A quantile-based remote sensing image relative radiometric calibration method and system

Abstract

The application discloses a kind of relative radiation calibration methods and systems of remote sensing image based on quantile, it is related to remote sensing image processing technical field, the method includes: collecting satellite historical image data and separating according to spectral band;The value of each column of single-band image detector is sequentially arranged in ascending order, and quantile value is uniformly extracted to generate the quantile feature sequence of each column;The arithmetic mean value of all columns under the same quantile level is calculated, and the global reference quantile distribution is constructed;Based on least square method, the linear mapping relationship between each column quantile sequence and reference distribution is constructed, and the radiation correction gain and bias coefficient of each detector column are obtained;The original image is corrected column by column using the coefficient;The present application effectively eliminates the vertical strip noise caused by the non-uniformity of sensor response, overcomes the limitations of traditional calibration under the meteorological conditions and platform maneuvering, significantly improves the timeliness and image radiation quality of high-frequency and high-precision calibration.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a method and system for relative radiometric calibration of remote sensing images based on quantiles. Background Technology

[0002] With the rapid development of Earth observation technology, hyperspectral remote sensing technology has been widely used due to its ability to provide rich and continuous narrow-band spectral information. To acquire sufficient signals under narrow-band, low-entry-pupil radiation energy conditions, hyperspectral cameras widely employ linear pushbroom imaging mechanisms. However, due to objective factors such as slight differences in detector manufacturing processes, bias variations in electronic links, and differences in dark current, the output responses of different detector elements in the sensor exhibit significant inconsistencies when receiving the same radiation. This inconsistency manifests as vertical stripes or fringe noise across the entire image, severely affecting the visual quality of the image and subsequent advanced quantitative applications. Therefore, on-orbit relative radiometric calibration has become a crucial step in eliminating systematic errors and ensuring the radiometric quality of the image.

[0003] Existing technologies for relative calibration of in-orbit optical remote sensing satellites have significant limitations: uniform field calibration is constrained by geographical and meteorological conditions, making it difficult to conduct at high frequencies; yaw calibration relies on platform maneuverability and encroaches on operational observation time, affecting operational efficiency. To overcome these limitations, statistical calibration based on historical data has been widely used. However, when faced with the requirements of high-frequency, high-precision calibration, this method is susceptible to interference from outliers such as clouds and is extremely dependent on the accumulation of massive samples over long periods, severely restricting the timeliness of calibration. Summary of the Invention

[0004] To address the above technical problems, this invention provides a method for relative radiometric calibration of remote sensing images based on quantiles, comprising the following steps: S1. Perform initial preprocessing on hyperspectral remote sensing images, collect historical image data acquired by the satellite in orbit, and separate and store these raw images according to different spectral bands. S2. Acquire the preprocessed single-band image. Within the limited effective radiation range, sort the pixel values ​​of each column of the detector in a monotonically increasing manner. Then, uniformly extract a specified number of quantile values ​​from the sorted sequence to generate the quantile feature sequence of each detector column. S3. For the quantile feature sequence, calculate the arithmetic mean of all detector columns at each same quantile level, and use the arithmetic mean as the target reference quantile at that quantile level, thereby constructing a global reference quantile distribution. S4. Based on the principle of least squares, calculate the covariance and variance between the quantiles extracted by each detector column and the reference quantiles, thereby constructing a linear mapping relationship, and then obtaining the radiation correction gain and bias coefficient corresponding to each detector column. S5. Using the radiometric correction gain and bias coefficient corresponding to each detector column, perform relative radiometric correction processing on the original image data column by column, and finally output a hyperspectral image that has been corrected and whose vertical strip noise of the detector has been eliminated.

[0005] The technical solution further defined in this invention is: Furthermore, step S1 specifically includes the following sub-steps: S1.1 Collect historical image data acquired by the satellite in orbit, screen and remove various types of abnormal images. Abnormal images include those that are completely invalid or have severely damaged data packets. Abnormal images with severely damaged data packets include those with a mean that exceeds the limit, an extreme value ratio that exceeds the standard, or a standard deviation that is lower than the threshold. S1.2 Separate the selected original images according to different spectral bands, and construct a two-dimensional image matrix for each spectral band that is stored independently.

[0006] As described above, the relative radiometric calibration method for remote sensing images based on quantiles includes the following sub-steps in step S2: S2.1 Calculate the total number of quantiles to be extracted based on the product of the number of images and the height dimension of the image, according to the preset quantile ratio coefficient and the height dimension of the image. S2.2. Generate a uniformly distributed quantile position sequence within the effective radiation range of 1-16384 of the 14-bit digital quantization value. S2.3. The image data of each column of the detector are sorted monotonically in ascending order, and the corresponding quantile values ​​are extracted by direct sampling according to the quantile position sequence to generate the quantile feature sequence of each detector column.

[0007] As described above, in a quantile-based remote sensing image relative radiometric calibration method, in step S2, the total number of quantiles is denoted as n; for the k-th quantile level, k=0,1,...,n-1, its corresponding cumulative probability position is... The calculation formula is The image data of each column of the detector is monotonically sorted in ascending order to obtain the one-dimensional sequence of the i-th column. Let i = 0, 1, ..., W-1, where W represents the number of columns in the image, i.e., the number of detector elements. Then, based on the quantile position sequence, the corresponding quantile values ​​are extracted through direct sampling to generate the quantile feature sequence for each detector column, i.e., the k-th quantile value of the i-th column is... H represents the number of rows in the image, i.e., the number of valid pixels. This indicates a round-down operation.

[0008] As described above, the relative radiometric calibration method for remote sensing images based on quantiles includes the following sub-steps in step S3: S3.1 Obtain the quantile feature sequences of all extracted detector columns, and align them horizontally according to the same quantile level to form a quantile feature matrix; S3.2 For each given quantile level k, extract the specific quantile values ​​of all detector columns at that level. This forms an independent one-dimensional array; S3.3 Calculate the arithmetic mean of all quantile values ​​in the one-dimensional array, that is, perform the mean operation in the row direction of the quantile matrix; S3.4. Use the calculated arithmetic mean as the target reference quantile for the corresponding quantile level, aggregate the target reference quantiles for all levels, and construct a global reference quantile distribution.

[0009] As described above, in a quantile-based remote sensing image relative radiometric calibration method, in step S3, the size of the quantile feature matrix is ​​n×W, where n represents the total number of quantiles and W represents the number of columns in the image, i.e., the number of detector elements; the target reference quantile is denoted as... The calculation formula is: , This represents the kth quantile value in the i-th column.

[0010] As described above, the relative radiometric calibration method for remote sensing images based on quantiles includes the following sub-steps in step S4: S4.1 For each specific detector column i, its extracted quantile feature sequence As the independent variable, This represents the k-th quantile value in the i-th column, where k represents the quantile position level and n represents the total number of quantiles; using the global reference quantile distribution sequence constructed in step S3. Using this as the dependent variable, a univariate linear regression model is established. Indicates the target reference quantile; S4.2. According to the principle of least squares, calculate the covariance between the independent variable and the dependent variable, as well as the variance of the independent variable, and use the ratio of the covariance to the variance as the radiation correction gain coefficient of the detector array. S4.3 Calculate the mean of the dependent variable and the mean of the independent variable. Subtract the product of the radiation correction gain coefficient and the mean of the independent variable from the mean of the dependent variable to obtain the radiation correction bias coefficient of the detector array, thereby completing the calculation and storage of the correction coefficients of all detector arrays.

[0011] As described above, the relative radiometric calibration method for remote sensing images based on quantiles includes the following sub-steps in step S5: S5.1 Read the original hyperspectral remote sensing image data to be corrected, and extract the original pixel gray values ​​of each column of the detector in the corresponding single-band image. S5.2 Multiply the pixel gray value of each column with the radiation correction gain coefficient corresponding to that column obtained in step S4 to complete the gain calibration of detector response difference. S5.3. Based on the numerical results of the gain calibration, add the corresponding radiometric correction bias coefficient for that column to output the relative radiometrically corrected image after eliminating vertical strip noise.

[0012] As described above, in the quantile-based remote sensing image relative radiometric calibration method, in step S5, the corrected pixel gray values ​​of each column of the detector are... The calculation formula is: , This represents the radiation correction gain coefficient. Represents the original pixel grayscale value. This represents the radiation correction bias coefficient.

[0013] A quantile-based remote sensing image relative radiometric calibration system includes: The preprocessing module is used to perform initial preprocessing on hyperspectral remote sensing images, collect historical image data acquired by the satellite in orbit, and separate and store these raw images according to different spectral bands. The quantile feature sequence generation module is used to acquire preprocessed single-band images. Within the limited effective radiation range, the pixel values ​​of each column of the detector are monotonically sorted in ascending order. Then, a specified number of quantile values ​​are uniformly extracted from the sorted sequence to generate the quantile feature sequence of each detector column. The reference quantile distribution construction module is used to calculate the arithmetic mean of all detector columns at each same quantile level for the quantile feature sequence, and use the arithmetic mean as the target reference quantile at that quantile level, thereby constructing a global reference quantile distribution. The correction coefficient calculation module is used to calculate the covariance and variance between the quantiles extracted by each detector column and the reference quantiles based on the least squares method, thereby constructing a linear mapping relationship and obtaining the radiation correction gain and bias coefficient corresponding to each detector column. The relative radiometric correction module is used to perform column-by-column relative radiometric correction on the original image data using the radiometric correction gain and bias coefficient corresponding to each detector column, and finally outputs a corrected hyperspectral image with the vertical strip noise of the detector eliminated.

[0014] The beneficial effects of this invention are: (1) This invention has significant advantages in noise resistance robustness and image quality improvement. From a technical perspective, traditional histogram statistical calibration methods rely heavily on the absolute frequency distribution of pixel gray values. When extreme outliers such as clouds (extremely bright), shadows, or bad pixels (extremely dark) exist in the image, these extreme values ​​will severely distort the cumulative probability density function. During histogram matching, this nonlinear distribution distortion will be forcibly transmitted to the correction coefficients, thereby introducing nonlinear mapping artifacts such as "dark stripes" in normal terrain areas. In contrast, this invention innovatively introduces the quantile statistical principle. The essence of quantiles is based on the spatial ordering position of the data rather than the absolute numerical frequency. This means that extreme outliers can significantly improve image quality. The brightness values ​​at the ends of the sequence will only be distributed at both ends, without interfering with the statistical characteristics of the core radiation region. This invention mathematically shields against interference from clouds and shadows, ensuring that all extracted feature points originate from valid observation targets. Therefore, the linear mapping relationship fitted based on effective quantiles is extremely robust, not only completely removing dense vertical stripes from the original image but also avoiding nonlinear artifacts caused by traditional methods. Quantitative test results based on real images show that this method significantly reduces the image stripe coefficient from 0.2339% to 0.0175% and the light response non-uniformity from 1.8964% to 0.7556%, achieving a leapfrog improvement in radiometric correction accuracy. (2) This invention greatly reduces the algorithm's dependence on massive long-term sample data. Since quantiles can be directly extracted from the sorted sequence at equal intervals, the number of feature points provided to the linear regression model is directly related to the sampling interval, rather than being constrained by the coverage of gray levels. Even when the sample data volume is small, based on nonparametric statistical theory, the quantile values ​​of the core interval can still maintain extremely high stability in approximating the real distribution. Experiments have shown that the method of this invention only requires 20% to 40% of the sample data volume required by traditional scenario statistics, and the extracted quantile features are sufficient to support the least squares method to obtain high-precision gain and bias coefficients. This technological breakthrough completely eliminates the stringent requirements of traditional methods for large-scale uniform terrain and long-term data, greatly alleviates the data acquisition and storage pressure during high-frequency on-orbit operational calibration, and greatly expands the applicability of the algorithm in complex business scenarios. (3) The present invention demonstrates great advantages in terms of computational efficiency and engineering deployment. The radiometric calibration algorithm proposed in this invention has an extremely simple and clear structure. In the feature extraction stage, only one-dimensional monotonic sorting operation is required for the effective pixels of each detector column, and the fixed-length quantile values ​​are directly extracted by index. In the coefficient calculation stage, only the least squares linear regression is completed through basic covariance and variance operations. This lightweight computational framework with free distribution not only greatly reduces the spatial and temporal complexity of the algorithm, but also makes it very easy to deploy engineering code and accelerate hardware in the limited onboard or ground preprocessing system of micro-nano hyperspectral satellites. The measured data shows that the calculation time of single-band calibration coefficients is greatly reduced from 16.6 seconds in the traditional histogram statistics to 10.5 seconds, and the computational efficiency has soared by 58.1%, which perfectly meets the stringent engineering requirements of high frequency, real-time / quasi-real-time on-orbit calibration of massive data of the new generation of hyperspectral remote sensing platform. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the original image before correction in an embodiment of the present invention; Figure 3 This is a schematic diagram of the corrected image in an embodiment of the present invention. Detailed Implementation

[0016] This embodiment provides a method for relative radiometric calibration of remote sensing images based on quantiles, such as... Figure 1 As shown, it includes the following steps: S1. This embodiment uses a 23-orbit image matrix acquired by a micro / nano hyperspectral satellite. For example (where H=327168 is the number of rows in the image and W=3072 is the number of columns in the image, i.e. the number of detectors), the hyperspectral remote sensing image is initially preprocessed, historical image data acquired by the satellite in orbit is collected, and these raw images are separated and stored independently according to different spectral bands to provide data support for subsequent calibration.

[0017] S1.1 Collect historical image data acquired by the satellite in orbit, and screen and remove a very small number of abnormal images of various types that are completely invalid or have severely damaged data packets (mean value exceeds the limit, i.e., less than 50 or greater than 16300; extreme value proportion exceeds the standard, i.e., DN value of 0 or 16383 accounts for more than 5% or 10%; standard deviation is too low, i.e. less than 10 or 5). The method of this embodiment has natural robustness to conventional clouds and shadows, so there is no need to perform complex cloud and snow masking processing.

[0018] S1.2 Separate the filtered original images according to different spectral bands, and construct a two-dimensional image matrix for each spectral band that is stored independently. This provides data support for subsequent independent relative radiometric calibration on a band-by-band basis.

[0019] S2. Acquire the preprocessed single-band image. Within the limited effective radiation range of 1-16384 for 14-bit digital quantization values, sort the pixel values ​​of each column of the detector in a monotonically increasing manner. Then, uniformly extract a specified number of quantile values ​​from the sorted sequence to generate the quantile feature sequence of each detector column to characterize the distribution characteristics of each column.

[0020] S2.1. Based on the preset quantile ratio coefficient and the height dimension of the image, calculate the total number of quantiles to be extracted, n = 23 × 327168, based on the product of the number of images and the height.

[0021] S2.2 Within the effective radiation range, calculate and generate a uniformly distributed sequence of quantile positions; for the k-th quantile level (k=0,1,...,n-1), its corresponding cumulative probability position... The calculation formula is .

[0022] S2.3. Sort the image data of each column (i.e., the i-th column, i=0,1,...,W-1) of the detector in a monotonically increasing order to obtain the one-dimensional sequence of the i-th column. Subsequently, based on the quantile position sequence, the corresponding quantile values ​​are extracted through direct sampling to generate the quantile feature sequence for each detector column, i.e., the k-th quantile value of the i-th column is... H represents the number of rows in the image, i.e., the number of valid pixels. This indicates a round-down operation.

[0023] S3. For the quantile feature sequence, calculate the arithmetic mean of all detector columns at each same quantile level, and use the arithmetic mean as the target reference quantile at that quantile level, thereby constructing the global reference quantile distribution.

[0024] S3.1 Obtain the quantile feature sequences of all extracted detector columns, and align them horizontally at the same quantile level to form a quantile feature matrix of size n×W.

[0025] S3.2 For each given quantile level k, extract the specific quantile values ​​of all detector columns at that level. This forms an independent one-dimensional array.

[0026] S3.3 Calculate the arithmetic mean of all quantile values ​​in the one-dimensional array (i.e., perform the mean operation in the row direction of the quantile matrix).

[0027] S3.4. Use the calculated arithmetic mean as the target reference quantile for the corresponding quantile level. The calculation formula is: Aggregate all levels (k=0,...,n-1), constructing a standardized global reference quantile distribution.

[0028] S4. Based on the principle of least squares, calculate the covariance and variance between the quantiles extracted from each detector array and the reference quantiles, thereby constructing a linear mapping relationship, and then accurately obtaining the radiation correction gain and bias coefficient corresponding to each detector array.

[0029] S4.1 For each specific detector column i, its extracted quantile feature sequence The global reference quantile distribution sequence constructed in step S3 is used as the independent variable. A univariate linear regression model is established using the variable as the dependent variable.

[0030] S4.2. Calculate the covariance between the independent and dependent variables according to the least squares method. and the variance of the independent variable. The ratio of covariance to variance is used as the radiation correction gain coefficient for the detector array. ,Right now .

[0031] S4.3 Calculate the mean of the dependent variable. with the mean of the independent variable The radiation correction bias coefficient of the detector array is obtained by subtracting the product of the radiation correction gain coefficient and the independent variable from the mean of the dependent variable. ,Right now This allows for the calculation and storage of correction coefficients for all detector arrays.

[0032] S5. Using the radiometric correction gain and bias coefficient corresponding to each detector column, the original image data is subjected to column-by-column relative radiometric correction processing, and finally a high-quality hyperspectral image with correction completed and detector vertical strip noise eliminated is output.

[0033] S5.1 Read the original hyperspectral remote sensing image matrix to be calibrated Iterate through and extract the original pixel gray values ​​of each column (i-th column) of the detector in the corresponding single-band image. .

[0034] S5.2. Compare the pixel grayscale value of each column with the radiometric correction gain coefficient corresponding to that column obtained in step S4. Perform multiplication operations to complete the gain calibration of the detector response differences.

[0035] S5.3. Based on the numerical results of the gain calibration, add the corresponding radiation correction bias coefficient for this column. That is, the corrected pixel gray value of each column (i-th column) of the detector. The calculation formula is: The final output is a relative radiometrically corrected image after eliminating vertical stripe noise. .

[0036] To overcome the shortcomings of existing histogram statistical calibration methods, which rely on long-term massive data, are susceptible to extreme value interference, and have low computational efficiency, this embodiment provides a hyperspectral relative radiometric calibration method based on quantile statistics. This method does not require assumptions about the data distribution. It constructs a linear mapping by extracting equidistant quantiles within the effective radiation interval, thereby achieving high-precision and fast-convergence characterization of hyperspectral image detector response characteristics and robustly eliminating strip noise.

[0037] The method described in this embodiment has significant advantages in terms of noise robustness and image quality improvement. From a technical perspective, traditional histogram statistical calibration methods rely heavily on the absolute frequency distribution of pixel gray values. When extreme outliers such as clouds (extremely bright), shadows, or bad pixels (extremely dark) exist in the image, these extreme values ​​severely distort the cumulative probability density function. During histogram matching, this nonlinear distribution distortion is forcibly transmitted to the correction coefficients, thereby introducing nonlinear mapping artifacts such as "dark stripes" in normal terrain areas.

[0038] This embodiment innovatively introduces the principle of quantile statistics. The essence of quantiles is based on the spatial ordering position of data rather than absolute numerical frequency. This means that extreme brightness values ​​will only be distributed at the two ends of the sequence, without interfering with the statistical characteristics of the core radiation region. The method in this embodiment mathematically shields against interference from clouds and shadows, ensuring that all extracted feature points come from valid observation targets. Therefore, the linear mapping relationship fitted based on effective quantiles is extremely robust, not only completely removing dense vertical stripes from the original image but also avoiding nonlinear artifacts caused by traditional methods. Figure 2 and Figure 3 As shown, based on the quantitative test results of real images, the method in this embodiment significantly reduces the image fringe coefficient from 0.2339% to 0.0175% and the light response non-uniformity from 1.8964% to 0.7556%, thus achieving a significant improvement in radiometric correction accuracy.

[0039] In addition to the embodiments described above, the present invention may have other implementations. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by the present invention.

Claims

1. A method for relative radiometric calibration of remote sensing images based on quantiles, characterized in that: Includes the following steps: S1. Perform initial preprocessing on hyperspectral remote sensing images, collect historical image data acquired by the satellite in orbit, and separate and store these raw images according to different spectral bands. S2. Acquire the preprocessed single-band image. Within the limited effective radiation range, sort the pixel values ​​of each column of the detector in a monotonically increasing manner. Then, uniformly extract a specified number of quantile values ​​from the sorted sequence to generate the quantile feature sequence of each detector column. S3. For the quantile feature sequence, calculate the arithmetic mean of all detector columns at each same quantile level, and use the arithmetic mean as the target reference quantile at that quantile level, thereby constructing a global reference quantile distribution. S4. Based on the principle of least squares, calculate the covariance and variance between the quantiles extracted by each detector column and the reference quantiles, thereby constructing a linear mapping relationship, and then obtaining the radiation correction gain and bias coefficient corresponding to each detector column. S5. Using the radiometric correction gain and bias coefficient corresponding to each detector column, the original image data is subjected to column-by-column relative radiometric correction processing, and finally outputs a hyperspectral image that has been corrected and whose vertical strip noise of the detector has been eliminated.

2. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 1, characterized in that: Step S1 specifically includes the following sub-steps: S1.1 Collect historical image data acquired by the satellite in orbit, screen and remove various types of abnormal images. Abnormal images include those that are completely invalid or have severely damaged data packets. Abnormal images with severely damaged data packets include those with a mean exceeding the limit, an extreme value ratio exceeding the standard, or a standard deviation below the threshold. S1.2 Separate the selected original images according to different spectral bands, and construct a two-dimensional image matrix for each spectral band that is stored independently.

3. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 1, characterized in that: Step S2 specifically includes the following sub-steps: S2.1 Calculate the total number of quantiles to be extracted based on the product of the number of images and the height dimension of the image, according to the preset quantile ratio coefficient and the height dimension of the image. S2.

2. Generate a uniformly distributed quantile position sequence within the effective radiation range of 1-16384 of the 14-bit digital quantization value. S2.

3. The image data of each column of the detector are sorted monotonically in ascending order, and the corresponding quantile values ​​are extracted by direct sampling according to the quantile position sequence to generate the quantile feature sequence of each detector column.

4. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 3, characterized in that: In step S2, the total number of quantiles is denoted as n; for the k-th quantile level, k=0,1,...,n-1, its corresponding cumulative probability position The calculation formula is: ; The image data of each column of the detector is monotonically sorted in ascending order to obtain a one-dimensional sequence of the i-th column. Let i = 0, 1, ..., W-1, where W represents the number of columns in the image, i.e., the number of detector elements. Then, based on the quantile position sequence, the corresponding quantile values ​​are extracted through direct sampling to generate the quantile feature sequence for each detector column, i.e., the k-th quantile value of the i-th column is... H represents the number of rows in the image, i.e., the number of valid pixels. This indicates a round-down operation.

5. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 1, characterized in that: The specific steps S3 are as follows It includes the following steps: S3.1 Obtain the quantile feature sequences of all extracted detector columns, and align them horizontally according to the same quantile level to form a quantile feature matrix; S3.2 For each given quantile level k, extract the specific quantile values ​​of all detector columns at that level. This forms an independent one-dimensional array; S3.3 Calculate the arithmetic mean of all quantile values ​​in the one-dimensional array, that is, perform the mean operation in the row direction of the quantile matrix; S3.

4. Use the calculated arithmetic mean as the target reference quantile for the corresponding quantile level, aggregate the target reference quantiles for all levels, and construct a global reference quantile distribution.

6. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 5, characterized in that: In step S3, the size of the quantile feature matrix is ​​n×W, where n represents the total number of quantiles and W represents the number of columns in the image, i.e., the number of detector elements; the target reference quantile is denoted as... The calculation formula is: , This represents the kth quantile value in the i-th column.

7. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 1, characterized in that: Step S4 specifically includes the following sub-steps: S4.1 For each specific detector column i, its extracted quantile feature sequence As the independent variable, This represents the k-th quantile value in the i-th column, where k represents the quantile position level and n represents the total number of quantiles; using the global reference quantile distribution sequence constructed in step S3. Using this as the dependent variable, a univariate linear regression model is established. Indicates the target reference quantile; S4.

2. According to the principle of least squares, calculate the covariance between the independent variable and the dependent variable, as well as the variance of the independent variable, and use the ratio of the covariance to the variance as the radiation correction gain coefficient of the detector array. S4.3 Calculate the mean of the dependent variable and the mean of the independent variable. Subtract the product of the radiation correction gain coefficient and the mean of the independent variable from the mean of the dependent variable to obtain the radiation correction bias coefficient of the detector array, thereby completing the calculation and storage of the correction coefficients of all detector arrays.

8. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 1, characterized in that: Step S5 specifically includes the following sub-steps: S5.1 Read the original hyperspectral remote sensing image data to be corrected, and extract the original pixel gray values ​​of each column of the detector in the corresponding single-band image. S5.2 Multiply the pixel gray value of each column with the radiation correction gain coefficient corresponding to that column obtained in step S4 to complete the gain calibration of detector response difference. S5.

3. Based on the numerical results of the gain calibration, add the corresponding radiometric correction bias coefficient for that column to output the relative radiometrically corrected image after eliminating vertical strip noise.

9. The method for relative radiometric calibration of remote sensing images based on quantiles according to claim 8, characterized in that: In step S5, the corrected pixel grayscale values ​​of each column of the detector are... The calculation formula is: , This represents the radiation correction gain coefficient. Represents the original pixel grayscale value. This represents the radiation correction bias coefficient.

10. A quantile-based remote sensing image relative radiometric calibration system, used to perform the method according to any one of claims 1-9, characterized in that: include: The preprocessing module is used to perform initial preprocessing on hyperspectral remote sensing images, collect historical image data acquired by the satellite in orbit, and separate and store these raw images according to different spectral bands. The quantile feature sequence generation module is used to acquire preprocessed single-band images. Within the limited effective radiation range, the pixel values ​​of each column of the detector are monotonically sorted in ascending order. Then, a specified number of quantile values ​​are uniformly extracted from the sorted sequence to generate the quantile feature sequence of each detector column. The reference quantile distribution construction module is used to calculate the arithmetic mean of all detector columns at each same quantile level for the quantile feature sequence, and use the arithmetic mean as the target reference quantile at that quantile level, thereby constructing a global reference quantile distribution. The correction coefficient calculation module is used to calculate the covariance and variance between the quantiles extracted by each detector column and the reference quantiles based on the least squares method, thereby constructing a linear mapping relationship and obtaining the radiation correction gain and bias coefficient corresponding to each detector column. The relative radiometric correction module is used to perform column-by-column relative radiometric correction on the original image data using the radiometric correction gain and bias coefficient corresponding to each detector column, and finally outputs a corrected hyperspectral image with the vertical strip noise of the detector eliminated.

Images