Training Method and Atmospheric Correction Method for Hyperspectral Thermal Infrared Atmospheric Correction Model

Abstract

The present disclosure provides a training method for a hyperspectral thermal infrared atmospheric correction model, which can be applied to the field of remote sensing technology. The method includes: inputting hyperspectral thermal infrared image simulation data into an attention module to obtain predicted weight values corresponding to the simulated values of the initial pupil radiation parameters for multiple bands respectively; based on the simulated values of the initial pupil radiation parameters for multiple bands respectively and the multiple predicted weight values, obtaining candidate input predicted values corresponding to multiple bands respectively; determining a target input predicted value from the candidate input predicted values corresponding to multiple bands respectively; inputting the target input predicted value into a prediction module to obtain a predicted value of the atmospheric radiation parameter for the target output band; based on the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band, respectively adjusting the parameters of the attention module and the prediction module to obtain a trained hyperspectral thermal infrared atmospheric correction model.

Description

Technical Field

[0001] The present disclosure relates to the field of remote sensing technology, and in particular to a training method for a hyperspectral thermal infrared atmospheric correction model and an atmospheric correction method. Background Art

[0002] Thermal infrared remote sensing is the main means of obtaining infrared radiation characteristics of global or regional targets. With the advancement of satellite and sensor technology, hyperspectral remote sensing has developed into the thermal infrared band. Hyperspectral thermal infrared remote sensing data contains richer ground and atmosphere information, which is conducive to obtaining accurate atmospheric radiation parameters.

[0003] In the process of obtaining atmospheric radiation parameters through hyperspectral thermal infrared remote sensing data, atmospheric correction is usually required. During the research process, the inventors found that the atmospheric correction methods in related technologies usually have problems such as strict assumptions and poor performance in atmospheric correction for medium and low spatial resolution hyperspectral thermal infrared remote sensing data. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a training method for a hyperspectral thermal infrared atmosphere correction model and an atmosphere correction method, device, equipment, medium and program product.

[0005] According to one aspect of the present disclosure, a training method for a hyperspectral thermal infrared atmosphere correction model is provided, comprising: inputting hyperspectral thermal infrared image simulation data into an attention module to obtain prediction weight values ​​corresponding to initial entrance pupil radiation parameter simulation values ​​of multiple bands, wherein the hyperspectral thermal infrared image simulation data includes initial entrance pupil radiation parameter simulation values ​​of multiple bands; based on the initial entrance pupil radiation parameter simulation values ​​of the multiple bands and the prediction weight values ​​corresponding to the initial entrance pupil radiation parameter simulation values ​​of the multiple bands, obtaining candidate input prediction values ​​corresponding to the multiple bands; determining a target input prediction value from the candidate input prediction values ​​corresponding to the multiple bands; inputting the target input prediction value into a prediction module to obtain an atmospheric radiation parameter prediction value of a target output band, wherein the target output band is a band containing the highest effective information within the target band range; based on the atmospheric radiation parameter prediction value and the atmospheric radiation parameter simulation value corresponding to the target output band, adjusting parameters of the attention module and the prediction module respectively to obtain a trained hyperspectral thermal infrared atmosphere correction model.

[0006] Another aspect of the present disclosure provides a hyperspectral thermal infrared atmosphere correction method, characterized in that the method includes: inputting hyperspectral thermal infrared image data into a trained hyperspectral thermal infrared atmosphere correction model to obtain a predicted value of atmospheric radiation parameters of a target output band.

[0007] Another aspect of the present disclosure provides a training device for a hyperspectral thermal infrared atmospheric correction model, including: a weight determination module, configured to input hyperspectral thermal infrared image simulation data into an attention module to obtain predicted weight values corresponding to initial pupil radiation parameter simulation values of multiple bands respectively, where the hyperspectral thermal infrared image simulation data includes initial pupil radiation parameter simulation values of multiple bands respectively; a first determination module, configured to obtain candidate input predicted values corresponding to multiple bands respectively based on the initial pupil radiation parameter simulation values of multiple bands and the predicted weight values corresponding to the initial pupil radiation parameter simulation values of multiple bands respectively; a second determination module, configured to determine a target input predicted value from the candidate input predicted values corresponding to multiple bands respectively; an atmospheric parameter determination module, configured to input the target input predicted value into a prediction module to obtain a predicted value of the atmospheric radiation parameter of a target output band, where the target output band is the band with the highest effective information in a target band range; an output module, configured to adjust the parameters of the attention module and the prediction module respectively based on the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band to obtain a trained hyperspectral thermal infrared atmospheric correction model.

[0008] Another aspect of the present disclosure provides a hyperspectral thermal infrared atmospheric correction device, characterized in that the method includes: a correction module, configured to input hyperspectral thermal infrared image data and hyperspectral thermal infrared image simulation data into a trained hyperspectral thermal infrared atmospheric correction model to obtain a predicted value of the atmospheric radiation parameter of a target output band.

[0009] Another aspect of the present disclosure provides an electronic device, including: one or more processors; a memory, configured to store one or more computer programs, where the above one or more processors execute the above one or more computer programs to implement the steps of the above method.

[0010] Another aspect of the present disclosure further provides a computer-readable storage medium, on which a computer program or instruction is stored, and when the computer program or instruction is executed by a processor, the steps of the above method are implemented.

[0011] Another aspect of the present disclosure further provides a computer program product, including a computer program or instruction, and when the computer program or instruction is executed by a processor, the steps of the above method are implemented. Description of the Drawings

[0012] Through the following description of the embodiments of the present disclosure with reference to the drawings, the above content and other objects, features and advantages of the present disclosure will become clearer. In the drawings:

[0013] Figure 1Schematically shows an application scenario diagram of a training method for a hyperspectral thermal infrared atmospheric correction model and a hyperspectral thermal infrared atmospheric correction method, apparatus, device, medium, and program product according to an embodiment of the present disclosure;

[0014] Figure 2 Schematically shows a flowchart of a training method for a hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure;

[0015] Figure 3 Schematically shows a schematic diagram of a hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure;

[0016] Figure 4 Schematically shows a schematic diagram of a training method for a hyperspectral thermal infrared atmospheric correction model according to another embodiment of the present disclosure.

[0017] Figure 5 Schematically shows a structural block diagram of a training apparatus for a hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure; and

[0018] Figure 6 Schematically shows a block diagram of an electronic device suitable for implementing a training method for a hyperspectral thermal infrared atmospheric correction model and a hyperspectral thermal infrared atmospheric correction method according to an embodiment of the present disclosure. Detailed implementation manners

[0019] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present disclosure.

[0020] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components. All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0021] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0022] The inventor discovered during the research that thermal infrared remote sensing is the main means of obtaining infrared radiation characteristics of global or regional targets, and a multi-series, multi-sensor satellite networking system has been formed. Multispectral thermal infrared remote sensing has a history of nearly 50 years of development, and has derived many thermal infrared surface temperature products, but multispectral thermal infrared sensors have fewer channels and provide limited information. Under specific conditions such as high atmospheric water vapor content, it is difficult to obtain satisfactory accuracy in the inversion results. With the advancement of satellite and sensor technology, hyperspectral remote sensing has developed to the thermal infrared band. Hyperspectral thermal infrared remote sensing data contains richer ground-atmosphere information, which is conducive to the accurate inversion of ground-atmosphere parameters. In recent years, satellites equipped with hyperspectral thermal infrared sensors such as the Infrared Atmospheric Sounding Interferometer (IASI), the Cross-Track Infrared Sensor (CrIS), and the Infrared Hyperspectral Atmospheric Vertical Sounder (HIRAS) have been launched and put into operation at home and abroad, and a large amount of data has been accumulated. However, the development of hyperspectral remote sensing technology in the thermal infrared spectrum is still lagging behind. Using the advantages of hyperspectral technology to invert ground-atmosphere characteristic parameters has become a major focus of hyperspectral thermal infrared research.

[0023] High-precision atmospheric correction is a necessary condition for obtaining accurate surface temperature and emissivity. The nanometer-level resolution of the hyperspectral spectrum means strong correlation between channels, which makes the hyperspectral thermal infrared data more seriously affected by the atmosphere. In essence, it no longer conforms to the mathematical derivation assumptions of traditional multispectral thermal infrared atmospheric correction methods such as the split window algorithm, and the difficulty of atmospheric correction has increased. At present, the atmospheric correction methods specifically for hyperspectral thermal infrared data appeared relatively late and have achieved few results. The widely used ones are mainly the automatic atmospheric compensation method (Autonomous Atmospheric Compensation, AAC), the scene correction method (In-Scene Atmospheric Correction, ISAC) and the atmospheric correction method based on atmospheric profiles. Among them, the AAC and ISAC methods have problems such as strict assumptions and poor performance on medium and low spatial resolution data; the atmospheric correction method based on atmospheric profiles requires the introduction of auxiliary data, and the radiation transfer model needs to be repeatedly called for multiple images, resulting in complicated calculations.

[0024] Figure 1 Schematically shows an application scenario diagram of a training method of a hyperspectral thermal infrared atmospheric correction model and a hyperspectral thermal infrared atmospheric correction method, apparatus, device, medium, and program product according to an embodiment of the present disclosure.

[0025] As Figure 1 shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0026] Users may use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0027] The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, and desktop computers, etc.

[0028] The server 105 may be a server providing various services, such as a background management server that supports websites browsed by users using the terminal devices 101, 102, 103 (only as an example). The background management server may analyze and process data such as received user requests, etc., and feedback the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0029] It should be noted that the training method of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction method provided by the embodiments of the present disclosure can generally be executed by the server 105. Correspondingly, the training device of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction device provided by the embodiments of the present disclosure can generally be set in the server 105. The training method of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction method provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and / or the server 105. Correspondingly, the training device of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction device provided by the embodiments of the present disclosure can also be set in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and / or the server 105.

[0030] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0031] are merely illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers. Figure 1 The following will be based on Figures 2 to 4 the described scenario to describe in detail the training method of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction method of the disclosure embodiments through

[0032] Figure 2 FIG. schematically shows a flowchart of the training method of the hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure.

[0033] As Figure 2 shown, the method includes operation S210 to operation S250.

[0034] In operation S210, the hyperspectral thermal infrared image simulation data is input into the attention module to obtain predicted weight values corresponding to the simulated values of the initial pupil radiation parameters for multiple bands, where the hyperspectral thermal infrared image simulation data includes the simulated values of the initial pupil radiation parameters for each of the multiple bands.

[0035] According to an embodiment of the present disclosure, the hyperspectral thermal infrared image simulation data can be data simulated by a hyperspectral thermal infrared payload sensor, and the hyperspectral thermal infrared image simulation data can be used as a training sample to train the hyperspectral thermal infrared atmospheric correction model.

[0036] According to an embodiment of the present disclosure, a band can also be referred to as a channel.

[0037] According to an embodiment of the present disclosure, the hyperspectral thermal infrared payload sensor may be a spaceborne hyperspectral thermal infrared payload that meets the spectral resolution requirements, has a relatively high calibration accuracy, and stable performance.

[0038] According to an embodiment of the present disclosure, the hyperspectral thermal infrared atmospheric correction model includes an attention module and a prediction module, and the connection manner between the two is not limited. For example, the attention module is embedded in the front end of the prediction module.

[0039] According to an embodiment of the present disclosure, the simulated value of the initial pupil radiation parameter may be the simulated value of the radiation brightness temperature value at the pupil of the sensor, or may also be the simulated value of the radiation brightness value at the pupil of the sensor.

[0040] According to an embodiment of the present disclosure, the hyperspectral thermal infrared image simulation data includes simulated values of initial pupil radiation parameters in different bands. The simulated values of the initial pupil radiation parameters may include simulated values of initial pupil radiation pixels under different underlying surfaces and / or different ground temperatures. Specifically, multiple simulated values of initial pupil radiation pixels in the same band and under the same underlying surface are obtained at different surface temperatures, and vice versa, multiple simulated values of initial pupil radiation pixels in the same band and at the same surface temperature are obtained under different underlying surfaces.

[0041] According to an embodiment of the present disclosure, the attention module is not limited to being a Convolutional Block Attention Module (CBAM), and specifically, it may be the channel attention module in CBAM.

[0042] According to an embodiment of the present disclosure, by using the attention module to extract spectral features from the hyperspectral thermal infrared image simulation data, prediction weight values corresponding to the simulated values of the initial pupil radiation parameters in multiple bands can be obtained, so as to determine the importance degree of the simulated values of the initial pupil radiation parameters corresponding to multiple bands, that is, the higher the prediction weight value, the higher the importance degree of the simulated value of the initial pupil radiation parameter.

[0043] In operation S220, based on the simulated values of the initial pupil radiation parameters of multiple bands and the prediction weight values corresponding to the simulated values of the initial pupil radiation parameters of multiple bands, candidate input prediction values corresponding to multiple bands are obtained.

[0044] According to an embodiment of the present disclosure, for each simulated value of the initial entrance pupil radiation parameter, candidate input prediction values corresponding to multiple bands can be obtained by multiplying the simulated value of the initial entrance pupil radiation parameter and the prediction weight values corresponding to the simulated values of the initial entrance pupil radiation parameters of the multiple bands band by band, that is, multiplying the simulated value of the initial entrance pupil radiation parameter in the a band and the prediction weight value corresponding to the simulated value of the initial entrance pupil radiation parameter in the a band band by band.

[0045] In operation S230, a target input prediction value is determined from the candidate input prediction values corresponding to multiple bands respectively.

[0046] According to an embodiment of the present disclosure, based on the candidate input prediction values, an importance value of the candidate input prediction value is determined. Based on the importance values of the multiple candidate input prediction values, a target input prediction value is determined from the multiple candidate input prediction values, and the target input prediction value is a candidate input prediction value with a higher importance value. Among them, the importance value can be a quantization value of the importance degree of the candidate input prediction value.

[0047] According to an embodiment of the present disclosure, candidate input prediction values with lower importance degrees are screened out, which can reduce the processing amount of the prediction module. Only the valid data, that is, the target input prediction value, is input into the prediction module, thereby realizing the dimensionality reduction of the input data and accelerating the processing efficiency of the prediction module.

[0048] In operation S240, the target input prediction value is input into the prediction module to obtain a predicted value of the atmospheric radiation parameter in the target output band, where the target output band is the band with the highest effective information in the target band range.

[0049] According to an embodiment of the present disclosure, the prediction module can be composed of a neural network model. Specifically, it can be a residual neural network (ResNet), for example: the ResNet18 network structure, or other neural network models such as a perceptron, a support vector machine, an eXtreme Gradient Boosting (abbreviated as XGBoost), and a random forest.

[0050] According to an embodiment of the present disclosure, the atmospheric radiation parameter value can include: atmospheric transmittance, upward and downward atmospheric radiation values. The predicted value of the atmospheric radiation parameter is the predicted value of the atmospheric radiation parameter value by the prediction module based on the simulated data of the hyperspectral thermal infrared image, and the predicted value of the atmospheric radiation parameter is a value after atmospheric correction.

[0051] According to an embodiment of the present disclosure, the target band range can be a band sensitive to surface parameters, that is, one band or a continuous multiple bands in the bands that are more sensitive to surface parameters. The target band range can be one or more. The target output band can be the band with the highest effective information determined from the target band range.

[0052] According to an embodiment of the present disclosure, the target output band can be a band suitable for subsequent separation of land surface temperature and land surface emissivity, and the land surface parameters can include land surface temperature, land surface reflectivity, etc.

[0053] According to an embodiment of the present disclosure, inputting the target input prediction value into the prediction module can obtain the value of the atmospheric radiation parameter corresponding to the band with the highest effective information in the target band range, thereby realizing the output of the predicted value of the atmospheric radiation parameter after atmospheric correction while outputting the effective band.

[0054] According to an embodiment of the present disclosure, by not distinguishing between the atmospheric window and the non-atmospheric window for the simulated data of the hyperspectral thermal infrared image, and inputting the simulated values of the initial pupil radiation parameters of both the atmospheric window and the non-atmospheric window into the hyperspectral thermal infrared atmospheric correction model, the atmospheric correction accuracy and data utilization rate of the hyperspectral thermal infrared atmospheric correction model can be effectively improved.

[0055] According to an embodiment of the present disclosure, due to the adoption of the attention module and the prediction module, and the prediction module can be obtained from a neural network model, it is possible to fully learn and represent the quantitative relationship between the input data, so that it has a more accurate effect in dealing with high-complexity and non-linear calculation problems such as atmospheric correction of the simulated data of the hyperspectral thermal infrared image.

[0056] In operation S250, based on the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band, the parameters of the attention module and the prediction module are adjusted respectively to obtain a trained hyperspectral thermal infrared atmospheric correction model.

[0057] According to an embodiment of the present disclosure, since the simulated value of the atmospheric radiation parameter is data obtained by simulating a real hyperspectral thermal infrared payload sensor, therefore, the data obtained by simulating the hyperspectral thermal infrared payload sensor is equivalent to the data after relatively accurate atmospheric correction. Therefore, the simulated value of the atmospheric radiation parameter can be used as the true value corresponding to the simulated data of the hyperspectral thermal infrared image.

[0058] According to an embodiment of the present disclosure, by inputting the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band into the loss function, a loss value can be obtained. Then, based on the loss value, the parameters of the attention module and the prediction module are adjusted respectively to obtain a trained attention module and a trained prediction module, thereby obtaining a trained hyperspectral thermal infrared atmospheric correction model.

[0059] According to an embodiment of the present disclosure, during the training process of the hyperspectral thermal infrared atmospheric correction model, it can be realized by a data set including multiple hyperspectral thermal infrared image simulation data. Specifically, the data set can be divided, and 80% of the hyperspectral thermal infrared image simulation data under each atmospheric condition and each observation zenith angle and the simulated values of the atmospheric radiation parameters of their respective corresponding target output bands are used as the training set.

[0060] According to an embodiment of the present disclosure, the loss function for calculating the loss value is not limited. The sum of the squares of the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter can be used as the model loss function, specifically as shown in the following formulas (1) and (2):

[0061] ; (1)

[0062] ; (2)

[0063] Among them, and are respectively the simulated value and the predicted value of the atmospheric radiation parameter of the band , is the total number of bands of the target input prediction value, is the activation function, is the total number of neurons, is the input of the th neuron, is the corresponding weight, is the bias.

[0064] According to an embodiment of the present disclosure, the model loss function is not limited. In addition to the above loss function, the mean absolute error loss (L1) function, the anchor box loss function, the smooth L1 loss function, etc. can also be used.

[0065] According to an embodiment of the present disclosure, by inputting the simulated data of the hyperspectral thermal infrared image into the attention module, the prediction weight values, i.e., the importance, of the initial pupil radiation parameter simulation values of multiple bands are determined respectively. And the candidate input prediction values corresponding to the multiple bands are obtained through the multiple prediction weight values and the initial pupil radiation parameter simulation values corresponding to the multiple prediction weight values respectively, and the target input parameters are determined from the multiple candidate input prediction values, that is, the screening of the input data with higher importance is realized, and thus the dimensionality reduction of the input data of the prediction module and the improvement of the output accuracy are realized; and the target input parameters are input into the prediction module to obtain the atmospheric radiation parameter values of the target output band, that is, the atmospheric radiation parameter values of the band with high effective information content are obtained as the corrected atmospheric radiation parameter values. Therefore, at least partially, the technical problems in the related art such as the need for strict assumption conditions and poor performance of atmospheric correction are solved, and thus the technical effect of obtaining more accurate atmospheric radiation parameters without strict assumption conditions is realized.

[0066] According to an embodiment of the present disclosure, the above method may further include the following operations.

[0067] Input the target atmospheric profile data, the preset payload observation geometry, and the preset spectral response function into the thermal infrared radiation transfer model to obtain the simulated values of the atmospheric radiation parameters corresponding to the multiple bands respectively; based on the simulated values of the atmospheric radiation parameters corresponding to the multiple bands respectively, the multiple surface temperatures, the multiple surface emissivities, and the thermal infrared radiation transfer equation, obtain the initial pupil radiation parameter simulation values of the multiple bands respectively; based on the initial pupil radiation parameter simulation values of the multiple bands respectively, obtain the simulated data of the hyperspectral thermal infrared image.

[0068] According to an embodiment of the present disclosure, the target atmospheric profile data includes the data of the atmospheric conditions under clear and cloudless skies, for example: the target atmospheric profile data includes the air pressure values, temperature values, atmospheric density, water vapor content, ozone content, etc. at different heights.

[0069] According to an embodiment of the present disclosure, the thermal infrared radiation transfer model is not limited and can be any model that can simulate the simulated values of the atmospheric radiation parameters. For example: it can be the hyperspectral thermal infrared radiation transfer model 4A / OP. 4A / OP is a fast and accurate line-by-line radiation transfer model. Its core is a huge database of single-molecule monochromatic optical thickness GEISA. This database only needs to be generated once and can be used permanently. 4A / OP relies on this database and adopts a line-by-line and layer-by-layer calculation method to obtain quite accurate atmospheric radiation parameters with high spectral resolution.

[0070] According to an embodiment of the present disclosure, by inputting the simulated values of the atmospheric radiation parameters, the surface reflectivity, and the surface temperature into the thermal infrared radiation transfer equation, the initial pupil radiation parameter simulation values can be obtained.

[0071] According to an embodiment of the present disclosure, the thermal infrared radiation transfer equation is not limited and can be any equation capable of obtaining the simulated value of the initial pupil radiation parameter, as shown in formulas (3) and (4).

[0072]

[0073]

[0074] Among them, is the wavelength, is the simulated value of the initial pupil radiation parameter, is the radiation from the ground, is the surface emissivity of the target surface, is the surface temperature of the target surface, is the Planck function, is the atmospheric transmittance, and are the upward and downward atmospheric radiations respectively.

[0075] According to an embodiment of the present disclosure, the quantities of the target atmospheric profile data, the preset payload observation geometry, the preset spectral response function, the surface temperature, and the surface emissivity are not limited and can all be multiple.

[0076] According to an embodiment of the present disclosure, the surface temperature can be obtained from the clear-sky and cloud-free atmospheric profile.

[0077] According to an embodiment of the present disclosure, the surface emissivity can be obtained from the spectral library, and the spectral library is not limited. For example, it can be the JHU spectral library.

[0078] According to an embodiment of the present disclosure, the method for obtaining the target atmospheric profile data is not limited and can be to screen the clear-sky and cloud-free profiles from the atmospheric profile database through an atmospheric profile database with a wide and uniform spatio-temporal distribution for subsequent atmospheric correction.

[0079] According to an embodiment of the present disclosure, the relative humidity can be used as a judgment index. When there is a layer of atmospheric relative humidity greater than 90% or two consecutive layers of atmospheric relative humidity greater than 85%, the atmospheric profile data is determined to be a cloudy atmospheric profile. By eliminating the cloudy profiles, the clear-sky and cloud-free atmospheric profile data can be obtained.

[0080] According to an embodiment of the present disclosure, a suitable hyperspectral thermal infrared payload is selected as the target sensor. Based on the clear-sky and cloud-free atmospheric profile, combined with parameters such as the preset payload observation geometry and the preset spectral response function, the simulated value of the initial pupil radiation parameter and the simulated value of the atmospheric radiation parameter can be obtained.

[0081] According to an embodiment of the present disclosure, according to the bottom layer air temperature of the atmospheric profile Set the surface temperature. For a high-temperature atmospheric profile ( greater than 300 K), the change in surface temperature relative to is from -4 K to 6 K. For a normal-temperature atmospheric profile ( between 270 K and 300 K), it is from -5 K to 5 K. For a low-temperature atmosphere ( less than 270 K) it is from -6 K to 4 K.

[0082] According to an embodiment of the present disclosure, the simulated data of the hyperspectral thermal infrared image obtained through the above process may include the simulated values of the initial pupil radiation parameters for each of the multiple bands. Among the simulated values of the initial pupil radiation parameters, there are also the simulated values of the initial pupil radiation pixels with different underlying surfaces and different surface temperatures. The simulated value of the initial pupil radiation pixel can also be the simulated value of the radiation brightness temperature at the pupil of the sensor at a certain underlying surface and a certain surface temperature in a certain band, and can also be the simulated value of the radiation brightness value at the pupil of the sensor.

[0083] According to an embodiment of the present disclosure, through the above process, the relationship between the simulated values of the atmospheric radiation parameters and the simulated values of the initial pupil radiation parameters in each band under different atmospheric conditions and different observation angles can be established.

[0084] According to an embodiment of the present disclosure, noise screening can be performed on the simulated values of the initial pupil radiation parameters for each of the multiple bands, and the simulated values of the initial pupil radiation parameters corresponding to the bands with noise greater than the preset noise are screened out. Specifically, the calculation formula for the noise of each band can be as shown in the following formula (5), where can be the noise equivalent temperature difference, that is, it refers to the noise value, and the simulated value of the initial pupil radiation parameter can be the simulated value of the radiation brightness temperature at the pupil.

[0085] ; (5)

[0086] where refers to the noise value obtained at the simulated value of the initial pupil radiation parameter at the band , is the Planck radiance, and are the reference noise and the reference temperature respectively. The reference temperature is set to 280 K, and the reference noise is set with reference to the spectral parameters of the Infrared Atmospheric Sounding Interferometer (IASI) instrument.

[0087] According to an embodiment of the present disclosure, based on the noise values of the simulated values of the initial entrance pupil radiation parameters in each band, the simulated values of the initial entrance pupil radiation parameters with large noise values are deleted, thereby improving the training accuracy of the subsequent model.

[0088] According to an embodiment of the present disclosure, the above method may further include the following operations.

[0089] Based on a channel division criterion, the simulated values of the initial entrance pupil radiation parameters of each of the multiple bands are divided into a first simulated value of the entrance pupil radiation parameter corresponding to the atmospheric parameter sensitive band and a second simulated value of the entrance pupil radiation parameter corresponding to the surface parameter sensitive band, where the channel division criterion includes the sensitive types to which each of the multiple bands belongs, and the sensitive types include the atmospheric parameter sensitive type and the surface parameter sensitive type; multiplying the first simulated value of the entrance pupil radiation parameter by a first preset weight value to obtain a weighted first simulated value of the entrance pupil radiation parameter; multiplying the second simulated value of the entrance pupil radiation parameter by a second preset weight value to obtain a weighted second simulated value of the entrance pupil radiation parameter, where the first preset weight value is greater than the second preset weight value; inputting the weighted first simulated value of the entrance pupil radiation parameter and the weighted second simulated value of the entrance pupil radiation parameter into an attention module to obtain prediction weight values corresponding to the simulated values of the initial entrance pupil radiation parameters of each of the multiple bands.

[0090] According to an embodiment of the present disclosure, the channel division criterion includes the sensitive types to which each of the multiple bands belongs, and based on this sensitive type, the multiple bands can be divided into the atmospheric parameter sensitive bands and the surface parameter sensitive bands.

[0091] According to an embodiment of the present disclosure, since the atmospheric sensitive parameters usually include more effective information, making the first preset weight value greater than the second preset weight value, that is, assigning a high weight to the atmospheric sensitive parameters to highlight their contribution to the atmospheric correction parameters, can make the candidate input prediction values with greater importance more prominent among the multiple candidate input prediction values obtained by subsequently inputting the weighted first simulated value of the entrance pupil radiation parameter and the weighted second simulated value of the entrance pupil radiation parameter into the trained attention module, thereby improving the accuracy of the predicted value of the output atmospheric radiation parameter.

[0092] According to an embodiment of the present disclosure, during the training process of the hyperspectral thermal infrared atmospheric correction model, the first preset weight parameter value and the second preset weight parameter value can be used as one of the hyperparameters, and the first preset weight parameter value and the second preset weight parameter value are continuously adjusted during the training process to obtain a better result.

[0093] According to an embodiment of the present disclosure, the above method may further include the following operations.

[0094] Input the perturbed target atmospheric profile data, the preset payload observation geometry, and the preset spectral response function into the thermal infrared radiation transfer model to obtain the simulated values of the perturbed atmospheric radiation parameters corresponding to each of multiple bands; based on the simulated values of the perturbed atmospheric radiation parameters for each of the multiple bands, the surface temperature, the surface emissivity, and the thermal infrared radiation transfer equation, obtain the simulated values of the third pupil radiation parameters for each of the multiple bands; based on the simulated values of the multiple atmospheric radiation parameters, the multiple perturbed surface temperatures, and the multiple perturbed surface emissivities, and the thermal infrared radiation transfer equation, obtain the simulated values of the fourth pupil radiation parameters for each of the multiple bands; based on the simulated values of the third pupil radiation parameters for each of the multiple bands, the simulated values of the fourth pupil radiation parameters for each of the multiple bands, and the simulated values of the initial pupil radiation parameters for each of the multiple bands, obtain the interference signal-to-noise ratios corresponding one-to-one to the multiple bands; compare the interference signal-to-noise ratios corresponding one-to-one to the multiple bands with the preset thresholds respectively to determine the channel division criteria.

[0095] According to an embodiment of the present disclosure, one or more main atmospheric parameters such as the atmospheric temperature, water vapor, carbon dioxide (CO2) content, or ozone (O3) content in the target atmospheric profile data can be adjusted based on the true parameter values included in the target atmospheric profile data. For example, the true parameter value can be replaced with a preset parameter value, or the true value can be increased or decreased, so as to realize the process of adding perturbations to the target atmospheric profile data.

[0096] According to an embodiment of the present disclosure, the method of adding perturbations to the surface temperature and the surface emissivity is not limited. It can be replacing the surface temperature with a preset surface temperature, replacing the surface emissivity with a preset surface emissivity, or increasing or decreasing based on the original surface temperature or the original surface emissivity.

[0097] According to an embodiment of the present disclosure, for the simulated value of the third pupil radiation parameter of each band, compare the simulated value of the third pupil radiation parameter with the simulated value of the initial pupil radiation parameter in the same band as the simulated value of the third pupil radiation parameter. The simulated value of the initial pupil radiation parameter in the same band as the simulated value of the third pupil radiation parameter can be the simulated value of the initial pupil radiation parameter that can be obtained before adding perturbations to the target atmospheric profile data of the simulated value of the third pupil radiation parameter.

[0098] According to an embodiment of the present disclosure, for the simulated value of the fourth pupil radiation parameter of each band, compare the simulated value of the fourth pupil radiation parameter with the simulated value of the initial pupil radiation parameter in the same band as the simulated value of the fourth pupil radiation parameter to obtain a second comparison value. The simulated value of the initial pupil radiation parameter in the same band as the simulated value of the fourth pupil radiation parameter can be the simulated value of the initial pupil radiation parameter that can be obtained before adding perturbations to the surface temperature and the surface emissivity.

[0099] According to an embodiment of the present disclosure, based on a first comparison value and a second comparison value of the same waveband, an interference signal-to-noise ratio corresponding to the waveband is obtained. Specifically, the formula for determining the interference signal-to-noise ratio can be as shown in the following formula (6).

[0100] ; (6)

[0101] Wherein, STI is the interference signal-to-noise ratio, ( = , , , ) respectively represent the first comparison values obtained after adding perturbations to the atmospheric temperature, water vapor, CO 2 and O 3 respectively, and the second comparison value obtained after adding perturbations to the surface temperature and surface emissivity is ([ = LST&LSE). ( =LST&LSE) represents the second comparison value obtained after adding perturbations to the surface temperature and surface emissivity.

[0102] According to an embodiment of the present disclosure, the first comparison value and the second comparison value can be the brightness temperature change values obtained by comparing the simulated values of the third and fourth pupil radiation parameters after adding perturbations with the simulated value of the initial pupil radiation parameter. The brightness temperature change value generated by adding perturbations to the surface temperature and emissivity is used as the signal, and the brightness temperature change value generated by adding perturbations to the remaining atmospheric parameters is used as the noise. The interference signal-to-noise ratio is obtained based on the ratio of the sum of the signal and the noise.

[0103] According to an embodiment of the present disclosure, based on the interference signal-to-noise ratio corresponding to the simulated value of the initial pupil radiation parameter, the division of atmospheric sensitive parameters and surface sensitive parameters can be realized. Specifically, the sensitive type of the waveband corresponding to the interference signal-to-noise ratio less than or equal to the preset threshold is used as the sensitive type of atmospheric parameters, and the sensitive type of the waveband corresponding to the interference signal-to-noise ratio greater than the preset threshold is used as the sensitive type of atmospheric parameters. The preset threshold can be 1.

[0104] According to an embodiment of the present disclosure, the channel division standard can be obtained in advance to facilitate more convenient use in subsequent applications.

[0105] According to an embodiment of the present disclosure, the above method may include the following operations.

[0106] Based on the correlation between the simulated values of the initial pupil radiation parameters with adjacent wavebands, a correlation curve is determined; based on the correlation curve, multiple wavebands are divided into multiple target waveband ranges; for each waveband in the target waveband range, based on the simulated value of the initial pupil radiation parameter corresponding to each waveband, the information entropy of each waveband is determined; based on the information entropy within each waveband in the target waveband range, the target output waveband is determined.

[0107] According to an embodiment of the present disclosure, a band selection algorithm based on nearest neighbor subspace partitioning can be used to screen fluctuations, so as to determine the band with the highest effective information in the target band range, that is, the target output band.

[0108] According to an embodiment of the present disclosure, the correlation between the simulated values of the initial pupil radiation parameters of adjacent bands among multiple bands can be determined first. Specifically, the correlation calculation formula can be as shown in formula (7).

[0109] ; (7)

[0110] Wherein, represents the value of the correlation between the simulated values of the initial pupil radiation parameters of two adjacent bands, (i = 1, 2,..., C - 1) represents the band matrix, P and Q respectively represent the width and height of the matrix composed of multiple simulated values of the initial pupil radiation parameters, and C is the number of bands, is the mean value of the simulated values of the initial pupil radiation pixels in band i, that is, the pixel mean value, and w and h are the coordinates of the simulated values of the initial pupil radiation pixels in band i.

[0111] According to an embodiment of the present disclosure, the hyperspectral thermal infrared image simulation data can be hyperspectral image cube (Hyperspectral Image, HSI) simulation data, that is, three-dimensional data. Its z-axis represents the band, the x-axis represents the underlying surface, and the y-axis represents the surface temperature. Then, w and h represent the coordinates of the simulated values of the initial pupil radiation pixels in band i, that is, the positions of the simulated values of the initial pupil radiation pixels on the x and y axes.

[0112] According to an embodiment of the present disclosure, based on the correlation between multiple adjacent bands, a correlation curve can be generated with the band as the abscissa and the correlation as the ordinate.

[0113] According to an embodiment of the present disclosure, the maximum and minimum values are determined in the correlation curve, and the maximum or minimum value is used as the interval point, so as to group multiple simulated values of the initial pupil radiation parameters to obtain at least one target band range, and the target band range can include at least one band.

[0114] According to an embodiment of the present disclosure, for each band in each target band range, the information entropy of the band is calculated, and the calculation formula is as shown in formula (8) below.

[0115] ; (8)

[0116] Wherein, represents the entire sample space, Indicates the probability that the simulated value i of the initial entrance pupil radiation parameter appears in band j.

[0117] According to an embodiment of the present disclosure, the higher the information entropy of a band, the more effective information is considered to be contained in that band. Thus, the band with the highest information entropy is the band containing the highest amount of effective information, i.e., the target output band.

[0118] According to an embodiment of the present disclosure, it can be obtained by determining the simulated value of the atmospheric radiation parameter corresponding to the target output band from the simulated values of the atmospheric radiation parameters corresponding to multiple bands respectively.

[0119] According to an embodiment of the present disclosure, taking the simulated value of the atmospheric radiation parameter corresponding to the target output band as the true value corresponding to the hyperspectral thermal infrared image simulation data can, on the one hand, reduce the output redundancy during the training process of the hyperspectral thermal infrared atmospheric correction model, and on the other hand, facilitate determining one or more bands containing the highest or relatively high amount of effective information, which is convenient for subsequent analysis.

[0120] According to an embodiment of the present disclosure, the attention module includes: a pooling layer, a shared network layer, and an addition layer; inputting the hyperspectral thermal infrared image simulation data into the attention module to obtain the predicted weight values corresponding to the simulated values of the initial entrance pupil radiation parameters of multiple bands respectively may include the following operations.

[0121] Inputting the simulated values of the initial entrance pupil radiation parameters of multiple bands respectively into the pooling layer to obtain the spectral context descriptors corresponding to multiple bands respectively; inputting the spectral context descriptors corresponding to multiple bands respectively into the shared network layer to obtain the spectral feature values corresponding to multiple bands respectively, where the shared network is composed of a multi-layer perceptron with hidden layers; obtaining the predicted weight values corresponding to the simulated values of the initial entrance pupil radiation parameters of multiple bands respectively based on the spectral feature values corresponding to multiple bands respectively.

[0122] According to an embodiment of the present disclosure, for the simulated values of the initial entrance pupil radiation parameters of multiple bands, inputting the simulated value of the initial entrance pupil radiation parameter of each band into the pooling layer, two spectral context descriptors corresponding to the simulated value of the initial entrance pupil radiation parameter can be calculated through the average pooling layer and the max pooling layer.

[0123] According to an embodiment of the present disclosure, for each spectral context descriptor of each simulated value of the initial entrance pupil radiation parameter, inputting the spectral context descriptor into the shared network layer to generate the spectral features corresponding to the two spectral context descriptors respectively. The shared network layer is composed of a multi-layer perceptron with hidden layers, and the output dimension of the shared network is the same as the dimension of the input descriptor.

[0124] According to an embodiment of the present disclosure, the spectral features output by the multi-layer perceptron corresponding to each simulated value of the initial entrance pupil radiation parameter are added by an adder to generate a band attention spectrum, that is, the prediction weight values corresponding to the simulated values of the initial entrance pupil radiation parameter in multiple bands are obtained.

[0125] According to an embodiment of the present disclosure, the simulated data of the hyperspectral thermal infrared image is input into the trained attention module, and the prediction weight values corresponding to the simulated values of the initial entrance pupil radiation parameter in multiple bands can be as shown in the following formula (9).

[0126] ; (9)

[0127] where M b The prediction weight value of the simulated value of the initial entrance pupil radiation parameter corresponding to band b, H is the simulated data of the hyperspectral thermal infrared image, is the Sigmod activation function, and are the weights in the shared network layer, and are the spectral context descriptors output by the average pooling layer and the spectral context descriptors output by the max pooling layer respectively. MLP(AvePool()) and MLP(MaxPool()) represent the average pooling layer and the max pooling layer.

[0128] According to an embodiment of the present disclosure, the process of obtaining the candidate input prediction values corresponding to the multiple initial entrance pupil radiation parameter simulation values through the weight values corresponding to the multiple initial entrance pupil radiation parameter simulation values and the weight values corresponding to the multiple initial entrance pupil radiation parameter simulation values can be as shown in the following formula (10):

[0129] ; (10)

[0130] where is the candidate input prediction value corresponding to the multiple initial entrance pupil radiation parameter simulation values, is the prediction weight value corresponding to the multiple initial entrance pupil radiation parameter simulation values, represents element-wise multiplication.

[0131] According to an embodiment of the present disclosure, the simulated values of the initial entrance pupil radiation parameter include multiple simulated values of the initial entrance pupil radiation pixels; the candidate input prediction values include multiple candidate input pixel prediction values; determining the target input prediction value from the candidate input prediction values corresponding to multiple bands can include the following operations.

[0132] For each candidate input prediction value, based on the multiple candidate input pixel prediction values included in the candidate input prediction value, determine a first importance value; normalize the first importance value to obtain a second importance value; based on the second importance values of the multiple candidate input prediction values respectively, determine a target input prediction value from the multiple candidate input prediction values, where the second importance value of the target input prediction value is greater than the second importance value of each candidate input prediction value other than the target input prediction value.

[0133] According to an embodiment of the present disclosure, since the simulated values of the initial pupil radiation parameters for each band include multiple simulated values of initial pupil radiation pixels, and each simulated value of the initial pupil radiation pixel corresponds to a different underlying surface and / or a different surface temperature respectively, the candidate input prediction values for each band also correspond to the candidate input pixel prediction values of different underlying surfaces and / or different surface temperatures respectively.

[0134] According to an embodiment of the present disclosure, after adding the candidate input pixel prediction values included in each candidate input prediction value and taking the absolute value, a first importance value can be obtained.

[0135] According to an embodiment of the present disclosure, by using an activation function to normalize the first importance value, a second importance value can be obtained. Specifically, the Softmax function can be used to map the first importance value between 0 and 1 to obtain the second importance value.

[0136] According to an embodiment of the present disclosure, based on the second importance values respectively corresponding to the multiple candidate input prediction values, sort the multiple candidate input prediction values to obtain a sorting result; based on the sorting result, select the candidate input prediction values with sorting values greater than a preset value from the multiple candidate input prediction values as the target input prediction values, that is, select the candidate input prediction values with relatively high second importance values. The preset value is not limited and can be set according to the actual situation.

[0137] According to an embodiment of the present disclosure, the process of determining the first importance value based on the multiple candidate input pixel prediction values included in the candidate input prediction value and normalizing the first importance value to obtain the second importance value can be as shown in the following formula (11).

[0138] ; (11)

[0139] Wherein, represents the th candidate input pixel prediction value in the th band, is the total number of bands in the hyperspectral thermal infrared image simulation data, is the total number of candidate input pixel prediction values included in one band, and Softmax() is the activation function.

[0140] According to an embodiment of the present disclosure, by calculating the importance values of the candidate input prediction values for each band, the quantification of the importance degree of the candidate input prediction values is realized, that is, it is more intuitive and accurate to determine whether each candidate input prediction value is valid, and by the above method of determining the target input parameters, the effect can be evaluated through the inversion accuracy, and it has better adaptability and better performance with the inversion model.

[0141] According to an embodiment of the present disclosure, the prediction module includes a residual neural network, and the residual neural network includes residual blocks, and the residual blocks include activation layers, fully connected layers and normalization layers.

[0142] According to an embodiment of the present disclosure, there can be multiple fully connected layers and normalization layers.

[0143] According to an embodiment of the present disclosure, the target input prediction value is input into the prediction module, and inside the prediction module, the target input prediction value can undergo operations such as convolution, batch normalization, ReLU activation, max pooling, residual block processing, average pooling, etc., and finally output the predicted value of the atmospheric radiation parameter of the target output band.

[0144] According to an embodiment of the present disclosure, since the process of determining the predicted value of the atmospheric radiation parameter based on the simulated data of the hyperspectral thermal infrared image is a non-linear regression problem, by replacing both the convolutional layer and the pooling layer included in the residual block of the residual neural network with fully connected layers, the prediction module can more accurately determine the value of the atmospheric radiation parameter.

[0145] According to an embodiment of the present disclosure, the above method may further include the following operations.

[0146] Normalize the initial simulated values of the in-pupil radiation parameters for each of the multiple bands respectively to obtain the normalized initial simulated values of the in-pupil radiation parameters for each of the multiple bands; based on the predicted value of the atmospheric radiation parameter and the normalized initial simulated values of the in-pupil radiation parameters for each of the multiple bands, obtain the contribution values for each of the multiple bands.

[0147] According to an embodiment of the present disclosure, Shapley Additive exPlanations analysis (SHAP) can be used to quantify the contribution of each band to the predicted value of the atmospheric radiation parameter output by the hyperspectral thermal infrared atmospheric correction model.

[0148] According to an embodiment of the present disclosure, methods such as Max-Relevance and Min-Redundancy (MRMR) and spectral separability index method can also be used to quantify the contribution of each band to the value of the atmospheric radiation parameter output by the hyperspectral thermal infrared atmospheric correction model.

[0149] According to an embodiment of the present disclosure, SHAP can obtain the contribution of each band to the model prediction by quantifying the contribution of each input variable, i.e., the simulated value of the initial entrance pupil radiation parameter, to the model prediction. Specifically, a simple linear model can be generated as a physical explanation of the hyperspectral thermal infrared atmospheric correction model, as shown in the following formula (12).

[0150] ; (12)

[0151] Wherein, is the value of the atmospheric radiation parameter output by the hyperspectral thermal infrared atmospheric correction model, is the total number of features or bands, is the contribution value of the simulated value of the initial entrance pupil radiation parameter of the i-th band, is the simulated value of the initial entrance pupil radiation parameter of the i-th band after normalization.

[0152] According to an embodiment of the present disclosure, by using SHAP to generate an explanation of the hyperspectral thermal infrared atmospheric correction model, the contribution of each input band to the atmospheric correction parameter can be visually represented by a chart, so as to summarize and generalize the rules.

[0153] According to an embodiment of the present disclosure, by quantifying the contribution value of each simulated value of the initial entrance pupil radiation parameter, the influence of the simulated value of the initial entrance pupil radiation parameter of each band on the finally output value of the atmospheric radiation parameter can be determined more accurately, so as to facilitate the physical explanation of the finally output value of the atmospheric radiation parameter.

[0154] According to an embodiment of the present disclosure, since the contribution value of each band to the prediction of the hyperspectral thermal infrared atmospheric correction model obtained through the above steps can realize the physical explanation of the hyperspectral thermal infrared atmospheric correction model, the above method can solve the problems of over-reliance on training data, lack of physical explanation in the training process and results in the related art. And the above hyperspectral thermal infrared atmospheric correction model can also be called a hyperspectral thermal infrared atmospheric correction model that integrates physical models and deep learning.

[0155] Figure 3 Schematically shows a schematic diagram of a hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure.

[0156] As Figure 3As shown, the simulated hyperspectral thermal infrared image data 310 is input into the attention module 320 to obtain predicted weight values corresponding to the simulated values of the initial entrance pupil radiation parameters for multiple bands. The multiple initial simulated values of the entrance pupil radiation parameters and the predicted weight values corresponding to the simulated values of the initial entrance pupil radiation parameters for multiple bands are multiplied one by one respectively to obtain multiple candidate input predicted values 330. After screening the multiple candidate input predicted values 330, the target input predicted value is obtained or no screening is performed. The multiple candidate input predicted values 330 or the target input predicted value are input into the prediction module 340 to obtain the predicted value 350 of the atmospheric radiation parameter of the target output band.

[0157] According to an embodiment of the present disclosure, to ensure the stability and noise resistance of the hyperspectral thermal infrared atmospheric correction model, the trained hyperspectral thermal infrared atmospheric correction model can be tested, and sensitivity analysis can be performed on the model based on the test set.

[0158] According to an embodiment of the present disclosure, specifically, the sensitivity analysis process is as follows: 20% of the remaining data set is used as the test set. The process of adding perturbations to the main atmospheric parameters such as the above-mentioned atmospheric temperature, water vapor, carbon dioxide (CO2) content, or ozone (O3) content, as well as the surface temperature and surface emissivity to the test set obtains multiple new simulated values of the third entrance pupil radiation parameters and multiple new simulated values of the fourth entrance pupil radiation parameters. The multiple new simulated values of the third entrance pupil radiation parameters and the multiple new simulated values of the fourth entrance pupil radiation parameters are input into the trained hyperspectral thermal infrared atmospheric correction model to obtain new predicted values of the atmospheric radiation parameters. The difference between the new predicted values of the atmospheric radiation parameters and the predicted values of the atmospheric radiation parameters without perturbations is calculated, so as to obtain the sensitivity of the hyperspectral thermal infrared atmospheric correction model to the earth-atmosphere parameters and noise, and further evaluate the stability and noise resistance of the constructed hyperspectral thermal infrared atmospheric correction model.

[0159] According to an embodiment of the present disclosure, the prediction accuracy of the hyperspectral thermal infrared atmospheric correction model can be evaluated by applying the hyperspectral thermal infrared atmospheric correction model, and further the effectiveness of the hyperspectral thermal infrared atmospheric correction model can be determined.

[0160] According to an embodiment of the present disclosure, specifically, the process of application accuracy evaluation can be as follows: First, input the hyperspectral thermal infrared image data into the trained hyperspectral thermal infrared atmospheric correction model to output the predicted value of the first atmospheric radiation parameter for the target output band. Specifically, globally, select typical target areas covering underlying surfaces such as soil, vegetation, and water bodies, and high, medium, and low latitudes, and collect cloudless clear-sky transit images of the target hyperspectral thermal infrared payload sensors at different observation zenith angles in different seasons at each site. Extract the mean value of the uniform area where the site is located, and based on the extracted mean value, input the hyperspectral thermal infrared image data into the trained hyperspectral thermal infrared atmospheric correction model to output the predicted value of the first atmospheric radiation parameter for the target output band.

[0161] According to an embodiment of the present disclosure, secondly, combine the high-precision atmospheric profile to evaluate the prediction accuracy of the hyperspectral thermal infrared atmospheric correction model and verify the effectiveness of the constructed hyperspectral thermal infrared atmospheric correction model. Specifically, collect high-precision atmospheric profile data that is spatio-temporally matched with the above cloudless clear-sky transit images, such as the radiosonde data of the University of Wyoming. To ensure that the atmospheric profile data represents the true atmospheric conditions during the imaging of the target payload, the time difference between the atmospheric profile data and the imaging time of the target payload is limited within 1 hour. Screen out the cloudless clear-sky profiles, and perform height and spatial interpolation on the atmospheric profiles. Input the atmospheric profile data, sensor characteristics such as the spectral response function, and the actual observation geometry into the 4A / OP hyperspectral thermal infrared radiation transfer model to obtain the simulated values of the first atmospheric radiation parameter for multiple bands.

[0162] Finally, use the predicted value of the first atmospheric radiation parameter for the target output band and the simulated value of the first atmospheric radiation parameter corresponding to the target output band to evaluate the accuracy of the inversion result of the atmospheric radiation parameter model. Specifically, methods such as the mean absolute error (MAE), standard deviation (STD), root mean square error (RMSE), and correlation coefficient (R2) can be used to evaluate the accuracy of the hyperspectral thermal infrared atmospheric correction model. The accuracy evaluation formulas are as shown in (13) to (16) below:

[0163] ; (13)

[0164] ; (14)

[0165] ; (15)

[0166] ; (16)

[0167] Among them, is the i-th band, is the number of channels used for inversion, i.e., the total number of bands, is the predicted value of the first atmospheric radiation parameter, is the simulated value of the first atmospheric radiation parameter, is the average value of the simulated values of the first atmospheric radiation parameter within the target output band, MAE is the mean absolute error, STD is the standard deviation, RMSE is the root mean square error, and R2 is the correlation coefficient.

[0168] According to the embodiments of the present disclosure, combined with the sensitivity analysis, model interpretation, and effectiveness analysis of the trained hyperspectral thermal infrared atmospheric correction model integrating physical models and deep learning, a comprehensive performance evaluation can be carried out on the constructed trained atmospheric radiation parameters.

[0169] According to the embodiments of the present disclosure, the comprehensive evaluation is mainly divided into two aspects: First, effectiveness analysis. Verify the reliability of the model inversion through the test set and the values of the atmospheric radiation parameters output by the trained hyperspectral thermal infrared atmospheric correction model. Verify the stability and noise resistance of the model through sensitivity analysis; verify the effectiveness of band selection by comparing the predicted values of the second atmospheric radiation parameters output by the hyperspectral thermal infrared atmospheric correction model trained by band selection methods such as not performing band selection, i.e., not determining the target input prediction value from multiple candidate input prediction values, with the predicted values of the atmospheric radiation parameters output after band selection; verify the effectiveness of the atmospheric correction method considering non-atmospheric window bands by comparing the predicted values of the third atmospheric radiation parameters obtained by directly inputting the atmospheric window bands into the hyperspectral thermal infrared atmospheric correction model and the predicted values of the atmospheric radiation parameters obtained by inputting all bands into the hyperspectral thermal infrared atmospheric correction model. Second, physical meaning analysis. Combining the SHAP model interpretation, analyze the characteristics and laws of the channel combinations suitable for calculating atmospheric radiation parameters under different underlying surfaces, different atmospheric conditions, and different observation zenith angles.

[0170] According to the embodiments of the present disclosure, the hyperspectral thermal infrared atmospheric correction method may include the following operations.

[0171] Input the hyperspectral thermal infrared image data into the trained hyperspectral thermal infrared atmospheric correction model to obtain the predicted values of the atmospheric radiation parameters in the target output band.

[0172] According to the embodiments of the present disclosure, the hyperspectral thermal infrared image data may be data obtained by a hyperspectral thermal infrared payload sensor.

[0173] According to the embodiments of the present disclosure, the trained hyperspectral thermal infrared atmospheric correction model includes a trained attention module and a trained prediction module, and the connection method between the two is not limited. For example, the attention module is embedded in the front end of the prediction module.

[0174] According to an embodiment of the present disclosure, hyperspectral thermal infrared image data is input into a trained attention module to obtain predicted weight values corresponding to the initial pupil radiation parameter values of multiple bands, wherein the hyperspectral thermal infrared image data includes the initial pupil radiation parameter values of multiple bands respectively; based on the initial pupil radiation parameter values of multiple bands respectively and the predicted weight values corresponding to the initial pupil radiation parameter values of multiple bands respectively, candidate input predicted values corresponding to multiple bands respectively are obtained; a target input predicted value is determined from the candidate input predicted values corresponding to multiple bands respectively; the target input predicted value is input into a trained prediction module to obtain an atmospheric radiation parameter predicted value of a target output band, wherein the target output band is the band with the highest effective information in a target band range.

[0175] According to an embodiment of the present disclosure, inputting the target input predicted value into the prediction module can obtain the atmospheric radiation parameter value corresponding to the band with the highest effective information in the target band range, thereby realizing the output of the atmospheric radiation parameter value after atmospheric correction while outputting the effective band.

[0176] Figure 4 The figure schematically shows a schematic diagram of a training method of a hyperspectral thermal infrared atmospheric correction model according to another embodiment of the present disclosure.

[0177] As Figure 4 shown, the method includes screening clear-sky cloud-free atmospheric profile data 402 from an atmospheric profile database 401, inputting the clear-sky cloud-free atmospheric profile data 402, sensor characteristics 403 such as: a preset spectral response function, a preset payload observation geometry 404 into a thermal infrared radiation transfer model 405 to obtain multiple simulated atmospheric radiation parameter values, such as: an upwelling atmospheric radiation simulated value 406, a downwelling atmospheric radiation simulated value 407, an atmospheric transmittance simulated value 408, inputting the upwelling atmospheric radiation simulated value 406, the downwelling atmospheric radiation simulated value 407, the atmospheric transmittance simulated value 408, the surface emissivity 409 and the surface temperature 410 in a spectral library into a thermal infrared radiation transfer equation 411 to obtain hyperspectral thermal infrared image simulated data 412, performing a sensitivity analysis on the simulated hyperspectral thermal infrared image simulated data 412 to obtain a channel division criterion 412, based on the channel division criterion 413, dividing multiple bands in the hyperspectral thermal infrared image simulated data 412 into an atmospheric parameter sensitive band 414 and a surface parameter sensitive band 415, and determining a target output band 417 from the surface parameter sensitive band 414 based on the SEASP band selection method 416

[0178] According to an embodiment of the present disclosure, weights are respectively assigned to the atmospheric parameter sensitive band 414 and the surface parameter sensitive band 415 to obtain the weighted band 418. The weighted first pupil radiation parameter simulation value and the weighted second pupil radiation parameter simulation value corresponding to the weighted band 418 are input into the attention module 419 to obtain the predicted weight values corresponding to the initial pupil radiation parameter simulation values of multiple bands respectively. Based on the initial pupil radiation parameter simulation values of multiple bands and the predicted weight values corresponding to the initial pupil radiation parameter simulation values of multiple bands respectively, candidate input prediction values corresponding to multiple bands are obtained. The target input prediction value is determined from the candidate input prediction values corresponding to multiple bands. The target input prediction value is input into the prediction module 420 to obtain the predicted values of the atmospheric radiation parameters of the target output band, namely: the predicted value of the upward atmospheric radiation 421, the predicted value of the downward atmospheric radiation 422, and the predicted value of the atmospheric transmittance 423. The predicted value of the upward atmospheric radiation 421, the predicted value of the downward atmospheric radiation 422, and the predicted value of the atmospheric transmittance 423 and the simulated values of the atmospheric radiation parameters of the target output band, namely the simulated value of the upward atmospheric radiation, the simulated value of the downward atmospheric radiation, and the simulated value of the atmospheric transmittance, are input into the loss function to obtain the loss value 424. Based on the loss value 424, the model parameters are adjusted until the loss value 424 is less than the preset loss value, that is, the model converges, and a trained hyperspectral thermal infrared atmospheric correction model 425 is obtained. Sensitivity analysis 426, SHAP analysis 427, and application 428 of the trained atmospheric radiation parameters based on the above hyperspectral thermal infrared atmospheric correction method are respectively performed on the trained hyperspectral thermal infrared atmospheric correction model 425 to obtain the application results. Application accuracy evaluation 428 is performed based on the application results 428. A comprehensive evaluation 430 of the model performance is obtained based on the sensitivity analysis 426, SHAP analysis 427, and evaluation results 429.

[0179] Based on the above training method of the hyperspectral thermal infrared atmospheric correction model, the present disclosure also provides a training device for the hyperspectral thermal infrared atmospheric correction model. The following will be combined with Figure 5 to describe this device in detail.

[0180] Figure 5 Schematically shows a structural block diagram of a training device for a hyperspectral thermal infrared atmospheric correction model according to an embodiment of the present disclosure.

[0181] As Figure 5 shown, the training device for the hyperspectral thermal infrared atmospheric correction model includes a weight determination module 510, a first determination module 520, a second determination module 530, an atmospheric parameter determination module 540, and an output module 550.

[0182] The weight determination module 510 is configured to input the simulated data of the hyperspectral thermal infrared image into the attention module to obtain predicted weight values corresponding to the simulated values of the initial pupil radiation parameters of multiple bands respectively, wherein the simulated data of the hyperspectral thermal infrared image includes the simulated values of the initial pupil radiation parameters of multiple bands respectively;

[0183] The first determination module 520 is configured to obtain candidate input prediction values corresponding to multiple bands respectively based on the simulated values of the initial pupil radiation parameters of multiple bands respectively and the predicted weight values corresponding to the simulated values of the initial pupil radiation parameters of multiple bands respectively;

[0184] The second determination module 530 is configured to determine a target input prediction value from the candidate input prediction values corresponding to multiple bands respectively;

[0185] The atmospheric parameter determination module 540 is configured to input the target input prediction value into the prediction module to obtain a predicted value of the atmospheric radiation parameter of the target output band, wherein the target output band is the band with the highest effective information in the target band range;

[0186] The training module 550 is configured to adjust the parameters of the attention module and the prediction module respectively based on the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band to obtain a trained hyperspectral thermal infrared atmospheric correction model.

[0187] According to an embodiment of the present disclosure, any plurality of the weight determination module 510, the first determination module 520, the second determination module 530, the atmospheric parameter determination module 540, and the output module 550 may be combined and implemented in one module, or any one of them may be split into multiple modules. Or, at least part of the functions of one or more of these modules may be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the weight determination module 510, the first determination module 520, the second determination module 530, the atmospheric parameter determination module 540, and the output module 550 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on substrate, a system on package, an application specific integrated circuit (ASIC), or may be implemented by any other reasonable way of integrating or packaging circuits, etc., in hardware or firmware, or implemented in any one of the three implementation manners of software, hardware, and firmware, or in an appropriate combination of any several of them. Or, at least one of the weight determination module 510, the first determination module 520, the second determination module 530, the atmospheric parameter determination module 540, and the output module 550 may be at least partially implemented as a computer program module, and when the computer program module is run, it may execute the corresponding functions.

[0188] Based on the above hyperspectral thermal infrared atmospheric correction method, the present disclosure also provides a training device for a hyperspectral thermal infrared atmospheric correction model. According to an embodiment of the present disclosure, the hyperspectral thermal infrared atmospheric correction device includes: a correction module. The correction module is configured to input hyperspectral thermal infrared image data and hyperspectral thermal infrared image simulation data into a trained hyperspectral thermal infrared atmospheric correction model to obtain a predicted value of the atmospheric radiation parameter for the target output band.

[0189] Figure 6 Schematically shows a block diagram of an electronic device suitable for implementing the training method of the hyperspectral thermal infrared atmospheric correction model and the hyperspectral thermal infrared atmospheric correction method according to an embodiment of the present disclosure.

[0190] As Figure 6 shown, the electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage section 606 into a random access memory (RAM) 603. The processor 601 can include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor, and / or a related chipset, and / or a dedicated microprocessor (such as an application-specific integrated circuit (ASIC)), etc. The processor 601 can also include on-board memory for caching purposes. The processor 601 can include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0191] In the RAM 603, various programs and data required for the operation of the electronic device 600 are stored. The processor 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. The processor 601 performs various operations of the method flow according to an embodiment of the present disclosure by executing the programs in the ROM 602 and / or the RAM 603. It should be noted that the program can also be stored in one or more memories other than the ROM 602 and the RAM 603. The processor 601 can also perform various operations of the method flow according to an embodiment of the present disclosure by executing the programs stored in one or more memories.

[0192] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, and the input / output (I / O) interface 605 is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the I / O interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 608 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage portion 607 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. The drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read therefrom can be installed into the storage portion 607 as needed.

[0193] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist separately without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the methods according to the embodiments of the present disclosure are implemented.

[0194] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the above-described ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603.

[0195] Embodiments of the present disclosure also include a computer program product, which includes a computer program, and the computer program includes program codes for executing the methods shown in the flowcharts. When the computer program product runs in a computer system, the program codes are used to cause the computer system to implement the training method and the hyperspectral thermal infrared atmospheric correction method provided by the embodiments of the present disclosure.

[0196] When the computer program is executed by the processor 601, the above functions defined in the system / apparatus of the embodiments of the present disclosure are executed. According to the embodiments of the present disclosure, the above-described systems, apparatuses, modules, units, etc. can be implemented by computer program modules.

[0197] In one embodiment, the computer program can rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program can also be transmitted and distributed in the form of signals on a network medium, and be downloaded and installed through the communication part 609, and / or be installed from the removable medium 611. The program code included in the computer program can be transmitted by any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0198] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or be installed from the removable medium 611. When the computer program is executed by the processor 601, the above functions defined in the system of the embodiments of the present disclosure are executed. According to the embodiments of the present disclosure, the above-described systems, devices, apparatuses, modules, units, etc. can be implemented by computer program modules.

[0199] According to the embodiments of the present disclosure, the program code for executing the computer program provided by the embodiments of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include but are not limited to, such as Java, C++, Python, the "C" language, or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0200] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code, and the above-mentioned module, segment of a program, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, as well as combinations of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0201] Those skilled in the art can understand that the features described in various embodiments and / or claims of the present disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in various embodiments and / or claims of the present disclosure can be combined and combined in various ways. All such combinations and / or combinations fall within the scope of the present disclosure.

[0202] The embodiments of the present disclosure have been described above. However, these embodiments are only for illustrative purposes and not for limiting the scope of the present disclosure. Although the embodiments have been described separately above, this does not mean that the measures in each embodiment cannot be used advantageously in combination. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and all such substitutions and modifications should fall within the scope of the present disclosure.

Claims

1. A training method for a hyperspectral thermal infrared atmospheric correction model, characterized in that: The method comprises: Inputting the hyperspectral thermal infrared image simulation data into the attention module to obtain prediction weight values ​​corresponding to the initial entrance pupil radiation parameter simulation values ​​of the multiple bands, wherein the hyperspectral thermal infrared image simulation data includes the initial entrance pupil radiation parameter simulation values ​​of the multiple bands; Based on the respective initial entrance pupil radiation parameter simulation values ​​of the plurality of wavebands and the prediction weight values ​​corresponding to the respective initial entrance pupil radiation parameter simulation values ​​of the plurality of wavebands, obtaining candidate input prediction values ​​corresponding to the respective plurality of wavebands; Determine a target input prediction value from the candidate input prediction values ​​corresponding to each of the plurality of bands; Inputting the target input prediction value into the prediction module to obtain the atmospheric radiation parameter prediction value of the target output band, wherein the target output band is the band containing the highest effective information in the target band range, and the prediction module includes a residual neural network, the residual neural network includes a residual block, and the residual block includes an activation layer, a fully connected layer and a normalization layer; Based on the predicted value of the atmospheric radiation parameter and the simulated value of the atmospheric radiation parameter corresponding to the target output band, respectively adjusting the parameters of the attention module and the prediction module to obtain a trained hyperspectral thermal infrared atmospheric correction model; Wherein, the method further comprises: Inputting the target atmospheric profile data, the preset load observation geometry and the preset spectral response function into the thermal infrared radiation transmission model to obtain the simulated values ​​of the atmospheric radiation parameters corresponding to each of the plurality of bands; Based on the simulated values ​​of atmospheric radiation parameters corresponding to each of the multiple bands, multiple surface temperatures, multiple surface emissivities, and a thermal infrared radiation transmission equation, obtaining the simulated values ​​of initial entrance pupil radiation parameters of each of the multiple bands; Based on the initial entrance pupil radiation parameter simulation values ​​of the multiple bands, the hyperspectral thermal infrared image simulation data is obtained; The attention module includes: a pooling layer, a shared network layer and an addition layer; the hyperspectral thermal infrared image simulation data is input into the attention module to obtain the prediction weight values ​​corresponding to the initial pupil radiation parameter simulation values ​​of multiple bands, including: Inputting the initial entrance pupil radiation parameter simulation values ​​of each of the multiple bands into the pooling layer to obtain the spectral context descriptors corresponding to each of the multiple bands; Inputting the spectral context descriptors corresponding to the plurality of bands into the shared network layer respectively to obtain spectral feature values ​​corresponding to the plurality of bands respectively, wherein the shared network is composed of a multilayer perceptron with a hidden layer; Prediction weight values ​​corresponding to the initial entrance pupil radiation parameter simulation values ​​of the multiple bands are obtained based on the spectral feature values ​​corresponding to the multiple bands.

2. The method according to claim 1, characterized in that The method further comprises: Based on a channel division standard, the initial entrance pupil radiation parameter simulation values ​​of each of the multiple bands are divided into a first entrance pupil radiation parameter simulation value corresponding to an atmospheric parameter sensitive band and a second entrance pupil radiation parameter simulation value corresponding to a surface parameter sensitive band, wherein the channel division standard includes the sensitivity types to which the multiple bands belong, and the sensitive types include an atmospheric parameter sensitive type and a surface parameter sensitive type; Multiplying the first entrance pupil radiation parameter simulation value by a first preset weight value to obtain a weighted first entrance pupil radiation parameter simulation value; Multiplying the second entrance pupil radiation parameter simulation value by a second preset weight value to obtain a weighted second entrance pupil radiation parameter simulation value, wherein the first preset weight value is greater than the second preset weight value; The weighted first entrance pupil radiation parameter simulation value and the weighted second entrance pupil radiation parameter simulation value are input into the attention module to obtain prediction weight values ​​corresponding to the initial entrance pupil radiation parameter simulation values ​​of multiple bands.

3. The method according to claim 2, characterized in that The method further comprises: Inputting the target atmospheric profile data after disturbance, the preset load observation geometry and the preset spectral response function into the thermal infrared radiation transmission model to obtain the simulated values ​​of the atmospheric radiation parameters after disturbance corresponding to each of the multiple bands; Based on the disturbed atmospheric radiation parameter simulation values ​​of the multiple bands, the surface temperature, the surface emissivity and the thermal infrared radiation transfer equation, obtaining the third entrance pupil radiation parameter simulation values ​​of the multiple bands respectively; Based on the plurality of simulated values ​​of the atmospheric radiation parameters, the plurality of surface temperatures after disturbance, the plurality of surface emissivities after disturbance, and the thermal infrared radiation transfer equation, the simulated values ​​of the fourth entrance pupil radiation parameters of the plurality of wavebands are obtained; Based on the third entrance pupil radiation parameter simulation values ​​of the multiple bands, the fourth entrance pupil radiation parameter simulation values ​​of the multiple bands, and the initial entrance pupil radiation parameter simulation values ​​of the multiple bands, obtaining interference signal-to-noise ratios corresponding to the multiple bands one by one; The interference signal-to-noise ratios corresponding to the multiple bands are compared with preset thresholds to determine the channel division standard.

4. The method according to claim 1, characterized in that: The method further comprises: Determining a correlation curve based on correlations between respective simulated values ​​of initial entrance pupil radiation parameters having adjacent wavebands; Based on the correlation curve, the plurality of bands are divided into a plurality of target band ranges; For each band in the target band range, determining the information entropy of each band based on the initial entrance pupil radiation parameter simulation value corresponding to each band; The target output band is determined based on the information entropy within each band in the target band range.

5. The method according to claim 1, characterized in that The initial entrance pupil radiation parameter simulation value includes a plurality of initial entrance pupil radiation pixel simulation values; the candidate input prediction value includes a plurality of candidate input pixel prediction values; The step of determining a target input prediction value from the candidate input prediction values ​​corresponding to each of the plurality of bands comprises: For each of the candidate input prediction values, determining a first importance value based on a plurality of candidate input pixel prediction values ​​included in the candidate input prediction value; Normalizing the first importance value to obtain a second importance value; Based on the second importance values ​​of each of the multiple candidate input prediction values, the target input prediction value is determined from the multiple candidate input prediction values, wherein the second importance value of the target input prediction value is greater than the second importance value of each of the candidate input prediction values ​​except the target input prediction value.

6. The method according to claim 1, characterized in that The method further comprises: Normalizing the initial entrance pupil radiation parameter simulation values ​​of the multiple wavebands respectively to obtain normalized initial entrance pupil radiation parameter simulation values ​​of the multiple wavebands; Based on the predicted values ​​of the atmospheric radiation parameters and the normalized initial entrance pupil radiation parameter simulation values ​​of each of the multiple bands, contribution values ​​of each of the multiple bands are obtained.

7. A hyperspectral thermal infrared atmosphere correction method, characterized in that: The method comprises: Inputting the hyperspectral thermal infrared image data into the trained hyperspectral thermal infrared atmospheric correction model to obtain the predicted value of the atmospheric radiation parameter of the target output band; The hyperspectral thermal infrared atmosphere correction model is obtained by training using the hyperspectral thermal infrared atmosphere correction model training method described in any one of claims 1 to 6.

Images