An embedded hyperspectral atmospheric correction processing method

Abstract

This invention discloses an embedded hyperspectral atmospheric correction processing method, comprising: establishing a general lookup table for atmospheric correction using the MODTRAN radiative transfer model; constructing a neural network model based on the lookup table and training it to obtain optimal weight parameters; acquiring observation information from data images and calculating apparent reflectance based on radiometric calibration coefficients; setting initial values ​​for atmospheric parameters based on remote sensing image data, and using the atmospheric correction network model for collaborative inversion to obtain atmospheric water vapor content and visibility; using remote sensing image observation information and the calculated atmospheric parameters, calling the MODTRAN radiative transfer model to calculate atmospheric correction coefficients for all bands in parallel; since atmospheric parameters are similar at adjacent spatial locations, using the atmospheric parameters from the previous stage as the initial values ​​for atmospheric correction in the next stage; repeating the above process until correction is completed. This solves the problems of complex parameter lookup, real-time status influence, and low inversion accuracy in atmospheric correction techniques.

Description

Technical Field

[0001] This invention relates to the field of on-board data processing technology in quantitative optics, and in particular to an embedded hyperspectral atmospheric correction processing method. Background Technology

[0002] In the field of quantitative optics, hyperspectral remote sensing technology has become increasingly mature and boasts significant advantages in spectral resolution, leading to its widespread application in areas such as oceanography, vegetation ecology, agriculture, geology, and atmosphere. During imaging with spaceborne or airborne hyperspectral quantitative detection payloads, atmospheric scattering or absorption, as well as topographical factors, can alter the surface-reflected radiation energy received by the sensor, resulting in distortion of ground object radiation information and significantly impacting the accuracy of spectral information extraction. To conduct quantitative research on hyperspectral remote sensing images, atmospheric correction is essential to remove the influence of the atmosphere on the radiation signal and obtain accurate information about ground objects.

[0003] Currently, scholars both domestically and internationally have conducted extensive research on atmospheric correction of hyperspectral images. Existing atmospheric correction methods can be divided into two categories: empirical statistical methods based on image features and physics-based radiative transfer model methods. The former does not need to consider the influence of atmospheric parameters, is simple, and has high computational efficiency, but requires prior knowledge and a large amount of measured data, resulting in high algorithmic costs. The latter considers the influence of the atmosphere on radiative transfer, can calculate more realistic surface reflectance, and has higher correction accuracy, but requires prior inversion of atmospheric parameters such as water vapor content and visibility, resulting in higher computational costs and relatively lower efficiency. Currently, lookup tables built using atmospheric radiative transfer models are widely used in atmospheric correction of hyperspectral images. To ensure the accuracy of lookup table correction, multiple atmospheric influencing factors need to be considered when constructing the lookup table, and each factor needs to be considered in multiple scenarios. This can easily lead to problems such as a large lookup table data volume, poor portability, constraints of attitude and orbit parameters and imaging conditions, and irrelevance of long strip data inversion work. Therefore, research on atmospheric correction methods for hyperspectral remote sensing data has become of significant theoretical and practical value. It is necessary to improve the key technical methods of hyperspectral atmospheric correction, refine the identification process, and improve the efficiency and accuracy of atmospheric correction. Summary of the Invention

[0004] This invention provides an embedded hyperspectral atmospheric correction processing method. Addressing the issues of data dependence and low accuracy in traditional correction algorithms, this method focuses on atmospheric correction of multispectral and hyperspectral remote sensing images acquired by spaceborne or airborne systems. Based on the atmospheric radiative transfer mechanism and classical atmospheric radiative transfer model, it solves the problems of complex parameter lookup, real-time status influence, and low inversion accuracy in existing atmospheric correction techniques using radiative transfer models. This provides algorithmic support for improving real-time atmospheric correction of multispectral and hyperspectral remote sensing images on spaceborne or airborne systems.

[0005] In a first aspect, an embedded hyperspectral atmospheric correction processing method is provided, characterized by comprising:

[0006] Step 1: Using the MODTRAN radiative transfer model, establish a general lookup table for atmospheric correction for a specified band, which includes the aerosol inversion band, the water vapor absorption band, and the reference band.

[0007] Step 2: Based on the general lookup table for atmospheric correction, construct a neural network model and obtain the optimal weight parameters of the neural network model through training;

[0008] Step 3, set the atmospheric water vapor content (WVC) i-1 With visibility AOT i-1 The observation information from the remote sensing image data is input into the neural network model, and the WVC is determined based on the output of the neural network model. i-1 The corresponding atmospheric transmittance T' is used to adjust and invert the WVC by comparing T' with the atmospheric transmittance reference value T. i AOT is determined based on the output of the neural network model. i-1 The corresponding simulated ground reflectance ρ” is obtained by comparing ρ and ρ” and adjusting the inversion to obtain AOT. i Repeat the inversion process until the difference between T' and T is less than the preset error, and the difference between ρ and ρ” is less than the preset error, to determine the final inverted atmospheric water vapor content (WVC) and visibility (AOT); Step 4, based on the observation information in the remote sensing image data, and the inverted atmospheric water vapor content (WVC) and visibility (AOT), call MODTRAN to calculate the atmospheric correction coefficients x for all bands in parallel. a ,x b ,x c , respectively corresponding to L p G b +G t S, L p For atmospheric path radiation, G b G t The values ​​represent the effects of direct atmospheric radiation and scattering, respectively, with S being the atmospheric albedo.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, in step 1, the atmospheric radiative transfer process is simulated using the MODTRAN radiative transfer model to obtain the atmospheric transmittance, and then the atmospheric path radiation L is calculated. p Albedo S and radiation correction parameter G of the balloon surface b +G t The simulation only covered atmospheric transport processes in the aerosol inversion band, water vapor absorption band, and reference band.

[0010]

[0011] Aerosol inversion bands include blue light band, red light band, and near-infrared band;

[0012] The water vapor absorption bands include 1117–1143 nm, 935–955 nm, and 810–830 nm.

[0013] The upper reference band corresponding to the water vapor absorption band of 1117–1143 nm includes 1184–1210 nm, and the lower reference band corresponding to the water vapor absorption band of 1117–1143 nm includes 1050–1067 nm.

[0014] The upper reference band corresponding to the water vapor absorption band of 935–955 nm includes 870–890 nm, and the lower reference band corresponding to the water vapor absorption band of 935–955 nm includes 995–1020 nm.

[0015] The upper reference band corresponding to the water vapor absorption band of 810–830 nm includes 850–870 nm, and the lower reference band corresponding to the water vapor absorption band of 810–830 nm includes 770–790 nm.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, in step 1, simulations are performed for three cases where the surface reflectance ρ is 0, 0.25, and 0.5, respectively. The atmospheric correction coefficient is solved using the following set of equations:

[0017]

[0018] The atmospheric correction factor is:

[0019]

[0020] In the formula L' 0.25 =L 0.25 -L0, L' 0.5 =L 0.5 -L0.

[0021] In conjunction with the first aspect, in some implementations of the first aspect, step 1, the MODTRAN table creation process based on a multi-core processor, is as follows:

[0022] Different observation conditions and atmospheric states are enumerated, and each case is distributed as a separate case to multiple threads or processes through instruction distribution to generate a Tape5 file;

[0023] The lookup table is built by calling MODTRAN in parallel using a multi-core processor.

[0024] In the process of hyperspectral remote sensing data processing, the parallel and for commands in OpenMP are used to decompose the task and start thread groups to handle multiple situations at the same time.

[0025] After execution, the system reads the data from the result file Tape7, calculates the atmospheric correction factor, and packages the input conditions and calculation results into the database.

[0026] In conjunction with the first aspect, in some implementations of the first aspect, in step 2, a neural network model is constructed; the model inputs include visibility (AOT), atmospheric water vapor content (WVC), and observed zenith angle (θ). v Solar zenith angle θ s Relative azimuth The center wavelength of the red band is λ. red Blue band center wavelength λ blue Center wavelength λ of water vapor absorption band absorb and reference band; the output is the atmospheric correction factor x for the red and blue bands and the water vapor absorption band. a ,x b ,x c Atmospheric transmittance, x a ,x b ,x c L respectively p G b +G t The neural network model structure includes:

[0027] 1) Convolutional layer Conv1d 3×1×32

[0028] 2) Activation layer ReLU

[0029] 3) Convolutional layer Conv1d 3×32×32

[0030] 4) Activation layer ReLU

[0031] 5) Maxpooling layer Maxpool1d

[0032] 6) Convolutional layer Conv1d 3×32×32

[0033] 7) Activation layer ReLU

[0034] 8) Convolutional layer Conv1d 3×32×32

[0035] 9) Activation layer ReLU

[0036] 10) Maxpooling layer Maxpool1d

[0037] 11) Convolutional layer Conv1d 3×32×10

[0038] 12) Activation layer ReLU.

[0039] In conjunction with the first aspect, in some implementations of the first aspect, in step 3, the normalized difference vegetation index (NDVI) and the apparent reflectance at 2100 nm are used to determine the dark pixels of the hyperspectral image data, wherein the requirements for dark pixels are as follows:

[0040]

[0041] ρ' 860 ,ρ' 660 ,ρ' 2100 The apparent reflectance of the hyperspectral image at 860nm, 660nm, and 2100nm are respectively used; atmospheric water vapor content (WVC) and visibility (AOT) are inverted using the selected dark pixels.

[0042] In conjunction with the first aspect, in some implementations of the first aspect, in step 3, attitude and orbit parameters and imaging conditions are obtained using real-time monitoring, and the initial values ​​for visibility and water vapor content are set to AOT0 = AOT0. 660 =AOT0 470 WVC0; feeds the parameters back to the neural network model and outputs atmospheric correction coefficients for the red and blue bands and the water vapor absorption band. And atmospheric transmittance T';

[0043] The simulated ground reflectance ρ” is calculated based on the atmospheric correction coefficients and apparent radiance of the red and blue bands. 660 ,ρ” 470 ;

[0044] If ρ” 660 and ρ 660 If the difference is not within the allowable range, then adjust AOT. i 660 , ρ 660 ≈ρ' 2100 / 2; if ρ” 660 Less than ρ 660 Then AOT i 660 Increase σ1 660 If ρ” 660 Greater than ρ 660 Then AOT i 660 Decrease step size σ1 660 This continues until the difference between the two is within the allowable error range;

[0045] If ρ” 470 and ρ 470 If the difference is not within the allowable range, then adjust AOT. i 470 , ρ470 ≈ρ' 2100 / 4; if ρ” 470 Less than ρ 470 Then AOT i 470 Increase σ1 470 If ρ” 470 Greater than ρ 470 Then AOT i 470 Decrease step size σ1 470 This continues until the difference between the two is within the allowable error range;

[0046] Average visibility obtained from the red and blue bands As a result of this inversion, AOT i ;

[0047] For the reference band corresponding to the water vapor absorption band, the atmospheric transmittance reference value T satisfies:

[0048]

[0049] In the formula, ρ(λ), ρ(λ) up ),ρ(λ down ) represent the apparent reflectance of the water vapor absorption band, upper reference band, and lower reference band, respectively; c1 and c2 are the reflectance weights of the reference bands, respectively; if T is less than the transmittance T' obtained from the neural network model, then the atmospheric water vapor content is WVC. i Increase the step size σ2, and vice versa.

[0050] In conjunction with the first aspect, in some implementations of the first aspect, in step 4, based on the image observation information and the atmospheric parameters calculated in step 3, the hyperspectral image bands are grouped and MODTRAN is called to calculate the atmospheric correction coefficients L for all bands. p G b +G t The atmospheric radiative transfer equation is expressed as:

[0051]

[0052] In the formula, L represents the apparent radiance received by the sensor; L p Atmospheric path radiation; G b G t These represent the effects of direct atmospheric radiation and scattering, respectively; S is atmospheric albedo; ρ is surface reflectance or the true reflectance of ground features.

[0053] Combined with atmospheric correction factor L p G b +G tThen, substituting the apparent reflectance obtained from the sensor into the above formula, we can obtain the surface reflectance or the true reflectance of the ground object, and thus complete the atmospheric correction.

[0054] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes:

[0055] Step 5: Repeat steps 3-4 to perform atmospheric correction on the next remote sensing image, using the inversion result from step 4 as input for the next stage to speed up the inversion process.

[0056] In conjunction with the first aspect, in some implementations of the first aspect, in step 5, for adjacent pixels that have the same ground features or similar surface reflectance, the surface reflectance of the atmospheric-corrected adjacent pixels of the previous remote sensing image is subtracted from the atmospheric-corrected adjacent pixels of the current remote sensing image to obtain the dark pixel.

[0057] Compared with the prior art, the solution provided by the present invention has at least the following beneficial technical effects:

[0058] 1. This invention establishes a universal, lightweight lookup table based on the radiative transfer model, which can meet the atmospheric correction requirements under various imaging conditions and is no longer limited by the sensor.

[0059] 2. This invention utilizes the powerful nonlinear mapping and generalization capabilities of neural networks to design a lightweight atmospheric correction lookup table, construct an atmospheric correction network module, and enhance its portability.

[0060] 3. The present invention monitors attitude and orbit parameters and imaging conditions in real time. By synergistically inverting atmospheric water vapor content and visibility, it takes into account the combined influence of the two on radiative transfer, thereby improving the accuracy of atmospheric correction.

[0061] 4. This invention adopts the idea of ​​rolling correction, which utilizes the similarity of atmospheric parameters between adjacent pixels and uses the atmospheric parameters obtained from the previous stage inversion as the current inversion input value, reducing the number of iterations, accelerating result convergence, and eliminating the problem caused by the absence of dark pixels when inverting aerosols in this stage; and uses a multi-core processor to call MODTRAN in parallel to realize the rapid calculation of atmospheric correction coefficients for band-by-band and pixel-by-pixel hyperspectral data, effectively improving the accuracy and efficiency of real-time atmospheric parameter inversion of long strip data. Attached Figure Description

[0062] Figure 1 This is a flowchart of the method of the present invention.

[0063] Figure 2 Flowchart for constructing a lightweight lookup table for the entire lifecycle.

[0064] Figure 3 This is a flowchart of the combined inversion process of water vapor inversion and aerosol inversion. Detailed Implementation

[0065] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0066] like Figure 1 As shown, the present invention provides an embedded hyperspectral atmospheric correction processing method, which mainly includes the following steps.

[0067] Step S1: Using the MODTRAN radiative transfer model, establish a general lookup table for atmospheric correction for a specified band, which includes the aerosol inversion band, the water vapor absorption band, and the reference band.

[0068] Hyperspectral remote sensing images are characterized by diverse imaging conditions and atmospheric states. Furthermore, the data volume of hyperspectral remote sensing images is enormous, and processing them band by band is both voluminous and time-consuming. Among the gas molecules and aerosols that contribute to atmospheric effects through absorption and scattering, the main absorbers include water vapor, ozone, carbon dioxide, and oxygen. Atmospheric molecular scattering and absorption caused by gases such as ozone and oxygen are relatively easy to correct because the concentrations of these elements are relatively stable in time and space. The difficulty lies in estimating the parameters of aerosols and water vapor. Therefore, removing the influence of aerosols and water vapor becomes the main focus of atmospheric correction.

[0069] In step S1, to construct a lightweight and universal lookup table that can meet the requirements of the entire work cycle, the MODTRAN radiative transfer model is used to simulate the atmospheric radiative transfer process to obtain the atmospheric transmittance T, and then the atmospheric path radiation L is calculated. p Albedo S and radiation correction parameter G of the balloon surface b +G t .

[0070] To avoid unnecessary computational overhead and ensure the lightweight nature of the lookup table, only the atmospheric transport processes in the aerosol inversion band, water vapor absorption band, and reference band were simulated during the simulation. The specific parameter settings for the atmospheric correction coefficient lookup table are shown in Table 1.

[0071] Table 1

[0072]

[0073] Based on the above table, different atmospheric conditions and observation conditions were set up, and simulations were conducted for three cases where the surface reflectance ρ was 0, 0.25, and 0.5. The atmospheric correction coefficient can be solved using the following set of equations:

[0074]

[0075] The atmospheric correction factor is:

[0076]

[0077] In the formula L' 0.25 =L 0.25 -L0, L' 0.5 =L 0.5 -L0.

[0078] Reference Figure 2 The specific process of building MODTRAN tables based on a multi-core processor is as follows: All possibilities for different observation conditions and atmospheric states are enumerated. Each case is distributed as a separate case to multiple threads or processes through instruction dispatch, generating a Tape5 file. The multi-core processor uses parallel calls to MODTRAN to build the lookup table, accelerating execution and shortening the table building time. Combining OpenMP (Open Multi-Processing) technology for parallel programming on a shared-memory computer, the parallel and for instructions in OpenMP are used to decompose the task during hyperspectral remote sensing data processing, enabling thread groups to handle multiple cases simultaneously, further accelerating execution. After execution, the data in the result file Tape7 is read, atmospheric correction coefficients are calculated, and the input conditions and calculation results are packaged and stored in the database.

[0079] Step S2: Based on the atmospheric correction general lookup table obtained in step S1, construct a neural network model and train it to obtain the optimal weight parameters.

[0080] Once the lookup table is established, construct the atmospheric correction neural network model. Model inputs include visibility (AOT), atmospheric water vapor content (WVC), and observed zenith angle (θ). v Solar zenith angle θ s Relative azimuth The center wavelength of the red band is λ. red Blue band center wavelength λ blue Center wavelength λ of water vapor absorption band absorb and reference band; the output is the atmospheric correction factor x for the red and blue bands and the water vapor absorption band. a ,x b ,x c Atmospheric transmittance T, x a ,x b ,x c L respectively p G b +G t The network structure information of the S neural network model is shown in Table 2.

[0081] Table 2

[0082]

[0083] Step S3: Obtain observation information from the data image, including the solar zenith angle θ. s Observing the zenith angle θ v Relative azimuth Image imaging time, latitude and longitude, and radiometric calibration coefficients, etc.; the apparent reflectance ρ' is calculated based on the radiometric calibration coefficients.

[0084] In step S3, the hyperspectral image requiring atmospheric correction is preprocessed to extract observation information, including the solar zenith angle θ. s Observing the zenith angle θ v Relative azimuth Image imaging time, latitude and longitude, and radiometric calibration coefficients; obtain calibration coefficients, and perform radiometric calibration calculations on hyperspectral image data to obtain apparent reflectance ρ'; extract the calibrated water vapor band, reference band, and aerosol inversion band images and center wavelengths, with wavelength selection reference table shown in Table 3.

[0085] Table 3

[0086]

[0087] Step S4: Based on the remote sensing image data, set the initial values ​​of atmospheric parameters, and use the atmospheric correction network model to collaboratively invert and obtain the atmospheric water vapor content (WVC) and visibility (AOT).

[0088] In step S4, refer to Figure 3 The specific process of establishing a parallel architecture for real-time monitoring and real-time computing is as follows.

[0089] In step S4, the normalized difference vegetation index (NDVI) and apparent reflectance at 2100 nm were used to determine the dark pixels in the hyperspectral image data, where the requirements for dark pixels are as follows:

[0090]

[0091] In the formula ρ' 860 ,ρ' 660 ,ρ' 2100 These represent the apparent reflectance of the hyperspectral image at 860 nm, 660 nm, and 2100 nm, respectively. The near-infrared band is minimally affected by aerosols and can be disregarded; therefore, the apparent reflectance at the 2100 nm band is used to approximate the surface reflectance. For dark pixels, 2100 nm exhibits the following linear relationship with the red and blue light bands (660 nm, 470 nm):

[0092] ρ 660 ≈ρ' 2100 / 2

[0093] ρ 470 ≈ρ' 2100 / 4

[0094] Attitude and orbit parameters and imaging conditions are acquired through real-time monitoring. Initial values ​​for visibility and water vapor content are set as AOT0 and WVC0, respectively. These parameters are fed back to the atmospheric correction neural network model, which outputs atmospheric correction coefficients for the red-blue band and water vapor absorption band. And atmospheric transmittance T.

[0095] For water vapor absorption bands and nearby reference bands, the surface reflectance changes similarly or approximately linearly. Therefore, the band ratio method can be used to approximate the atmospheric water vapor transmittance. This invention employs a three-band ratio method to invert atmospheric water vapor content:

[0096]

[0097] In the formula, ρ(λ), ρ(λ) up ),ρ(λ down ) represent the apparent reflectance of the water vapor absorption band and the reference band, respectively; T represents the atmospheric transmittance; c1 and c2 are the weights of the reflectance of the reference band, respectively.

[0098] The simulated ground reflectance ρ” is calculated based on the atmospheric correction coefficients and apparent radiance of the red and blue bands. 660 ,ρ” 470 If ρ” 660 and ρ 660 If the difference is not within the allowable range, then AOT needs to be adjusted. i 660 ;wherein, if ρ” 660 Less than ρ 660 Then AOT i 660 Increase σ1 660 If ρ” 660 Greater than ρ 660 Then AOT i 660 Decrease step size σ1 660 This continues until the difference between the two is within the allowable error range. If ρ” 470 and ρ 470 If the difference is not within the allowable range, then adjust AOT. i 470 , ρ 470 ≈ρ' 2100 / 4; if ρ” 470 Less than ρ 470 Then AOT i 470 Increase σ1 470 If ρ” 470 Greater than ρ 470 Then AOT i 470 Decrease step size σ1470 Continue this process until the difference between the two values ​​is within the acceptable error range. Then, average the visibility values ​​obtained from the red and blue bands. As a result of this inversion, AOT i .

[0099] If T is less than the transmittance T' obtained from the atmospheric correction network model, then the atmospheric water vapor content is WVC. i Increase σ2, and vice versa.

[0100] If the difference between the two results is within the allowable error range ε1, ε2, proceed to step S5; otherwise, repeat this step.

[0101] Step S5: Based on the remote sensing image observation information and the atmospheric parameters obtained in step S4, MODTRAN is used to calculate the atmospheric correction coefficients x for all bands in parallel. a ,x b ,x c , respectively corresponding to L p G b +G t 、S.

[0102] In step S5, based on the image observation information and the atmospheric parameters calculated in step S4, the atmospheric correction coefficients L for all bands of the hyperspectral image are calculated using MODTRAN by grouping the hyperspectral image bands. p G b +G t The atmospheric radiative transfer equation can be expressed as:

[0103]

[0104] In the formula, L represents the apparent radiance received by the sensor; L p Atmospheric path radiation; G b G t These represent the effects of direct atmospheric radiation and atmospheric scattering, respectively; S is atmospheric albedo; and ρ is surface reflectance or the true reflectance of ground features.

[0105] The MODTRAN radiative transfer model was used to simulate atmospheric radiative transfer, and three atmospheric correction factors L were obtained. p G b +G t By substituting the apparent reflectance obtained from the sensor into the above formula, the surface reflectance or the true reflectance of ground objects can be obtained, and atmospheric correction can be completed.

[0106] Step S6: Repeat steps S3-S5 to perform atmospheric correction on the next remote sensing image, using the inversion result from step S5 as input for the next stage to speed up the inversion process.

[0107] In step S6, for adjacent pixels in the hyperspectral image with similar atmospheric patterns and parameters, the atmospheric parameter result of this correction is used as the initial value for atmospheric correction in the next stage to accelerate the convergence speed of the collaborative inversion. In this invention, it cannot be guaranteed that there are dark pixels in the hyperspectral image. For adjacent pixels, they have the same ground features or similar surface reflectance. Therefore, the surface reflectance of the adjacent pixels in this correction is subtracted from the atmospheric correction pixel in the next stage to obtain the dark pixel.

[0108] The method of this invention achieves high efficiency and high precision in atmospheric correction of hyperspectral images by designing rolling correction of adjacent pixels, providing a technical method for real-time atmospheric correction of long strip data or spaceborne and airborne hyperspectral cameras.

[0109] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims of the present invention.

Claims

1. An embedded hyperspectral atmospheric correction processing method, characterized in that, include: Step 1: Using the MODTRAN radiative transfer model, establish a general lookup table for atmospheric correction for a specified band, which includes the aerosol inversion band, the water vapor absorption band, and the reference band. Step 2: Based on the general lookup table for atmospheric correction, construct a neural network model and obtain the optimal weight parameters of the neural network model through training; The neural network model is constructed, and the model input includes visibility. Atmospheric water vapor content Observing the zenith angle Solar zenith angle Relative azimuth Red band center wavelength Blue band center wavelength Center wavelength of water vapor absorption band and reference band; the output is the atmospheric correction coefficient for the red and blue bands and the water vapor absorption band. Atmospheric transmittance They are respectively , , ; Step 3, set the atmospheric water vapor content With visibility The observation information, along with the data from remote sensing images, is input into the neural network model, and the results output by the neural network model are used to determine... The corresponding atmospheric transmittance T' is obtained by comparing T' with the atmospheric transmittance reference value T and adjusting accordingly. Determine based on the output of the neural network model Corresponding simulated ground reflectance By comparison and To obtain by adjusting the inversion ; Repeat the inversion process until the difference between T' and T is less than the preset error. and If the difference is less than the preset error, the final atmospheric water vapor content obtained from the inversion is determined. With visibility ; This refers to surface reflectance or the true reflectance of ground features. Step 4: Based on the observation information in the remote sensing image data, compare it with the atmospheric water vapor content obtained through inversion. and visibility The atmospheric correction coefficients for all bands were calculated in parallel using MODTRAN. , respectively corresponding , , ; Atmospheric path radiation, , The effects of direct atmospheric radiation and scattering are respectively. Atmospheric albedo.

2. The method according to claim 1, characterized in that, In step 1, the atmospheric radiative transfer process is simulated using the MODTRAN radiative transfer model to obtain atmospheric transmittance, and then the atmospheric path radiation is calculated. , Large balloon surface albedo and radiation correction parameters The simulation only covered atmospheric transport processes in the aerosol inversion band, water vapor absorption band, and reference band. Water vapor content 0.3~5 g cm 2 Visibility 0~100km Solar zenith angle 0°~90° Observation zenith angle 0°~90° Relative azimuth angle 0°~180° Aerosol inversion bands include blue light band, red light band, and near-infrared band; The water vapor absorption bands include 1117~1143nm, 935~955nm, and 810~830nm; The upper reference band corresponding to the water vapor absorption band of 1117~1143nm includes 1184~1210nm, and the lower reference band corresponding to the water vapor absorption band of 1117~1143nm includes 1050~1067nm. The upper reference band corresponding to the water vapor absorption band of 935~955nm includes 870~890nm, and the lower reference band corresponding to the water vapor absorption band of 935~955nm includes 995~1020nm. The upper reference band corresponding to the water vapor absorption band of 810~830nm includes 850~870nm, and the lower reference band corresponding to the water vapor absorption band of 810~830nm includes 770~790nm.

3. The method according to claim 1, characterized in that, In step 1, the surface reflectance was measured separately. Simulations were performed for values ​​of 0, 0.25, and 0.

5. The atmospheric correction coefficients were solved using the following set of equations: The atmospheric correction factor is: In the formula , .

4. The method according to claim 1, characterized in that, In step 1, the MODTRAN table creation process based on a multi-core processor is as follows: Different observation conditions and atmospheric states are enumerated, and each case is distributed as a separate case to multiple threads or processes through instruction distribution to generate a Tape5 file; The lookup table is built by calling MODTRAN in parallel using a multi-core processor. In the process of hyperspectral remote sensing data processing, the parallel and for commands in OpenMP are used to decompose the task and start thread groups to handle multiple situations at the same time. After execution, the system reads the data from the result file Tape7, calculates the atmospheric correction factor, and packages the input conditions and calculation results into the database.

5. The method according to claim 1, characterized in that, The network structure of a neural network model includes: 1) Convolutional layer Conv1d3×1×32 2) Activation layer ReLU 3) Convolutional layer Conv1d3×32×32 4) Activation layer ReLU 5) Maxpooling layer Maxpool1d 6) Convolutional layer Conv1d 3×32×32 7) Activation layer ReLU 8) Convolutional layer Conv1d 3×32×32 9) Activation layer ReLU 10) Maxpooling layer Maxpool1d 11) Convolutional layer Conv1d 3×32×10 12) Activation layer ReLU.

6. The method according to claim 1, characterized in that, In step 3, the normalized difference vegetation index (NDVI) and apparent reflectance at 2100 nm are used to determine the dark pixels in the hyperspectral image data. The requirements for dark pixels are as follows: , , The apparent reflectance of the hyperspectral image at 860 nm, 660 nm, and 2100 nm are respectively used; atmospheric water vapor content is determined by selecting dark pixels. With visibility The inversion.

7. The method according to claim 6, characterized in that, In step 3, attitude and trajectory parameters and imaging conditions are acquired through real-time monitoring, and initial values ​​for visibility and water vapor content are set. ; The parameters are fed back to the neural network model, which outputs atmospheric correction coefficients for the red-blue band and the water vapor absorption band. and atmospheric transmittance ; The simulated ground reflectance was calculated based on the atmospheric correction coefficients and apparent radiance of the red and blue bands. ; if and If the difference is not within the allowable range, then adjust. , ;if Less than Then Increase Step size, if Greater than Then Decrease step size This continues until the difference between the two is within the allowable error range; if and If the difference is not within the allowable range, then adjust. , ;if Less than Then Increase Step size, if Greater than Then Decrease step size This continues until the difference between the two is within the allowable error range; Average visibility obtained from the red and blue bands As a result of this inversion ; For the reference band corresponding to the water vapor absorption band, the atmospheric transmittance reference value is... satisfy: In the formula, These are the apparent reflectances of the water vapor absorption band, the upper reference band, and the lower reference band, respectively. These are the reference band reflectance weights; if Less than the transmittance obtained by the neural network model The atmospheric water vapor content is Increase step size Conversely, it decreases.

8. The method according to claim 1, characterized in that, In step 4, based on the image observation information and the atmospheric parameters calculated in step 3, the atmospheric correction coefficients for all bands of the hyperspectral image are calculated using MODTRAN by grouping the hyperspectral image bands. , , The atmospheric radiative transfer equation is expressed as: In the formula The apparent radiance received by the sensor; Atmospheric path radiation; , The effects of direct atmospheric radiation and scattering are respectively considered. Atmospheric albedo; This refers to surface reflectance or the true reflectance of ground features. Combined with atmospheric correction factor , , Then, the apparent reflectance obtained from the sensor is substituted into the above formula to obtain the surface reflectance or the true reflectance of the ground object, and then atmospheric correction is completed.

9. The method according to claim 1, characterized in that, The method further includes: Step 5: Repeat steps 3-4 to perform atmospheric correction on the next remote sensing image, using the inversion result from step 4 as input for the next stage to speed up the inversion process.

10. The method according to claim 1, characterized in that, In step 5, for adjacent pixels with the same ground features or similar surface reflectance, the surface reflectance of the atmospheric-corrected adjacent pixels in the previous remote sensing image is subtracted from the atmospheric-corrected neighboring pixels in the current remote sensing image to obtain the dark pixel.

Images