Satellite ultraviolet hyperspectral ozone detector - bottom of the data preprocessing despeckling method and system

Abstract

The present application relates to hyperspectral imaging data processing technical field, especially relates to a kind of based on satellite ultraviolet hyperspectral ozone detector-bottom data preprocessing striping method and system.The method reads L0 original data and extracts DN value, constructs three-dimensional data matrix and carries out dark current correction;According to wavelength, extract spatial two-dimensional matrix, based on spatial dimension, carry out adaptive hierarchical smoothing filtering, obtain the DN value after striping filtering;The DN value after striping filtering is converted into radiance, and the radiance filter matrix is obtained;Radiance filter matrix is reversed according to spectral dimension and spliced, and the reconstructed three-dimensional data matrix is obtained, and L1 data is generated by the integration of reconstructed three-dimensional data matrix, geographical positioning and quality mark.Through dark background correction and spatial dimension filtering band adaptation, the present application adapts different band spatial pixel characteristics and SMILE effect, improves data quality and inversion accuracy, avoids the influence of SMILE effect, and suppresses random striping noise.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral imaging data processing technology, and in particular to a method and system for preprocessing and removing banding from satellite ultraviolet hyperspectral ozone detector-nadir data. Background Technology

[0002] The FY-3F satellite is a low-Earth orbit satellite and one of the operational observation satellites in my country's Fengyun-3 series of meteorological satellites. Its main mission is to conduct high-precision observations of global atmospheric composition and meteorological elements. FY-3F carries an ultraviolet hyperspectral ozone detector-nadir, operating in the 250-495nm wavelength range. Using nadir observation mode, it can acquire hyperspectral radiation data of the global atmosphere, serving as a key data source for retrieving ozone vertical profiles, ozone column concentrations, and column concentrations of trace gases such as NO2 and SO2.

[0003] The core detection unit of this instrument is an area-array CCD detector with a pixel size of 256 rows × 1024 columns. Rows 1-9 and 248-256 are dark pixel regions, while rows 10-247 are scientific observation pixels. In nadir observation mode, the two detectors read one frame of hyperspectral data per second. Each detector contains radiance DN values ​​in the 250-320nm and 310-495nm wavelength ranges, respectively. Multiple frames of data are stitched together to form an observational data product covering a specific area. To ensure spectral response performance, the instrument's optical system uses high-transmittance ultraviolet optical elements coated with an ultraviolet anti-reflection film, which effectively improves the utilization rate of light energy in the ultraviolet band while suppressing out-of-band stray light interference.

[0004] However, due to the non-uniformity of the pixel response of the area array CCD detector, random stripes are commonly found in the nadir observation data of this detector—that is, irregular bright and dark stripes formed by the superposition of instantaneous noise. Such stripe noise will cover the details of the characteristic absorption bands of ozone and trace gases.

[0005] Current preprocessing methods for this type of data often suffer from several drawbacks. Traditional median filtering and mean filtering methods tend to over-smooth the spectral details of hyperspectral data, destroying the characteristic absorption peak shapes of ozone and trace gases, leading to a decrease in subsequent inversion accuracy. Wavelet transform denoising methods require manual adjustment of the decomposition level and threshold, making it difficult to match the specificity of different gas absorption bands within the 250-495nm band. Single flat-field correction methods can only partially suppress fixed bands, with limited effectiveness in removing random bands, and they are not adapted to the response drift characteristics of the detectors during FY-3F satellite operation, failing to meet the high data quality requirements of operational observations.

[0006] Existing data processing systems struggle to balance the need for strip noise removal and spectral feature preservation when processing nadir data from the FY-3F satellite's ultraviolet hyperspectral ozone detector. This results in residual stripes in the preprocessed data, directly impacting the spatial consistency and numerical accuracy of atmospheric composition inversion results.

[0007] Currently, in the L1 level data preprocessing stage, the mainstream methods for processing strip noise in imaging spectrometers include traditional filtering, flat-field correction, and wavelet transform. Traditional methods such as median filtering and mean filtering achieve noise suppression through spatial domain smoothing, but their effectiveness is limited for wavelength-dependent stripes caused by the SMILE effect. Flat-field correction is only suitable for correcting fixed stripes and is not effective in suppressing random stripes. Wavelet transform requires manual setting of parameters such as the number of decomposition layers and thresholds, resulting in poor adaptability and difficulty in matching the data complexity of ultra-wide-angle imaging spectrometers. Existing methods generally suffer from drawbacks such as difficulty in simultaneously achieving noise suppression and spectral feature preservation, and insufficient adaptability to the SMILE effect, making it difficult to achieve accurate and efficient strip correction for FY-3F satellite ultraviolet hyperspectral ozone detector-nadir (FY-3FOMS-N) data. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies, such as the difficulty in simultaneously achieving noise suppression and spectral feature preservation, and insufficient adaptability to the SMILE effect of the FY-3F satellite, thereby providing a method and system for preprocessing and destripping of satellite ultraviolet hyperspectral ozone detector-nadir data.

[0009] To solve the above-mentioned technical problems, the present invention provides a method for preprocessing and removing banding from satellite ultraviolet hyperspectral ozone detector-nadir data, the steps of which include: Read the raw L0 level data acquired by the detector and extract the raw DN values ​​to form a three-dimensional data matrix. Use the dark current distribution matrix to correct the raw DN values ​​by band to obtain the corrected three-dimensional data matrix. Spatial column pixels of the corrected three-dimensional data matrix are extracted according to wavelength to obtain a spatial two-dimensional matrix for each wavelength. Window filtering is then performed on the spatial two-dimensional matrix only for the spatial dimension. The steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset, and the first-round smoothing processing is performed on the first-round processing subset to obtain the first-round smoothed spatial matrix; based on the first-round smoothed spatial matrix, the spatial pixel subset in the spatial pixel sequence that did not participate in the first-round smoothing processing is subjected to iterative smoothing processing. The first row boundary pixels in the second round of processing subset are adaptively supplemented by the mean of the subsequent effective pixels until the strip noise optimization in the full spatial dimension is completed, and the DN value after strip filtering is obtained; the second round of processing subset is the subset of remaining pixels that did not participate in the first round of filtering; Based on preset radiometric calibration coefficients, the DN values ​​after striping and filtering are converted into radiance to obtain a radiance filtering matrix; the radiance filtering matrix is ​​then concatenated in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and the reconstructed three-dimensional data matrix is ​​integrated with geolocation data and quality identifiers to generate an L1 level data file.

[0010] According to the data processing method of the present invention, the bands for window filtering of the spatial dimension include: UV1 band and UV2 / VIS band.

[0011] According to the data processing method of the present invention, the spatial pixels of the UV1 band are subjected to adaptive filtering according to spatial rows, and the spatial pixels of the UV2 / VIS band are subjected to filtering according to spatial rows.

[0012] According to the data processing method of the present invention, the correction of the original DN value as dark background correction includes: using the acquired ground calibration dark background data to obtain a dark current distribution matrix consistent with the observation conditions to correct the original DN value, the formula of which is: ; in, For spectral dimension index, For spatial cell indexing, For frame number, The grayscale value after removing the background. The original grayscale value. The dark background count distribution during ground calibration.

[0013] According to the data processing method of the present invention, the dark background correction is based on the difference in the number of pixels in different bands of OMS-N. The dark current of the UV1 band is calculated according to spatial pixels, and the dark current of the UV2 / VIS band is calculated according to spatial pixels.

[0014] According to the data processing method of the present invention, the first-round smoothing process on the first-round processing subset includes: For regular pixels in the first-round processing subset, a local smoothing window is constructed, and strip noise is suppressed by neighborhood mean operation; For the last boundary pixels in the first-round processing subset, the mean of the previous valid pixels is used for adaptive supplementation to generate the spatial matrix after the first round of smoothing.

[0015] According to the data processing method of the present invention, the geographic positioning data includes: latitude and longitude and solar zenith angle calculated from attitude data.

[0016] According to the data processing method of the present invention, the spatial two-dimensional matrix is ​​represented as follows: ,in, Corresponding spatial pixels, Corresponding spectral dimension.

[0017] This invention also proposes a preprocessing and destriping system for satellite ultraviolet hyperspectral ozone detector-nadir data, used to perform any of the methods described above, including: The raw data acquisition module is used to acquire L0 level raw data; Data reading and basic correction are used to extract the original DN value, form a three-dimensional data matrix, and correct the original DN value by band using the dark current distribution matrix; The spatial column pixel extraction module is used to extract the spatial column pixels of the three-dimensional data matrix according to each wavelength to obtain a spatial two-dimensional matrix under each wavelength; The adaptive neighborhood filtering module performs window filtering on the spatial two-dimensional matrix only in terms of the spatial dimension. Its steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset. The first-round processing subset is then subjected to a first-round smoothing process to obtain a smoothed spatial matrix. Based on the smoothed spatial matrix, the spatial pixel subsets in the spatial pixel sequence that did not participate in the first-round smoothing process are subjected to iterative smoothing. The first-row boundary pixels in the second-round processing subset are adaptively supplemented by the mean of the subsequent effective pixels until the strip noise optimization of the entire spatial dimension is completed, and the DN value after strip filtering is obtained. The data integration module is used to convert the destriped DN values ​​into radiance based on preset radiometric calibration coefficients, obtaining a radiance filtering matrix; and to concatenate the radiance filtering matrix in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and The L1-level data file generation module is used to integrate the reconstructed 3D data matrix with geolocation data and quality identifiers to generate an L1-level data file.

[0018] The advantages of this invention are: 1. Dark background subtraction adaptation design: To address the differences in the number of pixels in different bands of OMS-N, the dark current is calculated based on 58 spatial pixels for the UV1 band and 238 spatial pixels for the UV2 / VIS band, avoiding correction deviations caused by abnormalities in single dark pixels. 2. Spatial dimension filtering band adaptation: The UV1 band (58 spatial pixels) performs adaptive filtering according to 58 rows and columns, and the UV2 / VIS band (238 spatial pixels) performs filtering according to 238 rows and columns, which fully matches the spatial resolution characteristics of each band and ensures that the filtering does not destroy spatial scale information. In summary, this invention can specifically solve the random striping problem of OMS-N nadir data, and is adapted to its large field-of-view SMILE effect and multi-band characteristics. The processed L1 data can improve the inversion accuracy and meet the operational needs of atmospheric composition inversion.

[0019] This invention takes DN data obtained from nadir observation using a satellite ultraviolet hyperspectral ozone detector as the object and proposes a strip correction method adapted to the ultra-wide-angle SMILE effect, including: ① designing a wavelength-by-wavelength spatial column pixel extraction strategy to avoid the influence of the SMILE effect; ② constructing a spatial dimension fractional filtering algorithm to effectively suppress random strip noise. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method proposed in this invention; Detailed Implementation

[0021] The technical solutions provided by the present invention will be further illustrated below with reference to the embodiments.

[0022] Example 1 The following description uses the FY-3F satellite ultraviolet hyperspectral ozone detector-nadir (OMS-N detector) as an example to illustrate this embodiment.

[0023] To address the issue of random banding caused by the non-uniformity of detector pixel response in nadir observation data from the OMS-N detector (covering three bands: UV1: 250-300nm, UV2: 300-320nm, and VIS: 310-493nm, with spatial resolutions of 21km×28km, 7km×7km, and 7km×7km, respectively), a complete workflow design of "basic preprocessing - wavelength-wise spatial dimension adaptive filtering - L1 data generation" was adopted. This approach removes random banding while retaining the characteristic absorption spectral information of gases such as ozone, SO2, and NO2, ultimately generating high-quality L1-level radiation spectral data.

[0024] like Figure 1 As shown, the core method of this invention includes the following steps: Step 1: Data Reading and Basic Correction (Dark Background Subtraction) Decode the L0 level raw data according to the OMS-N data format, extract the raw DN values ​​of 58 spatial pixels in the UV1 band and 238 spatial pixels in the UV2 / VIS band, and form the "spectral dimension ( ) × spatial pixel dimension ( ) × number of frames ( ) three-dimensional data matrix; In this embodiment, the FY-3F satellite ultraviolet hyperspectral ozone detector-nadir has two detectors and three spectral channels. The detector is an image acquisition device. UV1 and UV2 share one detector, but the number of spatial row direction merging is different. Therefore, when extracting DN, it is necessary to extract the spatial distribution of UV1, UV2 and VIS separately.

[0025] Dark current correction (dark background subtraction): A dark current distribution matrix, "spectral dimension," consistent with observation conditions, is obtained using ground-calibrated dark background data acquired by OMS-N. ) × spatial pixel dimension ( The original DN value is corrected according to the band, using the following formula:

[0026] in, For spectral dimension index (corresponding to different wavelengths in the band range). Spatial pixel index (UV1: UV2 / VIS: ), The grayscale value after removing the background. The original grayscale value. The dark background count distribution during ground calibration has a spectral dimension length of 1024. Dark background subtraction adaptation design: To address the differences in the number of pixels in different bands of OMS-N, the dark current is calculated based on 58 spatial pixels for the UV1 band and 238 spatial pixels for the UV2 / VIS band, avoiding correction deviations caused by abnormalities in single dark pixels. Step 2: Extract spatial column pixels wavelength by wavelength (adapting to the large field of view SMILE effect) For the DN_S three-dimensional data matrix, according to each wavelength (Covering the entire band) Extract spatial column pixels from the entire field of view individually to obtain a two-dimensional spatial matrix at that wavelength. Among them, row index Corresponding spatial pixels, a total of 64 rows and column pixels Corresponding spectral dimension, UV1 band UV2 / VIS band .

[0027] The core purpose of this step is to address the large field-of-view SMILE effect of the OMS-N detector, which causes significant differences in the spatial position of the same wavelength in different rows. Wavelength-by-wavelength extraction can avoid positional deviations caused by the SMILE effect and ensure the accuracy of spatial destripping.

[0028] Step 3: Spatial Row Order Adaptive Neighborhood Filtering For each wavelength Two-dimensional matrix of space Window filtering is performed only in the spatial dimension and does not involve spectral dimension processing. The specific rules are as follows: First, a first-round grouping smoothing process is performed. Based on the inherent characteristics of the row order of spatial pixels, the entire spatial pixel sequence is adaptively divided into a first-round processing subset. For regular pixels in the first-round processing subset, a three-point local smoothing window is constructed using the spatial correlation of their preceding and following pixels. The strip noise is initially suppressed by calculating the neighborhood mean. For the last boundary pixels in the first-round processing subset, since there are no subsequent effective neighboring pixels to support them, the mean of the preceding effective pixels is used for adaptive supplementation to ensure the continuity of the signal trend in the boundary region, thereby generating the spatial matrix after the first round of smoothing.

[0029] Then, a second round of iterative smoothing is performed. Based on the spatial matrix after the first round of smoothing, a second adaptive smoothing is performed on the remaining pixel subset in the spatial pixel sequence that did not participate in the first round of filtering. For the first row boundary pixel in the second round of processing subset, since there are no preceding effective neighbor pixels, the mean of the subsequent effective pixels is used for adaptive supplementation to ensure that there is no omission or distortion in the entire spatial dimension, and finally the strip noise optimization in the entire spatial dimension is completed.

[0030] Spatial dimension filtering band adaptation: UV1 band (58 spatial pixels) performs adaptive filtering according to 58 spatial rows, and UV2 / VIS band (238 spatial pixels) performs filtering according to 238 spatial rows, which fully matches the spatial resolution characteristics of each band and ensures that filtering does not destroy spatial scale information. Step 4: Data Integration and L1-Level Product Generation Radiometric calibration: Use the radiometric calibration coefficients in the auxiliary data to convert the corrected DN value into radiance.

[0031] For all wavelengths Corresponding radiance filtering matrix By reversing the spectral dimensions, a three-dimensional data matrix of "spectral dimension × spatial pixel dimension × radiance" is reconstructed; The reconstructed 3D data matrix R is integrated with geolocation data (latitude and longitude, solar zenith angle, calculated from attitude data, which is positioning data provided by the satellite platform) and quality labels to generate an L1-level data file in HDF format.

[0032] Through the above steps, this invention can specifically solve the random striping problem of OMS-N nadir data, and adapt to its large field-of-view SMILE effect and multi-band characteristics. The processed L1 data can improve the inversion accuracy and meet the operational needs of atmospheric composition inversion.

[0033] In this embodiment, through dark background subtraction adaptation design, considering the difference in the number of pixels in different bands of OMS-N, the dark current is calculated based on 58 spatial pixels for the UV1 band and 238 spatial pixels for the UV2 / VIS band, avoiding correction deviations caused by single dark pixel anomalies; and Spatial dimension filtering band adaptation: UV1 band (58 spatial pixels) performs adaptive filtering according to 58 spatial rows, and UV2 / VIS band (238 spatial pixels) performs filtering according to 238 spatial rows, which fully matches the spatial resolution characteristics of each band and ensures that filtering does not destroy spatial scale information. Through the above steps, this invention can specifically solve the random striping problem of OMS-N nadir data, and adapt to its large field-of-view SMILE effect and multi-band characteristics. The processed L1 data can improve the inversion accuracy and meet the operational needs of atmospheric composition inversion.

[0034] Example 2 A method for preprocessing and removing banding from satellite ultraviolet hyperspectral ozone detector-nadir data includes the following steps: Read the raw L0 level data acquired by the detector and extract the raw DN values ​​to form a three-dimensional data matrix. Use the dark current distribution matrix to correct the raw DN values ​​by band to obtain the corrected three-dimensional data matrix. Spatial column pixels of the corrected three-dimensional data matrix are extracted according to wavelength to obtain a spatial two-dimensional matrix for each wavelength. Window filtering is then performed on the spatial two-dimensional matrix only for the spatial dimension. The steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset, and the first-round smoothing processing is performed on the first-round processing subset to obtain the first-round smoothed spatial matrix; based on the first-round smoothed spatial matrix, the spatial pixel subset in the spatial pixel sequence that did not participate in the first-round smoothing processing is subjected to iterative smoothing processing. The first row boundary pixels in the second round of processing subset are adaptively supplemented by the mean of the subsequent effective pixels until the strip noise optimization in the full spatial dimension is completed, and the DN value after strip filtering is obtained; the second round of processing subset is the subset of remaining pixels that did not participate in the first round of filtering; Based on preset radiometric calibration coefficients, the DN values ​​after striping and filtering are converted into radiance to obtain a radiance filtering matrix; the radiance filtering matrix is ​​then concatenated in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and the reconstructed three-dimensional data matrix is ​​integrated with geolocation data and quality identifiers to generate an L1 level data file.

[0035] According to the data processing method of the present invention, the bands for window filtering of the spatial dimension include: UV1 band and UV2 / VIS band.

[0036] According to the data processing method of the present invention, the 58 spatial pixels of the UV1 band are subjected to adaptive filtering in 58 spatial rows, and the 238 spatial pixels of the UV2 / VIS band are subjected to filtering in 238 spatial rows.

[0037] According to the data processing method of the present invention, the correction of the original DN value as dark background correction includes: using the acquired ground calibration dark background data to obtain a dark current distribution matrix consistent with the observation conditions to correct the original DN value, the formula of which is: ; in, For spectral dimension index, For spatial cell indexing, For frame number, The grayscale value after removing the background. The original grayscale value. The dark background count distribution during ground calibration.

[0038] According to the data processing method of the present invention, the dark background correction is based on the difference in the number of pixels in different bands of OMS-N. The dark current is calculated based on 58 spatial pixels for the UV1 band and 238 spatial pixels for the UV2 / VIS band.

[0039] According to the data processing method of the present invention, the first-round smoothing process on the first-round processing subset includes: For regular pixels in the first-round processing subset, a local smoothing window is constructed, and strip noise is suppressed by neighborhood mean operation; For the last boundary pixels in the first-round processing subset, the mean of the previous valid pixels is used for adaptive supplementation to generate the spatial matrix after the first round of smoothing.

[0040] According to the data processing method of the present invention, the geographic positioning data includes: latitude and longitude and solar zenith angle calculated from attitude data.

[0041] According to the data processing method of the present invention, the spatial two-dimensional matrix is ​​represented as follows: ,in, Corresponding spatial pixels, Corresponding spectral dimension, UV1 band UV2 / VIS band .

[0042] Example 3 A satellite ultraviolet hyperspectral ozone detector-nadir data preprocessing and destriping system, used to execute the method described in any of the above embodiments, is characterized by comprising: The raw data acquisition module is used to acquire L0 level raw data; Data reading and basic correction are used to extract the original DN value, form a three-dimensional data matrix, and correct the original DN value by band using the dark current distribution matrix; The spatial column pixel extraction module is used to extract the spatial column pixels of the three-dimensional data matrix according to each wavelength to obtain a spatial two-dimensional matrix under each wavelength; The adaptive neighborhood filtering module performs window filtering on the spatial two-dimensional matrix only in terms of the spatial dimension. Its steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset. A first-round smoothing process is then performed on this subset to obtain a smoothed spatial matrix. Based on this smoothed spatial matrix, iterative smoothing is performed on the spatial pixel subset that did not participate in the first-round smoothing process. The first-row boundary pixels in the second-round processing subset are adaptively supplemented using the mean of subsequent effective pixels until stripe noise optimization across the entire spatial dimension is completed, yielding the destriped DN value. The second-round processing subset is the remaining pixel subset that did not participate in the first-round filtering. The data integration module is used to convert the destriped DN values ​​into radiance based on preset radiometric calibration coefficients, obtaining a radiance filtering matrix; and to concatenate the radiance filtering matrix in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and The L1-level data file generation module is used to integrate the reconstructed 3D data matrix with geolocation data and quality identifiers to generate an L1-level data file.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for preprocessing and removing banding from satellite ultraviolet hyperspectral ozone detector-nadir data, comprising the following steps: Read the raw L0 level data acquired by the detector and extract the raw DN values ​​to form a three-dimensional data matrix. Use the dark current distribution matrix to correct the raw DN values ​​by band to obtain the corrected three-dimensional data matrix. Spatial column pixels of the corrected three-dimensional data matrix are extracted according to wavelength to obtain a spatial two-dimensional matrix for each wavelength. Window filtering is then performed on the spatial two-dimensional matrix only for the spatial dimension. The steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset, and the first-round smoothing processing is performed on the first-round processing subset to obtain the first-round smoothed spatial matrix; based on the first-round smoothed spatial matrix, the spatial pixel subset in the spatial pixel sequence that did not participate in the first-round smoothing processing is subjected to iterative smoothing processing. The first-round smoothing process for the first-round processing subset includes: For regular pixels in the first-round processing subset, a local smoothing window is constructed, and strip noise is suppressed by neighborhood mean operation; For the last boundary pixels in the first-round processing subset, the mean of the previous effective pixels is used for adaptive supplementation to generate the first-round smoothed spatial matrix; The first row boundary pixels in the second round of processing subset are adaptively supplemented by the mean of the subsequent effective pixels until the strip noise optimization in the full spatial dimension is completed, and the DN value after strip filtering is obtained; the second round of processing subset is the subset of remaining pixels that did not participate in the first round of filtering; Based on preset radiometric calibration coefficients, the DN values ​​after striping and filtering are converted into radiance to obtain a radiance filtering matrix; the radiance filtering matrix is ​​then concatenated in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and the reconstructed three-dimensional data matrix is ​​integrated with geolocation data and quality identifiers to generate an L1 level data file.

2. The data preprocessing destriping method according to claim 1, characterized in that, The bands for the window filtering include: UV1 band and UV2 / VIS band.

3. The data preprocessing destriping method according to claim 2, characterized in that, The spatial pixels of the UV1 band are subjected to adaptive filtering according to spatial rows, and the spatial pixels of the UV2 / VIS band are subjected to filtering according to spatial rows.

4. The data preprocessing destriping method according to claim 2, characterized in that, The correction of the original DN value is a dark background correction, which includes: using the acquired ground calibration dark background data to obtain a dark current distribution matrix consistent with the observation conditions to correct the original DN value, the formula of which is: ； in, For spectral dimension index, For spatial cell indexing, For frame number, The grayscale value after removing the background. The original grayscale value. The dark background count distribution during ground calibration.

5. The data preprocessing destriping method according to claim 4, characterized in that, The aforementioned dark background correction is performed to correct for the difference in the number of pixels in different bands of OMS-N. The dark current of the UV1 band is calculated based on spatial pixels, and the dark current of the UV2 / VIS band is calculated based on spatial pixels.

6. The data preprocessing destriping method according to claim 1, characterized in that, The spatial two-dimensional matrix is ​​represented as follows: ,in, Corresponding spatial pixels, Corresponding spectral dimension.

7. The data preprocessing destriping method according to claim 1, characterized in that, The geolocation data includes latitude and longitude and solar zenith angle calculated from the attitude data.

8. A preprocessing and destriping system for satellite ultraviolet hyperspectral ozone detector-nadir data, used to execute the method described in any one of claims 1-7, characterized in that, include: The raw data acquisition module is used to acquire L0 level raw data; Data reading and basic correction are used to extract the original DN value, form a three-dimensional data matrix, and correct the original DN value by band using the dark current distribution matrix; The spatial column pixel extraction module is used to extract the spatial column pixels of the three-dimensional data matrix according to each wavelength to obtain a spatial two-dimensional matrix under each wavelength; The adaptive neighborhood filtering module performs window filtering on the spatial two-dimensional matrix only in terms of the spatial dimension. Its steps include: Based on the inherent characteristics of spatial pixel row order, the spatial pixel sequence is adaptively divided into a first-round processing subset. A first-round smoothing process is performed on this subset to obtain a smoothed spatial matrix. Based on this smoothed spatial matrix, iterative smoothing is performed on the spatial pixel subset that did not participate in the first-round smoothing process. The first-round smoothing process on the first-round processing subset includes: constructing a local smoothing window for the regular pixels in the first-round processing subset and suppressing stripe noise through neighborhood mean calculation; adaptively supplementing the last boundary pixels in the first-round processing subset using the mean of the preceding valid pixels to generate a smoothed spatial matrix; adaptively supplementing the first-row boundary pixels in the second-round processing subset using the mean of the following valid pixels until stripe noise optimization across the entire spatial dimension is completed, obtaining the de-striped filtered DN value; the second-round processing subset is the remaining pixel subset that did not participate in the first-round filtering. The data integration module is used to convert the destriped DN values ​​into radiance based on preset radiometric calibration coefficients, obtaining a radiance filtering matrix; and to concatenate the radiance filtering matrix in reverse along the spectral dimension to obtain a reconstructed three-dimensional data matrix; and The L1-level data file generation module is used to integrate the reconstructed 3D data matrix with geolocation data and quality identifiers to generate an L1-level data file.

Images