the following steps:

1) Read the number of observation angles in the entire orbit data, determine the effective total number of rows and columns, and divide the obtained orbit satellite data to obtain initial blocks;

2) Determine whether there is at least one complete data column in the initial block; if there is, as a data block to be processed, go to step 4)

3) The initial block is divided by row, and divided into several sub-blocks with at least one complete data column; the sub-blocks are used as the data blocks to be processed;

4) Extract the observation data in the data block to be processed;

5) From one end point in the complete data column as the starting point to the other end point, the current point is used as the reference pixel, and the next point is used as the to-be-processed pixel to form a reference pixel-to-be-processed pixel pair; all Reference pixel-to-be-processed pixel pair constitutes a reference column adjacent pixel pair dataset;

6) Take each point in the complete data column as the starting point, and traverse to the left in turn, the current point is used as the reference pixel, and the next point is used as the pixel to be processed, forming a reference pixel-to-be-processed pixel pair; all The reference pixel-to-be-processed pixel pair constitutes a left adjacent pixel pair dataset; traverse to the right in turn, and use the same method to form a right adjacent pixel pair dataset;

7) Calculate the data of all observation angles of the reference pixel and the data of the current observation angle of the pixel to be processed using formula (1) and formula (2), and determine whether each of the reference pixel-to-be-processed pixel pairs is formed. match;

δz=VZA _ref,i -VZA _test (1);

δa=VAAref _,i - _VAAtest (2);

Among them, δz is the observation zenith angle difference, i is the observation angle serial number, VZA _ref,i is the ith observation zenith angle of the reference pixel, VZA _test is the observation zenith angle of the current observation angle of the pixel to be processed, δa is the observation azimuth difference, VAA _ref,i is the ith observation azimuth of the reference pixel, and VAA _test is the observation azimuth of the current observation angle of the pixel to be processed;

If δz is smaller than the observation zenith angle threshold Tz and δa is smaller than the observation azimuth angle Ta; then it is judged to be a match and go to step 9);

Otherwise, it is judged that it does not match; go to step 8)

8) Select the data of the next observation angle of the pixel to be processed and match the reference pixel through step 7) until it matches;

9) The pixel to be processed is used as the reference pixel, and the matching observation angle is used as the starting observation angle of the reference pixel;

10) Use the same method as step 7) to process the reference column adjacent pixel pair dataset, the left adjacent pixel pair dataset and the right adjacent pixel pair dataset until all reference pixels in the dataset- The pixel pair to be processed is matched;

11) Replace the original data with the observation data corresponding to each observation angle of all reference pixels that have been matched and complete the multi-angle sequence reorganization.

Further, the range of Tz is 0.5-1.0°; the range of Ta is 1.0-2.0°.

Further, the observation data includes geographic coordinate domain information and data domain information of each band and each angle.

Further, the geographic coordinate domain information includes surface elevation, sea and land mask, longitude and latitude, and grid row and column numbers.

Further, the data domain information includes radiance, polarization Stokes component, and observation geometry.

Further, the following steps are also included:

12) According to the starting observation angle sequence number AnS and the effective observation angle number N of the pixel to be processed, calculate the observation angle moving order step size:

Transpose the data of the starting observation angle serial number AnS and later with the data before the starting observation angle serial number AnS, reorganize the data, and calculate the overall difference between VZA and VAA at each observation angle of the reference pixel and the pixel to be processed. Determine whether all observation angles have been matched:

Among them, i represents the observation angle serial number (N in total), the subscripts ref and test represent the reference pixel and the pixel to be processed, respectively, Tz is the observation zenith angle threshold and Ta is the observation azimuth angle threshold;

If it is satisfied, the data verification is successful; if it is not satisfied, the new data is processed from step 1) to step 12); the observation angle where the abnormal data is found and eliminated.

The invention processes the satellite data by dividing the data into blocks with a standard number of rows, forming a pair of reference pixels and pixels to be processed, and sorting, reorganizing and adjusting pixel by pixel; the multi-angle data reorganization and matching method is effective , stable, and the processing results can meet the data basis requirements for quantitative inversion of atmospheric parameters.

Description of drawings

Fig. 1 is the multi-angle data reorganization and matching method process flow proposed by the present invention;

2 is a schematic diagram of the process of further performing data segmentation according to the selection of reference pixels in the embodiment;

Figure 3 is the L1-level data image of DPC/GF-5 with standard line number data blocks in the embodiment (670 nm band, Stokes vectors I, Q, U of the first observation angle, and observation zenith angle, observation Azimuth);

Figure 4 shows the image data after multi-angle data recombination and matching in the embodiment (670 nm band, Stokes vectors I, Q, U of the first observation angle, and observation zenith angle and observation azimuth).

Detailed ways

The present invention will be more fully described below by means of examples. The present invention may be embodied in many different forms and should not be construed as limited to the exemplary embodiments set forth herein.

Taking the actual observation data of the domestic multi-angle polarization satellite sensor DPC/GF-5 as an example, the specific implementation of the present invention is described. The example data used in the present invention is DPC/GF-5 observation data (data time: May 5, 2019, data space range: Eastern China); the data is processed by a computer according to the method of the present invention. The technical flow chart of the method of the present invention is shown in FIG. 1 .

(1) The whole track image data is divided into blocks

First, read the observation angle number information in the entire track data, and judge the valid total number of rows and columns of the track data through invalid values or fill values. Secondly, set the standard block number (which can be set according to the processing performance of the computer), and perform block-by-band and angle-by-angle block processing on the entire track data according to the set value. Figure 3 shows some examples of the data of the DPC/GF-5 example using the standard number of lines (500 lines) to block the data, including the polarization Stokes vectors I670, Q670, U670 in the 670 nm band, the observation zenith angle VZA, and the observation azimuth angle. VAA.

The traditional data block processing can complete the above steps, but for the goal of multi-angle data reorganization that the present invention needs to finally achieve, the image after block may also face various unfavorable situations, especially in high latitude areas (close to the North and South Pole) The images are prone to irregular shapes, such as sharp edges in the images, and multiple columns of invalid data appearing at the edges due to an excessively large aspect ratio of the images, which requires further targeted processing. On the basis of the standard row number block, an innovation of the present invention is that the block processing is further performed according to the need of subsequent reference pixel selection, so as to avoid invalid results of the reference pixel during data reorganization. It is judged whether each data block after preliminary segmentation meets the requirements of reference pixel selection, that is, there is at least one complete data column in the divided image, and all the pixels in this column are valid data. It is noted that there is no complete data column in the block data of the DPC/GF-5 embodiment, so it is necessary to continue to perform block processing on the data block. At least one complete column of data exists in the next two sub-blocks. Figure 2 shows a schematic diagram of the process of further dividing the data in the DPC/GF-5 embodiment into data blocks. Of course, three or more row-wise splits can also be performed as needed. After splitting, the reference column is as close as possible to the middle of all complete columns.

Finally, the corresponding observation data are extracted according to the above block results, including data domain information such as radiance of each band and angle, polarization Stokes component, observation geometry, etc. Coordinate field information. Finally, according to the original L1 level full track data file storage format, the above-mentioned extracted block data information is written into a new block data file to complete the data block processing.

(2) Block data reference pixel selection

The core idea of multi-angle data reorganization and matching proposed by the present invention is to establish a one-to-one corresponding reference pixel-to-be-processed pixel pair for each pixel in the segmented image, and to sort the multi-angle order of the to-be-processed pixels according to the reference pixel make adjustments. Except for the image edge, each pixel is both a reference pixel and a pixel to be processed. The multi-angle order of a pixel is determined by its reference pixel, and it also determines the multi-angle order of the next pixel to be processed. Therefore, the correct selection of the reference pixels of the block data is an important prerequisite for the realization of the present invention. The core problem to be solved in this step is how to determine the reference cell for each cell in the image.

Traditionally, there are two methods for traversing remote sensing images pixel by pixel. One is to traverse vertically and horizontally from the corner point. However, due to the special imaging method and data projection characteristics of such loads as DPC/GF-5, the divided image data is not a standard rectangle or square, but It is a quadrilateral that approximates a parallelogram or trapezoid. This method will encounter a large number of invalid data conditions and cause processing failure. The other is to randomly select a pixel in the middle of the image to traverse in four directions, up, down, left, and right. In case of an invalid value, it is determined to reach the edge of the image and terminate. Although this method can establish a reference pixel for all pixels—the image to be processed However, it will make the subsequent multi-angle recombination processing more complicated and error-prone, and the amount of calculation is large. To solve this problem, the present invention innovatively proposes a reference pixel traversal method based on a double cycle of "point" (reference pixel) - "column" (reference data column) - "row" (each row). First, the starting reference pixel seed is determined according to the block data. Since the segmentation strategy that at least one complete data column exists in the foregoing segmentation step needs to be satisfied, the starting reference pixel is selected from all columns {C _i } that satisfy this requirement, and generally the middle column (C _i } of {C i } is selected. _m ) of the bottom row (R _b ) of the cell, as the starting reference cell seed (row number R _b , column number C _m ). After that, traverse the adjacent pixels from the seed upwards pixel by pixel, and construct the reference pixel-to-be-processed pixel pair for each row of cells in column C _m , namely (R _b , C _m )ßà(R _b -1, C _m ), (R _b -1, C _m )ßà(R _b -2, C _m ), … and so on. After that, take each row of pixels in the reference data column C _m as a reference, traverse left and right pixel by pixel, and construct the corresponding reference pixel-to-be-processed pixel pair, namely (R _b , C _m )ßà( R _b , C _m -1) and (R _b , C _m )ßà(R _b , C _m +1), (R _b -1, C _m )ßà(R _b -1, C _m -1) and ( R _b -1, C _m )ßà(R _b -1, C _m +1), … and so on. Finally, the corresponding relationship between the reference pixel and the pixel to be processed of all the pixels in the data block is established, and the adjacent pixel pair dataset is formed.

(3) Matching of observation angles of adjacent pixels

After the adjacent pixel pair dataset is formed, the observation angle matching is performed for each group of reference pixels-to-be-processed pixels. Subsequent multi-angle data reordering processing provides initial values. When the DPC/GF-5 sensor is imaged at the same observation angle, the observed zenith angle VZA changes very little, but after the aforementioned cropping and splicing, there will be obvious jumps at the edge of the splicing (imaging at different observation angles) , so it can be determined whether the imaging is under the same observation angle by setting the jump threshold of the observation zenith angle. However, each pixel of the DPC/GF-5 data has 9-12 observation angles, and the VZA basically presents front-sight and rear-sight symmetry with the nadir direction VZA as the minimum value according to the order of change from large to small and then large. Variation distribution, it is difficult to correctly match only relying on the observed zenith angle value.

The present invention proposes a new technical idea to solve the above-mentioned problems, and adds the observation azimuth angle to assist in the identification on the basis of the identification and matching of the observation zenith angle. First, according to the variation law of the observed zenith angle VZA and the observed azimuth angle VAA in the block data, the jump thresholds Tz and Ta of VZA and VAA are set, that is, the VZA difference between the reference pixel and the pixel to be processed is less than Tz and When the VAA difference is less than Ta, it is determined that the pixel to be processed and the reference pixel are imaged at the same observation angle, and no cropping and splicing occur. After that, match the VZA and VAA at other observation angles of the reference pixel with the VZA and VAA at the first observation angle of the pixel to be processed (that is, there is no jump), and perform the following processing according to the matching situation: j If there is only one matching result, the matching angle is the starting observation angle AnS of the pixel to be processed; if there is more than one matching result k, the matching angle with the overall smaller δz and δa is selected as the starting point of the pixel to be processed. If there is no matching result, it means that compared with the reference pixel, it is not the same observation angle as the first observation angle of the pixel to be processed. Change the observation angle of the pixel to be processed and repeat the above steps to continue the judgment. Check whether the VZA and VAA at each observation angle of the reference pixel match the VZA and VAA at the second observation angle of the pixel to be processed, so as to find the starting observation angle of the pixel to be processed. After the matching is successful, the data of the pixel to be processed replaces the data of the reference pixel in the next reference pixel-to-be-processed pixel pair for matching. When matching, firstly perform the matching of all reference pixel-to-be-processed pixel pairs in the reference column adjacent pixel pair dataset; All reference-to-be-processed cell pairs in the dataset are processed by the right neighbor cell pair described above. Finally, the initial observation angle sequence number of each pixel to be processed is established, which is used for subsequent pixel-by-pixel sequence reorganization. The jump calculation formula is:

δz=VZA _ref,i -VZA _test (1);

δa=VAAref _,i - _VAAtest (2);

(4) Pixel-by-pixel multi-angle data sequencing and reorganization

After the initial observation angle serial number of the pixel to be processed is determined, the data at each angle can be reorganized in sequence according to the change rule of the observation zenith angle and the observation azimuth angle. However, due to the missing individual viewing angles for some cells, just doing a sequential reorganization would result in a sequential misalignment of the angles following the missing angle. Therefore, in the actual data processing process of the present invention, the strategy of successive reorganization and matching detection is adopted, that is, matching detection is performed after each reorganization, the remaining unmatched observation angles are reprocessed, and the observation angles where abnormal data are located are found and eliminated. Until all valid observation angles are matched. Specific steps are as follows:

First, according to the initial observation angle sequence number AnS and the effective observation angle number N of the pixel to be processed, calculate the observation angle moving order step size:

Move the AnSth observation angle of the pixel to be processed to the last observation angle, move it to the first L+1 angle of the new data, and move the first AnS-1 angle of the pixel to be processed to the L+th angle of the new data. 2 to the last angle to complete the first reorganization. After that, by calculating the overall difference between VZA and VAA at each observation angle of the reference pixel and the pixel to be processed, it is judged whether all the observation angles have satisfied the matching:

Among them, i represents the observation angle serial number (N in total), the subscripts ref and test represent the reference pixel and the pixel to be processed, respectively, T _z is the observation zenith angle threshold and T _a is the observation azimuth threshold.

If there are still unmatched observation angles, further filter the unmatched observation angles according to the threshold, take the first one as the new starting observation angle serial number, and repeat the above steps until all the observation angles of the pixel to be processed are Match with the reference pixel, satisfy formula (4) and formula (5), and output the observation angle order adjustment array of the pixel to be processed. The range of Tz is preferably 0.5-1.0°; the range of Ta is 1.0-2.0°. If the threshold is too low, a complete match cannot be guaranteed; if the threshold is too high, the computation time will increase dramatically.

Finally, according to the ordered array, rewrite the multi-band intensity radiation, polarization Stokes vector, observation zenith angle, observation azimuth, solar zenith angle, solar azimuth, etc. under each observation angle of the pixel to be processed into a new Data files, complete multi-angle sequencing and reorganization. Figure 4 shows the results of the multi-angle sequencing and reorganization of the block data in the DPC/GF-5 embodiment. It can be seen that the Stokes vector I, The jumps of the Q and U images have been corrected, the observation zenith angle and the observation azimuth angle strips have been restored to the images of the sensor imaging, and the change law conforms to the physical scene, indicating that the multi-angle data reorganization and matching method proposed by the present invention is effective and stable. , which can meet the data base requirements for quantitative inversion of atmospheric parameters.

The above examples are only used to illustrate the present invention. In addition, there are many different implementations, and these implementations can be thought of by those skilled in the art after comprehending the idea of the present invention. Therefore, they are not repeated here. an enumeration.

Claims (6)

1. A data preprocessing method for a multi-angle polarized satellite is characterized by comprising the following steps:

1) reading observation angle number information in the whole-orbit data, judging effective total line number and total column number, and segmenting the acquired whole-orbit satellite data to obtain initial blocks;

2) judging whether at least one complete data column exists in the initial block; if yes, as the data block to be processed, executing the step 4)

3) Dividing the initial block into a plurality of sub-blocks with at least one complete data column; the sub-blocks are used as data blocks to be processed;

4) extracting observation data in the data block to be processed;

5) traversing from one end point in the complete data column as a starting point to the other end point in sequence, wherein the current point is used as a reference pixel and the next point is used as a pixel to be processed to form a reference pixel-pixel to be processed pair; all reference pixel-pixel pairs to be processed form a reference column adjacent pixel pair data set;

6) Traversing each point in the complete data column serving as a starting point in sequence leftwards, wherein the current point serves as a reference pixel and the next point serves as a pixel to be processed to form a reference pixel-pixel to be processed pair; all reference pixel-to-be-processed pixel pairs form a left adjacent pixel pair data set; traversing sequentially to the right, and forming a right adjacent pixel pair data set by adopting the same method;

7) calculating the data of all observation angles of the reference pixel and the data of the current observation angle of the pixel to be processed by adopting a formula (1) and a formula (2), and judging whether each formed reference pixel-pixel to be processed pair is matched;

δz=VZA_ref,i-VZA_test （1）；

δa=VAA_ref,i-VAA_test （2）；

wherein, delta z is the difference value of the observation zenith angle, i is the serial number of the observation angle, VZA_ref,iIs the ith observation of the reference pixelZenith angle, VZA_testThe observation zenith angle of the current observation angle of the pixel to be processed is delta a, the difference value of the observation azimuth angle is VAA_ref,iIs the i-th observation azimuth, VAA, of the reference pixel_testThe observation azimuth angle of the current observation angle of the pixel to be processed is obtained;

δ z is smaller than the observation zenith angle threshold Tz and δ a is smaller than the observation azimuth angle Ta; judging the matching and entering the step 9);

otherwise, judging as mismatching; entering step 8)

8) Selecting data of a next observation angle of the pixel to be processed and the reference pixel, and matching the data and the reference pixel in the step 7) until the data and the reference pixel are matched;

9) The pixel to be processed is used as a reference pixel, and the matched observation angle is used as an initial observation angle of the reference pixel;

10) processing the reference column adjacent pixel pair data set, the left adjacent pixel pair data set and the right adjacent pixel pair data set by adopting the same method in the step 7) until all reference pixel-to-be-processed pixel pairs in the data sets are matched;

11) and replacing the observation data corresponding to each observation angle of all the reference pixels subjected to matching processing with the original data to finish multi-angle sequencing recombination.

2. The method for pre-processing data for a multi-angle polarized satellite of claim 1, wherein Tz ranges from 0.5 to 1.0 °; ta ranges from 1.0 to 2.0 deg.

3. The method for pre-processing data of a multi-angle polarized satellite according to claim 1, wherein the observation data comprises geographical coordinate domain information and data domain information for each wave band and each angle.

4. The method for preprocessing data of a multi-angle polarized satellite of claim 3, wherein the geographic coordinate domain information comprises surface elevation, sea-land mask, latitude and longitude, grid row and column number.

5. The method for data pre-processing of a multi-angle polarized satellite of claim 3, wherein the data domain information comprises radiance, polarized Stokes components, observation geometry.

6. The method for pre-processing data for a multi-angle polarized satellite of claim 1, further comprising the steps of:

12) calculating the step length of the movement sequence of the observation angle according to the initial observation angle sequence number AnS and the effective observation angle number N of the pixel to be processed:

the data of the initial observation angle sequence number AnS and the subsequent data are transposed with the data of the initial observation angle sequence number AnS, the data are recombined, the overall difference value of VZA and VAA under each observation angle of the reference pixel and the pixel to be processed is calculated, and whether all the observation angles meet the matching condition is judged:

wherein i represents an observation angle sequence number; the number of the observation angles is N, subscripts ref and test respectively represent a reference pixel and a pixel to be processed, Tz is an observation zenith angle threshold value and Ta is an observation azimuth angle threshold value;

if yes, the data check is successful; if not, processing the new data from the step 1) to the step 12); and finding and eliminating the observation angle of the abnormal data.

Images