Short-wave infrared hyperspectral data on-orbit blind pixel repairing method

Abstract

The invention discloses a short-wave infrared hyperspectral data in-orbit blind pixel restoration method. The method comprises the following steps: preprocessing satellite original data; partitioning the preprocessed data, and averaging each block of data along the track; aiming at a piece of along-track average data, carrying out spatial dimension Gaussian fitting on probe elements at the left side and the right side of the blind pixel position to obtain a fitting gray value at the blind pixel position; calculating the standard deviation of the gray values of nine probe elements in the sudoku range around the blind element position; corresponding standard deviations of the fitting gray values are calculated respectively, the fitting gray value with the minimum standard deviation is selected as a blind pixel initial restoration value, and the fitting gray value and the gray values of the surrounding eight probe elements form a gray value combination; and repeating the calculation on all block data to obtain a plurality of groups of gray value combinations, performing multiple linear regression by taking the blind pixel gray value as a dependent variable and taking the gray values of the surrounding eight detection elements as independent variables to obtain a regression coefficient and a residual error, and calculating the gray value at the blind pixel position. According to the method, blind pixel repair can be quickly and effectively realized.

Description

Technical Field

[0001] The invention belongs to the field of hyperspectral remote sensing image processing and relates to an on-orbit blind pixel repair method for on-orbit short-wave infrared hyperspectral data. Background Art

[0002] As an important technology in the field of earth observation, imaging spectroscopy can obtain high-resolution spectral information of surface objects while clearly imaging the surface. With the further development of aerospace remote sensing technology, satellite-borne hyperspectral imagers and their data processing technology have become one of the important research hotspots in the field of remote sensing.

[0003] my country's hyperspectral comprehensive observation satellite is equipped with a hyperspectral imager, which can obtain imaging spectral data in the visible and short-wave infrared bands. It is widely used in environmental monitoring, land and resources surveys, crop growth monitoring, water quality monitoring and other fields. Due to the casting process of detection components or environmental changes, there are blind pixels in the detector. The existence of blind pixels will greatly reduce the spatial resolution and temperature resolution of the imaging system, seriously inhibit the performance of the imaging system, and greatly affect the imaging quality. It is necessary to develop data processing methods for repair. The general method is to use median filtering, which has the disadvantages of poor stability, large loss of image information after repair, and inability to repair continuous blind pixels, which brings difficulties to subsequent radiation processing. Summary of the invention

[0004] The technical problem solved by the present invention is to overcome the deficiencies of the prior art, provide an on-orbit blind pixel repair method for on-orbit short-wave infrared hyperspectral data, and solve the problem of radiation defects in on-orbit satellite short-wave infrared hyperspectral imaging data.

[0005] The technical solution of the present invention is: an on-orbit blind pixel repair method for on-orbit short-wave infrared hyperspectral data, comprising:

[0006] Step 1, obtaining original level 0 scene segmentation data, parsing and preprocessing it, and obtaining preprocessed level 0 scene segmentation data;

[0007] Step 2, obtaining the blind pixel position according to the blind pixel distribution file;

[0008] Step 3, dividing the pre-processed level 0 scene data into blocks, and averaging each block of data along the track;

[0009] Step 4, select a block of data after averaging along the track, perform spatial Gaussian fitting on the detectors on the left and right sides of the blind pixel position, and obtain the fitting grayscale value DN0 at the blind pixel position;

[0010] Step 5, add a bias to the obtained fitting grayscale value DN0 in the interval [-P, P] with a step size of 1, calculate the standard deviation of the grayscale values ​​in the 3*3 nine-square grid around the blind pixel, select the fitting grayscale value with the smallest standard deviation as the initial repair value of the blind pixel, and form a group of grayscale value combinations with the remaining 8 detector grayscale values;

[0011] Step 6, for each block of track average data, repeat steps 4-5 to obtain multiple groups of gray value combinations;

[0012] Step 7: Perform multiple linear regression on the obtained multiple gray value combinations to obtain the regression coefficients and residuals, and then calculate the blind pixel repair value DN result .

[0013] The method of obtaining the original level 0 scene segmentation data, parsing and preprocessing the original level 0 scene segmentation data to obtain the preprocessed level 0 scene segmentation data includes:

[0014] According to the satellite-to-ground interface file, the original level 0 scene data is parsed to obtain image data and auxiliary data; then the time information in the auxiliary data is further extracted;

[0015] Obtain the dark current time-varying parameters of each detector of the short-wave infrared hyperspectral array, specifically including two time-varying parameters including slope a and cutoff b;

[0016] Based on the row-time information and time-varying parameters, the grayscale value of each detector after dark current deduction in each frame of data is calculated:

[0017] DN=DN 原始 -Time*ab

[0018] Where Time is the imaging time corresponding to each frame of image, DN 原始 is the detector grayscale value before deducting the dark current, and DN is the detector grayscale value after deducting the dark current, that is, the 0-level scene data after preprocessing.

[0019] When DN<0, set DN to 0.

[0020] The method selects a block of block data after averaging along the track, performs spatial Gaussian fitting on the detectors on the left and right sides of the blind pixel position, and obtains the fitting grayscale value DN0 at the blind pixel position, including:

[0021] Calculate the mean and standard deviation of the data in the same row where the blind pixel is located, filter the grayscale value of the detector in this row, that is, remove the singular value, and obtain the grayscale value of the detector after removing the singular value;

[0022] The Gaussian model is used to fit the grayscale values ​​of each detector after removing the singular value, and the fitting grayscale value DN0 at the blind pixel position is calculated based on the fitting coefficient.

[0023] The screening criteria for the gray value of the detector are:

[0024] |DNj-Mean|>3*STD

[0025] Among them, DNj is the gray value of the jth detector in this row after removing the singular value.

[0026] The fitting gray value DN0 at the blind pixel position is calculated based on the fitting coefficient:

[0027] DN0=a0*e^(-z*z / 2)+a3+a4*x+a5*x*x

[0028] z=(x-a1) / a2

[0029] Among them, a0, a1, a2, a3, a4, a5 are fitting coefficients, and x is the detector column number of the blind pixel position.

[0030] The repair value DN of the blind pixel result =F(DN1, DN2, ...DNi, ...DN8); where DN result is the restoration value of the blind pixel; F represents the regression coefficient; DNi represents the gray value of one of the eight adjacent pixels around the blind pixel.

[0031] Said P=3-5.

[0032] The advantages of the present invention compared with the prior art are:

[0033] 1. The present invention fully considers the response relationship between adjacent detectors in the spatial dimension and the spectral dimension, and performs unified modeling, thereby overcoming the unstable problems such as blind pixel information jump after repair caused by traditional repair methods.

[0034] 2. The present invention is universally applicable because it is trained through multiple along-track average data and can effectively repair the continuous blind pixel problem.

[0035] 3. The present invention takes into account the response relationship of adjacent detectors in the spatial dimension, and selects model coefficients with smaller differences in spatial-spectral grayscale values ​​through grayscale equal-step iteration method, which can effectively restore imaging information and maintain a good amount of image information after restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of the blind pixel repair method of the present invention.

[0037] Figure 2 Divide the data into blocks and obtain the average schematic diagram along the track.

[0038] Figure 3 This is a schematic diagram of the GF5 01A satellite shortwave infrared sensor array.

[0039] Figure 4 This is the blind element parameter file for the shortwave infrared sensor array of the GF5 01A satellite. DETAILED DESCRIPTION

[0040] The specific implementation is described by taking the GF501A short-wave infrared hyperspectral sensor as an example.

[0041] The detector array of GF501A short-wave infrared hyperspectral sensor consists of 4 modules, each module consists of 512 detectors in the spatial dimension and 180 detectors in the spectral dimension. The image size of one frame (push-scan line) of the sensor imaging the ground is 2048 columns * 180 rows. Figure 3 shown.

[0042] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings and examples. Figure 1 As shown, the process is as follows:

[0043] 1. Obtain original level 0 scene segmentation data, parse and preprocess the data, and obtain preprocessed level 0 scene segmentation data.

[0044] (1) First, the level 0 scene data is parsed to obtain image data and auxiliary data; then the line time information in the auxiliary data is further extracted.

[0045] (2) Read the dark current time-varying parameter file. Take the linear time-varying parameter file as an example. The file contains two time-varying parameters: slope a and cutoff b. Read a and b. Based on the row time information and time-varying parameters of each frame of data, calculate the gray value of each detector after deducting the dark current in each frame of data according to the following formula:

[0046] DN=DN 原始 -Time*ab

[0047] When DN<0, set DN to 0.

[0048] Where Time is the imaging time corresponding to each frame of image, DN 原始 is the detector grayscale value before deducting the dark current, and DN is the detector grayscale value after deducting the dark current.

[0049] 2. Obtain the blind pixel distribution file and get the blind pixel position, such as Figure 4 As shown in the figure, each row represents a blind pixel position, the first field is the blind pixel column number (spatial dimension), and the second field is the blind pixel row number (spectral dimension). The following steps take the repair of the fourth blind pixel as an example, and finally the gray value of the 120th column and 11th row of each frame of hyperspectral array detector data after repair is obtained: DN k(120，11)， Where k = 1, 2, 3...M (M is the frame number).

[0050] 3. Such as Figure 2 As shown, the pre-processed 0-level scene data is divided into N blocks along the track direction, and each block of data is averaged along the track to obtain the grayscale values ​​of 8 detectors in the averaged nine-square grid range.

[0051] 4. For each piece of data, perform spatial Gaussian fitting on the detectors on the left and right sides of the blind pixel position to obtain the fitted grayscale value at the blind pixel position. Finally, adjust the fitted grayscale value by adjusting the step size to obtain the final fitted grayscale value at the blind pixel position. The method is as follows:

[0052] (1) First, calculate the mean and standard deviation of the data in the same row where the blind pixel is located, and then filter the grayscale values ​​of the pixel in the row to remove the singular values. The standard is:

[0053] |DNj-Mean|>3*STD

[0054] Among them, DNj is the gray value of the jth detector in this row after removing the singular value.

[0055] (2) Using a cubic polynomial model, perform Gaussian fitting on the grayscale values ​​of each detector after removing the singular values ​​in step (1), and calculate the fitting grayscale value at the blind pixel position of the block of data based on the fitting coefficient.

[0056] DN' i(120，11) =a0*e^(-z*z / 2)+a3+a4*x+a5*x*x

[0057] z=(x-a1) / a2

[0058] Among them, a0, a1, a2, a3, a4, a5 are fitting coefficients, and x is the detector column number of the blind pixel position.

[0059] (3) The gray value DN' at the blind pixel obtained in step (2) i(120，11) In the interval [-3, +3], the bias is increased in steps of 1 to obtain the grayscale values ​​of the seven nine-square grid ranges, and then the standard deviation STDj of the grayscale values ​​of the nine detectors in each nine-square grid range is calculated, j = 1, 2, ..7. The grayscale value of the blind pixel corresponding to the minimum standard deviation is selected as the grayscale value DN of the initial repair of the blind pixel of the block of data i(120，11) , DN' i(120，11) Together with the grayscale values ​​of the eight detectors around the nine-square grid, a set of grayscale value combinations is constructed, namely {DN i(120，11) , DNi1, DNi2, DNi3, DNi4, DNi5, DNi6, DNi7, DNi8}.

[0060] 5. Repeat step 4 to obtain N groups of gray value combinations: {DN i(120，11), DNi1, DNi2, DNi3, DNi4, DNi5, DNi6, DNi7, DNi8}, i=1, 2, 3...N

[0061] 6. Perform multiple linear regression on the multiple gray value combinations obtained in step 5 to obtain the regression coefficients and residuals. The regression formula is as follows:

[0062] DN0=F(DN1,DN2,DN3,DN4,DN5,DN6,DN7,DN8)

[0063] Among them, DN0 corresponds to the grayscale value at the blind pixel, DN1, DN2, DN3, DN4, DN5, DN6, DN7, and DN8 correspond to the grayscale values ​​of each detector in the nine-square grid around the blind pixel, and F is the regression coefficient.

[0064] 7. According to the regression coefficient, the blind pixels of each frame of the pre-processed 0-level scene data are repaired:

[0065] DN k(120，11) =F(DNK k 1.DNK k 2.DNK k 3.DNK k 4.DNK k 5.DNK k 6.DNK k 7.DNK k 8)

[0066] Where DN k 1.DN k 2.DN k 3.DN k 4.DN k 5.DN k 6.DN k 7.DN k 8 are the grayscale values ​​of the detectors in the nine-square grid around the detector in the 120th column and the 11th row of the hyperspectral array detector data of the kth frame.

[0067] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention by using the technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for on-orbit blind pixel repair of on-orbit shortwave infrared hyperspectral data, characterized in that: include: Step 1, obtaining original level 0 scene segmentation data, parsing and preprocessing it, and obtaining preprocessed level 0 scene segmentation data; Step 2, obtaining the blind pixel position according to the blind pixel distribution file; Step 3, dividing the pre-processed level 0 scene data into blocks, and averaging each block of data along the track; Step 4, select a block of data after averaging along the track, perform spatial Gaussian fitting on the detectors on the left and right sides of the blind pixel position, and obtain the fitting grayscale value DN0 at the blind pixel position; Step 5, add a bias to the obtained fitting grayscale value DN0 in the interval [-P, P] with a step size of 1, calculate the standard deviation of the grayscale value in the 3*3 nine-square grid around the blind pixel position, select the fitting grayscale value with the smallest standard deviation as the initial repair value of the blind pixel, and form a group of grayscale value combinations with the remaining 8 detector grayscale values; Step 6, for each block of track average data, repeat steps 4-5 to obtain multiple groups of gray value combinations; Step 7: Perform multiple linear regression on the obtained multiple gray value combinations to obtain the regression coefficients and residuals, and then calculate the blind pixel repair value DN result .

2. The on-orbit blind pixel repair method for on-orbit shortwave infrared hyperspectral data according to claim 1, characterized in that: The method of obtaining the original level 0 scene segmentation data, parsing and preprocessing the original level 0 scene segmentation data to obtain the preprocessed level 0 scene segmentation data includes: According to the satellite-to-ground interface file, the original level 0 scene data is parsed to obtain image data and auxiliary data; then the time information in the auxiliary data is further extracted; Obtain the dark current time-varying parameters of each detector of the short-wave infrared hyperspectral array, specifically including two time-varying parameters including slope a and cutoff b; Based on the row-time information and time-varying parameters, the grayscale value of each detector after dark current deduction in each frame of data is calculated: DN=DN 原始 -Time*a-b Where Time is the imaging time corresponding to each frame of image, DN 原始 is the detector grayscale value before deducting the dark current, and DN is the detector grayscale value after deducting the dark current, that is, the 0-level scene data after preprocessing.

3. The on-orbit blind pixel repair method for on-orbit shortwave infrared hyperspectral data according to claim 2 is characterized in that: When DN<0, set DN to 0.

4. The on-track blind pixel repair method for shortwave infrared hyperspectral data according to claim 1 is characterized in that: The method selects a block of block data after averaging along the track, performs spatial Gaussian fitting on the detectors on the left and right sides of the blind pixel position, and obtains the fitting grayscale value DN0 at the blind pixel position, including: Calculate the mean and standard deviation of the data in the same row where the blind pixel is located, filter the grayscale value of the detector in this row, that is, remove the singular value, and obtain the grayscale value of the detector after removing the singular value; The Gaussian model is used to fit the grayscale values ​​of each detector after removing the singular value, and the fitting grayscale value DN0 at the blind pixel position is calculated based on the fitting coefficient.

5. The on-track blind pixel repair method for shortwave infrared hyperspectral data according to claim 4 is characterized in that: The screening criteria for the gray value of the detector are: |DNj-Mean|>3*STD Among them, DNj is the gray value of the jth detector in this row after removing the singular value.

6. The on-track blind pixel repair method for shortwave infrared hyperspectral data according to claim 5 is characterized in that: The fitting gray value DN0 at the blind pixel position is calculated based on the fitting coefficient: DN0=a0*e^(-z*z / 2)+a3+a4*x+a5*x*x z=(x-a1) / a2 Among them, a0, a1, a2, a3, a4, a5 are fitting coefficients, and x is the detector column number of the blind pixel position.

7. The on-track blind pixel repair method for shortwave infrared hyperspectral data according to claim 1, characterized in that: The repair value DN of the blind pixel result =F(DN1, DN2, ...DNi, ...DN8); where DN result is the restoration value of the blind pixel; F represents the regression coefficient; DNi represents the gray value of one of the eight adjacent pixels around the blind pixel.

8. The on-track blind pixel repair method for shortwave infrared hyperspectral data according to claim 1, characterized in that: Said P=3-5.

Images