Hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs

Abstract

The present application relates to a hyperspectral remote sensing image radiation normalization method, in particular to a hyperspectral remote sensing image relative radiation normalization method based on pseudo-invariant region pairs, which solves the technical problems of low spatial matching accuracy and spectral feature deviation of the existing relative radiation normalization method based on pixels. The present application adopts superpixel segmentation, morphological processing and superpixel guided method to screen pseudo-invariant regions, then adopts a self-correction model based on spatial-spectral constraints to adaptively learn the relative radiation normalization of the high light remote sensing spectral image to be corrected, adopts a quantitative method, which can not only fully utilize the spatial-spectral features of the hyperspectral remote sensing image, but also adaptively correct the radiation intensity of the hyperspectral remote sensing image in different time phases, improve the consistency and quantitative inversion accuracy between time sequences, thereby improve the accuracy and consistency of the relative radiation normalization, and is suitable for the radiation normalization of high spatial resolution, complex ground object scene hyperspectral remote sensing images.

Description

Technical Field

[0001] This invention relates to a method for radiometric normalization of hyperspectral remote sensing images, specifically a method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs. Background Technology

[0002] Airborne hyperspectral remote sensing images possess rich spectral information and high-resolution spatial information, making them highly promising for applications such as dynamic monitoring of land cover attributes, wastewater detection, and quantitative inversion of land cover elements. Relative radiometric normalization is an important prerequisite for their quantitative application.

[0003] Traditional relative radiometric normalization methods can be divided into two categories: pixel-based relative radiometric normalization and distribution-based relative radiometric normalization. Currently, for the same hyperspectral payload, researchers typically choose pixel-based relative radiometric normalization methods to obtain better results. However, with the continuous improvement of hyperspectral detector performance, the relative radiometric normalization of high spatial resolution hyperspectral remote sensing images faces the following shortcomings:

[0004] (1) It is highly subjective. The pixel-based relative radiometric normalization method often relies on manually set parameters or thresholds during the processing, which leads to a highly subjective result of normalization, affecting its accuracy and consistency. This subjectivity is even more pronounced when processing high spatial resolution images due to the increase in image details.

[0005] (2) After extracting pseudo-invariant feature points, the least squares method or orthogonal regression method is usually chosen to calculate the mapping relationship between each band of the hyperspectral remote sensing image to be corrected and the reference hyperspectral remote sensing image. However, for multi-temporal hyperspectral remote sensing images after atmospheric correction, the main reasons for the spectral inconsistency between time phases include the incomplete removal of the atmosphere and the influence of noise from instruments and observations. Therefore, the relationship between the two is nonlinear. When processing nonlinear data, if the least squares method or orthogonal regression method that assumes a linear relationship between the data is used, the fitting results will be inaccurate, thus affecting the accuracy of the subsequent normalization results.

[0006] (3) The band-by-band normalization method is usually adopted, which ignores the consistency of spectral shape between bands, resulting in a shift in the spectral characteristics after normalization. Summary of the Invention

[0007] The purpose of this invention is to solve the technical problems of low spatial matching accuracy and spectral feature shift in existing pixel-based relative radiometric normalization methods, and to provide a relative radiometric normalization method for hyperspectral remote sensing images based on pseudo-invariant region pairs.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs, characterized by the following steps:

[0010] Step 1: Use a spectrometer to acquire the reference hyperspectral remote sensing image DN_T1 and the hyperspectral remote sensing image to be calibrated DN_T2 at different times, and then calculate the entrance pupil radiance of DN_T1 and DN_T2 respectively according to the calibration coefficient of the spectrometer.

[0011] Step 2: Perform atmospheric correction on DN_T1 using the entrance pupil radiance of DN_T1, and perform atmospheric correction on DN_T2 using the entrance pupil radiance of DN_T2. Then, perform image registration on the two to achieve feature space alignment, resulting in feature space aligned DN_T1 and DN_T2.

[0012] Step 3: Preprocess the feature space aligned DN_T1 and DN_T2 respectively, and remove the regions that change significantly over time to obtain the enhanced DN_T1 and DN_T2;

[0013] Step 4: Use a feature band selection network based on spectral response function to perform dimensionality reduction on the enhanced DN_T1 and DN_T2 respectively, extract their spectral features, and obtain the dimensionality-reduced DN_T1 and DN_T2.

[0014] Step 5: Using an iterative weighted multivariate change detection transformation algorithm, obtain the pseudo-invariant feature point groups of DN_T1 and DN_T2 in the enhanced DN_T1 and DN_T2 respectively;

[0015] Step 6: Perform ground feature clustering on the dimensionality-reduced DN_T1 and DN_T2 respectively using superpixel segmentation. At the same time, perform morphological processing on the pseudo-invariant feature point groups of DN_T1 and DN_T2 respectively. Then, based on the ground feature clustering results and morphological processing results, filter the pseudo-invariant regions of DN_T1 and DN_T2 respectively.

[0016] Step 7: Pair the pseudo-invariant regions of DN_T1 and DN_T2, and then input the paired pseudo-invariant regions into the self-calibrating initial model based on spatial spectrum constraints. The self-calibrating initial model normalizes the pseudo-invariant region of DN_T2, and then adjusts the parameters of the self-calibrating initial model according to the loss function value between the normalization result and the pseudo-invariant region of DN_T1 to obtain the self-calibrating model based on spatial spectrum constraints.

[0017] Step 8: Input the DN_T2 aligned with the feature space into the self-calibration model based on spatial spectral constraints to obtain the normalized hyperspectral remote sensing image, thus completing the relative radiometric normalization of the hyperspectral remote sensing image.

[0018] Furthermore, step 6 specifically includes:

[0019] 6.1. The superpixel segmentation algorithm is used to divide the ground feature information in the dimension-reduced DN_T1 and DN_T2 into superpixel regions to obtain the superpixel region sets of DN_T1 and DN_T2 respectively;

[0020] 6.2. Set the spectral angular distance threshold and the root mean square error threshold, and calculate the spectral angular distance and root mean square error of two adjacent superpixel regions in the DN_T1 and DN_T2 superpixel region sets respectively. If the spectral angular distance of two adjacent superpixel regions is less than the spectral angular distance threshold and the root mean square error is less than the root mean square error threshold, then merge the two adjacent superpixel regions to obtain the optimized region sets of DN_T1 and DN_T2 respectively.

[0021] 6.3. Use morphological methods to process the pseudo-invariant feature point groups of DN_T1 and DN_T2 to obtain the initial set of pseudo-invariant regions of DN_T1 and DN_T2;

[0022] 6.4. Take the intersection of the optimized region set and the initial pseudo-invariant region set of DN_T1, and the optimized region set and the initial pseudo-invariant region set of DN_T2, respectively. Then calculate the maximum inscribed rectangle of the intersection to obtain the pseudo-invariant regions of DN_T1 and DN_T2.

[0023] Furthermore, step 4 specifically involves:

[0024] 4.1 Construct a feature band selection network based on the spectral response function, and set a loss function threshold for the feature band selection network; the feature band network includes an encoding network and a decoding network connected in sequence, and the decoding network is based on the spectral Transformer technique;

[0025] 4.2 Input the enhanced DN_T1 or DN_T2 into the encoding network. The encoding network simulates the spectral response curve and selects the spectral features that are sensitive to the spectral response of the enhanced DN_T1 or DN_T2 to reduce the dimensionality of the enhanced DN_T1 or DN_T2 and obtain the low-dimensional features of DN_T1 or DN_T2.

[0026] 4.3 The decoding network reconstructs the low-dimensional features of DN_T1 or DN_T2 to obtain the feature cube of DN_T1 or DN_T2;

[0027] 4.4 Calculate the loss function value of the feature band selection network based on the feature cube of DN_T1 or DN_T2, and adjust the parameters of the encoding and decoding networks according to the loss function value. Then return to step 4.2 for iteration until the loss function value is less than the loss function threshold to obtain the trained feature band selection network.

[0028] 4.5 Input the enhanced DN_T1 and DN_T2 into the trained feature band selection network to obtain the dimensionality-reduced DN_T1 and DN_T2.

[0029] Further, in step 4.1, the encoding network includes a 1×1 3D convolutional unit, a LeakyReLU activation unit, and a residual linking unit connected in sequence.

[0030] Furthermore, in step 4.1, the loss function of the feature band selection network is the root mean square error between the output of the decoding network and the input of the encoding network.

[0031] Furthermore, in step 7, the self-calibration model based on spatial spectral constraints includes an input layer, a feature extraction layer, a spectral constraint layer, a radiation difference learning layer, and a radiation correction output layer connected in sequence.

[0032] The feature extraction layer includes 3D convolutional remnants for extracting spatial and spectral features of pseudo-invariant regions, and the spectral constraint layer is based on a spectral attention mechanism.

[0033] Furthermore, in step 7, the loss function L of the self-calibrating model based on spatial spectrum constraints * for:

[0034] L * =α*D KL (T2',T1)+β*SAM(mean(T2'),mean(T1))+γ*SSIM(T1,T1')

[0035] Among them, D KL (T2′,T1) represents the KL divergence distribution between the pseudo-invariant regions of DN_T2 and DN_T1 after relative radiative normalization, where T2′ represents the pseudo-invariant region of DN_T2 after relative radiative normalization, and T1 represents the pseudo-invariant region of DN_T1.

[0036] SAM(mean(T2′),mean(T1)) represents the spectral angular distance between the spectral mean of the pseudo-invariant DN_T2 region after relative radiative normalization and the spectral mean of DN_T1, mean(T2′) represents the spectral mean of the pseudo-invariant DN_T2 region after relative radiative normalization and the spectral mean of DN_T1, and mean(T1) represents the spectral mean of DN_T1.

[0037] SSIM(T1′,T1) represents the structural similarity between DN_T1 and DN_T1 after relative radiometric normalization, and T1′ represents the pseudo-invariant region of DN_T1 after relative radiometric normalization.

[0038] α, β, and γ are D KL The weight parameters of (T2′,T1), SAM(mean(T2′),mean(T1)), and SSIM(T1′,T1).

[0039] Furthermore, in step 3, the areas that change significantly over time include vegetation-covered areas, water bodies, and shaded areas. Therefore, step 3 specifically involves:

[0040] 3.1 Extract the characteristic spectral segments of vegetation cover areas and water body areas from the database respectively;

[0041] 3.2 Calculate the normalized vegetation index and the normalized differential water index based on the characteristic spectral segments of the vegetation cover area and the water body area, respectively;

[0042] 3.3 Divide the DN_T1 and DN_T2 regions aligned in the feature space into regions, and set the vegetation cover threshold, water body threshold, and shadow threshold.

[0043] 3.4. Calculate the ratio of the difference between the reflectance of the near-infrared band and the red band to the sum of the two based on the normalized vegetation index. If the ratio is greater than the vegetation coverage area threshold, the corresponding area will be removed.

[0044] The reflectance difference in the shortwave infrared band and green light band of each region is calculated based on the normalized difference water index. If the reflectance difference is greater than the water threshold, the corresponding region is removed.

[0045] Calculate the shadow index for each region. If the shadow index is greater than the shadow threshold, the corresponding region is removed. Further, in step 1, the entrance pupil radiance is calculated using the following formula:

[0046] L = a * DN + b

[0047] Where L is the entrance pupil radiance, DN is the energy intensity of the hyperspectral remote sensing image, and a and b are the calibration coefficients of the spectrometer.

[0048] Furthermore, in step 2, the image registration method is as follows: image registration is performed by combining POS information and image SURF feature points.

[0049] Compared with the prior art, the present invention has the following beneficial technical effects:

[0050] 1. The relative radiometric normalization method for hyperspectral remote sensing images based on pseudo-invariant region pairs provided by this invention is applicable to hyperspectral remote sensing images with high spatial resolution and complex ground features. It uses superpixel segmentation and morphological processing to screen pseudo-invariant regions, and employs a self-calibration model based on spatial-spectral constraints to adaptively learn and normalize the relative radiometric normalization of the hyperspectral remote sensing image to be calibrated. It adopts a quantitative method, which can not only make full use of the spatial-spectral features of hyperspectral remote sensing images, but also adaptively correct the radiometric intensity of hyperspectral remote sensing images at different time phases. It can improve the consistency between time phases and the accuracy of quantitative inversion, solve the problems of low spatial matching accuracy and spectral feature shift, and improve the accuracy and consistency of relative radiometric normalization.

[0051] 2. In the hyperspectral remote sensing image relative radiation normalization method based on pseudo-invariant region pairs provided by the present invention, morphological processing and superpixel segmentation methods are used to guide the screening of pseudo-invariant regions through superpixels, which can preserve rich spatial spectral features and extract the feature relationships of spatial neighborhoods, thereby reducing the interference of the proximity effect on the screening of pseudo-invariant regions.

[0052] 3. In the hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs provided by this invention, the encoding network simulates the spectral response curve and selects spectral features that are sensitive to the spectral response, which can ensure the maximization of information after dimensionality reduction; the decoding network is based on spectral Transformer technology, which can better understand the complex relationship between spectral features, thereby generating more accurate reconstructed data.

[0053] 4. In the hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs provided by the present invention, the self-calibration model based on spatial spectral constraints includes a spectral constraint layer and introduces a spectral attention mechanism to focus on regions or bands that have a greater impact on radiometric normalization, thereby improving the accuracy of relative radiometric normalization.

[0054] 5. In the hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs provided by the present invention, the loss function of the self-calibration model based on spatial-spectral constraints includes KL divergence and spectral angular distance, which can not only maintain spatial and spectral consistency, but also avoid spatial misalignment that causes the hyperspectral remote sensing image to be calibrated to undergo spatial transformation guided by the reference hyperspectral remote sensing image.

[0055] 6. In the hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs provided by the present invention, the reference hyperspectral remote sensing image and the hyperspectral remote sensing image to be corrected are screened by normalized vegetation index, normalized differential water index and shadow index. This can eliminate interference information in the process of extracting pseudo-invariant feature point groups and ensure the accuracy of subsequent relative radiometric normalization. Attached Figure Description

[0056] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0057] Figure 2 The reference hyperspectral remote sensing image obtained in step 1 of this embodiment of the invention;

[0058] Figure 3 The hyperspectral remote sensing image to be corrected is obtained in step 1 of this embodiment of the invention;

[0059] Figure 4 This refers to the normalized hyperspectral remote sensing image obtained in step 8 of this embodiment of the invention.

[0060] Figure 5 This is a spectral curve diagram of a reference hyperspectral remote sensing image, a hyperspectral remote sensing image to be corrected, and a normalized hyperspectral remote sensing image in an embodiment of the present invention. Detailed Implementation

[0061] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a more comprehensive explanation of the relative radiometric normalization method for hyperspectral remote sensing images based on pseudo-invariant region pairs proposed in this invention. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this invention and are not intended to limit the scope of protection of this invention.

[0062] A method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs, such as Figure 1 As shown, it includes the following steps:

[0063] Step 1: Use a spectrometer to acquire data at different times, such as... Figure 2 The reference hyperspectral remote sensing image DN_T1 shown and as follows Figure 3 The hyperspectral remote sensing image DN_T2 to be calibrated is shown below. Then, the entrance pupil radiance of DN_T1 and DN_T2 is calculated respectively based on the spectrometer's calibration coefficients. The entrance pupil radiance is calculated using the following formula:

[0064] L = a * DN + b

[0065] Where L is the entrance pupil radiance, DN is the energy intensity of the hyperspectral remote sensing image, and a and b are the calibration coefficients of the spectrometer.

[0066] Step 2: Perform atmospheric correction on DN_T1 using the entrance pupil radiance of DN_T1, and perform atmospheric correction on DN_T2 using the entrance pupil radiance of DN_T2. Then, combine the POS information and the image SURF feature points to perform image registration on the two to achieve feature space alignment, resulting in feature space aligned DN_T1 and DN_T2.

[0067] Step 3: Preprocess the feature space aligned DN_T1 and DN_T2 respectively, and remove the regions that change significantly over time to obtain the enhanced DN_T1 and DN_T2.

[0068] Among them, the areas that change significantly over time include vegetation cover areas, water bodies, and shaded areas. Therefore, step 3 specifically involves:

[0069] 3.1 Extract the characteristic spectral segments of vegetation cover areas and water body areas from the database respectively;

[0070] 3.2 Calculate the normalized vegetation index and the normalized differential water index based on the characteristic spectral segments of the vegetation cover area and the water body area, respectively;

[0071] 3.3 Divide the DN_T1 and DN_T2 regions aligned in the feature space into regions, and set the vegetation cover threshold, water body threshold, and shadow threshold.

[0072] 3.4. Calculate the ratio of the difference between the reflectance of the near-infrared band and the red band to the sum of the two based on the normalized vegetation index. If the ratio is greater than the vegetation coverage area threshold, the corresponding area will be removed.

[0073] The reflectance difference in the shortwave infrared band and green light band of each region is calculated based on the normalized difference water index. If the reflectance difference is greater than the water threshold, the corresponding region is removed.

[0074] Calculate the shadow index for each region. If the shadow index is greater than the shadow threshold, the corresponding region is removed.

[0075] Step 4: Use a feature selection band network based on spectral response function to reduce the dimensionality of the enhanced DN_T1 and DN_T2, extract their spectral features, and obtain the dimensionality-reduced DN_T1 and DN_T2. Specifically:

[0076] 4.1 Construct a feature band selection network based on the spectral response function, and set a loss function threshold for the feature band selection network; the feature band network includes an encoding network and a decoding network connected in sequence, and the decoding network is based on the spectral Transformer technique;

[0077] 4.2 Input the enhanced DN_T1 or DN_T2 into the encoding network. The encoding network simulates the spectral response curve and selects the spectral features that are sensitive to the spectral response of the enhanced DN_T1 or DN_T2 to reduce the dimensionality of the enhanced DN_T1 or DN_T2 and obtain the low-dimensional features of DN_T1 or DN_T2.

[0078] 4.3 The decoding network reconstructs the low-dimensional features of DN_T1 or DN_T2 to obtain the feature cube of DN_T1 or DN_T2;

[0079] 4.4 Calculate the loss function value of the feature band selection network based on the feature cube of DN_T1 or DN_T2, and adjust the parameters of the encoding and decoding networks according to the loss function value. Then return to step 4.2 for iteration until the loss function value is less than the loss function threshold to obtain the trained feature band selection network.

[0080] 4.5 Input the enhanced DN_T1 and DN_T2 into the trained feature band selection network to obtain the dimensionality-reduced DN_T1 and DN_T2.

[0081] In this step, the encoding network is composed of stacked residual modules, each consisting of a sequentially connected 1×1 3D convolutional unit, a LeakyReLU activation unit, and a residual linking unit. The encoding network compresses the input data into a low-dimensional feature representation. The decoding network, based on spectral Transformer technology, captures long-range dependencies in the data, helping it better understand the complex relationships between spectral features and thus generate more accurate reconstructed data. The loss function of the feature band selection network uses the root mean square error between the output of the decoding network and the input of the encoding network to guide the feature band selection network in extracting low-dimensional features.

[0082] Step 5: Using an iterative weighted multivariate change detection transformation algorithm, obtain the pseudo-invariant feature point groups of DN_T1 and DN_T2 in the enhanced DN_T1 and DN_T2 respectively.

[0083] Step 6: Perform land feature clustering on the dimensionality-reduced DN_T1 and DN_T2 using superpixel segmentation. Simultaneously, perform morphological processing on the pseudo-invariant feature point groups of DN_T1 and DN_T2 respectively. Then, based on the land feature clustering results and morphological processing results, filter the pseudo-invariant regions of DN_T1 and DN_T2 respectively. Specifically:

[0084] 6.1. The superpixel segmentation algorithm is used to divide the ground feature information in the dimension-reduced DN_T1 and DN_T2 into superpixel regions to obtain the superpixel region sets of DN_T1 and DN_T2 respectively;

[0085] 6.2. Set the spectral angular distance threshold and the root mean square error threshold, and calculate the spectral angular distance and root mean square error of two adjacent superpixel regions in the DN_T1 and DN_T2 superpixel region sets respectively. If the spectral angular distance of two adjacent superpixel regions is less than the spectral angular distance threshold and the root mean square error is less than the root mean square error threshold, then merge the two adjacent superpixel regions to obtain the optimized region sets of DN_T1 and DN_T2 respectively.

[0086] 6.3. Use morphological methods to process the pseudo-invariant feature point groups of DN_T1 and DN_T2 to obtain the initial set of pseudo-invariant regions of DN_T1 and DN_T2;

[0087] 6.4. Take the intersection of the optimized region set and the initial pseudo-invariant region set of DN_T1, and the optimized region set and the initial pseudo-invariant region set of DN_T2, respectively. Then calculate the maximum inscribed rectangle of the intersection to obtain the pseudo-invariant regions of DN_T1 and DN_T2.

[0088] Step 7: Pair the pseudo-invariant regions of DN_T1 and DN_T2, and then input the paired pseudo-invariant regions into the self-calibrating initial model based on spatial spectrum constraints. The self-calibrating initial model normalizes the pseudo-invariant region of DN_T2, and then adjusts the parameters of the self-calibrating initial model according to the loss function value between the normalization result and the pseudo-invariant region of DN_T1 to obtain the self-calibrating model based on spatial spectrum constraints.

[0089] The self-calibration model based on spatial-spectral constraints comprises an input layer, a feature extraction layer, a spectral constraint layer, a radiometric difference learning layer, and a radiometric correction output layer, connected sequentially. The feature extraction layer includes 3D convolutional remnants for extracting spatial and spectral features of pseudo-invariant regions. The input layer takes pseudo-invariant regions DN_T1 and DN_T2 as input; the feature extraction layer extracts spatial and spectral features of these pseudo-invariant regions; the spectral constraint layer introduces a spectral attention mechanism to help the autoencoder model focus on regions or bands that significantly affect radiometric normalization; the radiometric difference learning layer captures global radiometric variations; and the radiometric correction output layer outputs the radiometrically normalized image, ensuring that the normalized hyperspectral remote sensing image to be corrected maintains the same intensity as the reference hyperspectral remote sensing image while preserving the spatial and spectral dimensions of the original image.

[0090] The loss function L of the self-calibrating model * for:

[0091] L * =α*D KL (T2',T1)+β*SAM(mean(T2'),mean(T1))+γ*SSIM(T1,T1')

[0092] Among them, D KL (T2′,T1) represents the KL divergence distribution between the pseudo-invariant regions of DN_T2 and DN_T1 after relative radiative normalization, where T2′ represents the pseudo-invariant region of DN_T2 after relative radiative normalization, and T1 represents the pseudo-invariant region of DN_T1.

[0093] SAM(mean(T2′),mean(T1)) represents the spectral angular distance between the spectral mean of the pseudo-invariant DN_T2 region after relative radiative normalization and the spectral mean of DN_T1, mean(T2′) represents the spectral mean of the pseudo-invariant DN_T2 region after relative radiative normalization and the spectral mean of DN_T1, and mean(T1) represents the spectral mean of DN_T1.

[0094] SSIM(T1′,T1) represents the structural similarity between DN_T1 and DN_T1 after relative radiometric normalization, and T1′ represents the pseudo-invariant region of DN_T1 after relative radiometric normalization.

[0095] α, β, and γ are D KL The weight parameters of (T2′,T1), SAM(mean(T2′),mean(T1)), and SSIM(T1′,T1) are initially set to 0.6, 0.2, and 0.2 respectively in this embodiment.

[0096] In the loss function of the self-calibration model, KL divergence distribution is introduced to ensure that the intensity distribution of the normalized DN_T2 is consistent with that of DN_T1; spectral angular distance is introduced to avoid changes in the spectral characteristics of DN_T2 during the calibration process; and structural similarity is introduced to ensure that the spatial texture of the normalized DN_T2 and the original image remains unchanged.

[0097] Step 8: Input the feature space-aligned DN_T2 into the self-calibrating model based on spatial spectrum constraints to obtain the following... Figure 4 The normalized hyperspectral remote sensing image shown is a result of the relative radiometric normalization of the hyperspectral remote sensing image.

[0098] like Figures 2-4 As shown, compared to the hyperspectral remote sensing image DN_T2 to be corrected, the radiance of the corrected image is closer to that of the reference hyperspectral remote sensing image DN_T1. Figure 5 As shown, the spectrum of the selected invariant region is compared. It can be seen that, compared with the hyperspectral remote sensing image DN_T2 to be corrected, the spectrum of the normalized hyperspectral remote sensing image is closer to the spectrum of the reference hyperspectral remote sensing image DN_T1. It can be seen that the relative radiation normalization method provided in this embodiment has high accuracy and consistency.

[0099] The hyperspectral remote sensing image relative radiometric normalization method based on pseudo-invariant region pairs provided in this embodiment makes full use of the spatial-spectral features of high spatial resolution hyperspectral remote sensing images. At the same time, it constructs a spatial-spectral constraint loss function for relative radiometric normalization under spatially incomplete alignment. It adopts a quantitative method, which does not require complicated operations and can adaptively correct the radiometric intensity of hyperspectral remote sensing images at different time phases, thereby improving the consistency between time series and the accuracy of relative radiometric normalization.

Claims

1. A method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs, characterized in that, Includes the following steps: Step 1: Use a spectrometer to acquire the reference hyperspectral remote sensing image DN_T1 and the hyperspectral remote sensing image to be calibrated DN_T2 at different times, and then calculate the entrance pupil radiance of DN_T1 and DN_T2 respectively according to the calibration coefficient of the spectrometer. Step 2: Perform atmospheric correction on DN_T1 using the entrance pupil radiance of DN_T1, and perform atmospheric correction on DN_T2 using the entrance pupil radiance of DN_T2. Then, perform image registration on the two to achieve feature space alignment, resulting in feature space aligned DN_T1 and DN_T2. Step 3: Preprocess the feature space aligned DN_T1 and DN_T2 respectively, and remove the regions that change significantly over time to obtain the enhanced DN_T1 and DN_T2; Step 4: Use a feature band selection network based on spectral response function to perform dimensionality reduction on the enhanced DN_T1 and DN_T2 respectively, extract their spectral features, and obtain the dimensionality-reduced DN_T1 and DN_T2. Step 5: Using an iterative weighted multivariate change detection transformation algorithm, obtain the pseudo-invariant feature point groups of DN_T1 and DN_T2 in the enhanced DN_T1 and DN_T2 respectively; Step 6: Perform ground feature clustering on the dimensionality-reduced DN_T1 and DN_T2 respectively using superpixel segmentation. At the same time, perform morphological processing on the pseudo-invariant feature point groups of DN_T1 and DN_T2 respectively. Then, based on the ground feature clustering results and morphological processing results, filter the pseudo-invariant regions of DN_T1 and DN_T2 respectively. Step 7: Pair the pseudo-invariant regions of DN_T1 and DN_T2, and then input the paired pseudo-invariant regions into the self-calibrating initial model based on spatial-spectral constraints. The self-calibrating initial model normalizes the pseudo-invariant region of DN_T2, and then adjusts the parameters of the self-calibrating initial model according to the loss function value between the normalization result and the pseudo-invariant region of DN_T1, to obtain the self-calibrating model based on spatial-spectral constraints. The self-calibrating model based on spatial-spectral constraints includes an input layer, a feature extraction layer, a spectral constraint layer, a radiometric difference learning layer, and a radiometric correction output layer connected in sequence. The feature extraction layer includes 3D convolutional remnants for extracting spatial and spectral features of the pseudo-invariant regions, and the spectral constraint layer is based on a spectral attention mechanism. Step 8: Input the DN_T2 aligned with the feature space into the self-calibration model based on spatial spectral constraints to obtain the normalized hyperspectral remote sensing image, thus completing the relative radiometric normalization of the hyperspectral remote sensing image.

2. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 1, characterized in that, Step 6 specifically involves: 6.

1. The superpixel segmentation algorithm is used to divide the ground feature information in the dimension-reduced DN_T1 and DN_T2 into superpixel regions to obtain the superpixel region sets of DN_T1 and DN_T2 respectively; 6.

2. Set the spectral angular distance threshold and the root mean square error threshold, and calculate the spectral angular distance and root mean square error of two adjacent superpixel regions in the DN_T1 and DN_T2 superpixel region sets respectively. If the spectral angular distance of two adjacent superpixel regions is less than the spectral angular distance threshold and the root mean square error is less than the root mean square error threshold, then merge the two adjacent superpixel regions to obtain the optimized region sets of DN_T1 and DN_T2 respectively. 6.

3. Use morphological methods to process the pseudo-invariant feature point groups of DN_T1 and DN_T2 to obtain the initial set of pseudo-invariant regions of DN_T1 and DN_T2; 6.

4. Take the intersection of the optimized region set and the initial pseudo-invariant region set of DN_T1, and the optimized region set and the initial pseudo-invariant region set of DN_T2, respectively. Then calculate the maximum inscribed rectangle of the intersection to obtain the pseudo-invariant regions of DN_T1 and DN_T2.

3. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 2, characterized in that, Step 4 is as follows: 4.1 Construct a feature band selection network based on the spectral response function, and set a loss function threshold for the feature band selection network; the feature band selection network includes an encoding network and a decoding network connected in sequence, and the decoding network is based on the spectral Transformer technology; 4.2 Input the enhanced DN_T1 or DN_T2 into the encoding network. The encoding network simulates the spectral response curve and selects the spectral features that are sensitive to the spectral response of the enhanced DN_T1 or DN_T2 to reduce the dimensionality of the enhanced DN_T1 or DN_T2 and obtain the low-dimensional features of DN_T1 or DN_T2. 4.3 The decoding network reconstructs the low-dimensional features of DN_T1 or DN_T2 to obtain the feature cube of DN_T1 or DN_T2; 4.4 Calculate the loss function value of the feature band selection network based on the feature cube of DN_T1 or DN_T2, and adjust the parameters of the encoding and decoding networks according to the loss function value. Then return to step 4.2 for iteration until the loss function value is less than the loss function threshold to obtain the trained feature band selection network. 4.5 Input the enhanced DN_T1 and DN_T2 into the trained feature band selection network to obtain the dimensionality-reduced DN_T1 and DN_T2.

4. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 3, characterized in that: In step 4.1, the encoding network includes a 1×1 3D convolutional unit, a LeakyReLU activation unit, and a residual linking unit connected in sequence.

5. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 4, characterized in that: In step 4.1, the loss function of the feature band selection network is the root mean square error between the output of the decoding network and the input of the encoding network.

6. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to any one of claims 1-5, characterized in that: In step 7, the loss function of the self-calibrating model based on spatial spectrum constraints L * for: L *= α * D KL ( T 2′, T 1)+ β * SAM ( mean ( T 2′), mean ( T 1))+ γ * SSIM ( T 1, T 1′) in, D KL ( T 2′, T 1) represents the KL divergence distribution between the pseudo-invariant regions DN_T2 and DN_T1 after relative radiative normalization. T 2′ represents the pseudo-invariant region of DN_T2 after relative radiation normalization. T 1 represents the pseudo-invariant region of DN_T1; SAM ( mean ( T 2′), mean ( T 1)) represents the spectral angular distance between the spectral mean of the pseudo-invariant region DN_T2 after relative radiation normalization, calculated along the channel dimension, and the spectral mean of DN_T1. mean ( T 2′) represents the spectral mean obtained along the channel dimension of the pseudo-invariant region of DN_T2 after relative radiative normalization. mean ( T 1) Represents the spectral mean of DN_T1; SSIM ( T 1, T 1′) represents the structural similarity between DN_T1 and DN_T1 after relative radiometric normalization. T 1′ represents the pseudo-invariant region of DN_T1 after relative radiation normalization; α , β , γ They are respectively D KL ( T 2′, T 1) SAM ( mean ( T 2′), mean ( T 1)) SSIM ( T 1, T The weight parameters of 1′).

7. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 6, characterized in that: In step 3, the areas that change significantly over time include vegetation-covered areas, water bodies, and shaded areas. Therefore, step 3 specifically involves: 3.1 Extract the characteristic spectral segments of vegetation cover areas and water body areas from the database respectively; 3.2 Calculate the normalized vegetation index and the normalized differential water index based on the characteristic spectral segments of the vegetation cover area and the water body area, respectively; 3.3 Divide the DN_T1 and DN_T2 regions aligned in the feature space into regions, and set the vegetation cover threshold, water body threshold, and shadow threshold. 3.

4. Calculate the ratio of the difference between the reflectance of the near-infrared band and the red band to the sum of the two based on the normalized vegetation index. If the ratio is greater than the vegetation cover threshold, the corresponding area will be removed. The reflectance difference in the shortwave infrared band and green light band of each region is calculated based on the normalized difference water index. If the reflectance difference is greater than the water threshold, the corresponding region is removed. Calculate the shadow index for each region. If the shadow index is greater than the shadow threshold, the corresponding region is removed.

8. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 7, characterized in that: In step 1, the entrance pupil radiance is calculated using the following formula: L = a * DN + b in, L DN represents the entrance pupil radiance, and DN represents the energy intensity of the hyperspectral remote sensing image. a and b These are the calibration coefficients for the spectrometer.

9. The method for relative radiometric normalization of hyperspectral remote sensing images based on pseudo-invariant region pairs according to claim 8, characterized in that, In step 2, the image registration method is as follows: image registration is performed by combining POS information and image SURF feature points.

Images