Artificial intelligence correction-based cross radiation calibration method and device

Abstract

The invention discloses a cross radiation calibration method and device for artificial intelligence correction, and belongs to the technical field of electronic information, and the method comprises the steps: S1, obtaining a sample pair through a space-time matching method, and carrying out the fitting of sample data, and obtaining a fitting curve; s2, a series of time points are selected, and temperature measurement values of the light path component of the infrared remote sensing instrument at the time points are obtained; inputting the output values of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain corresponding cross calibration result values; s3, taking the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of a neural network, taking the cross calibration result value as the output of the neural network, and training the neural network; and S4, obtaining a calibration result of the remote sensing instrument at any moment through network output. According to the invention, any short-time single-time radiometric calibration result can be provided.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic information, and in particular to a method and device for cross-radiation calibration corrected by artificial intelligence. Background Art

[0002] Infrared imagers are based on the principle of infrared radiation and can observe by receiving infrared light from objects. In one application scenario, an infrared imager can be carried on a satellite to observe the earth. The infrared imager receives infrared light radiated by the object and then outputs an output value to perform infrared imaging. In the process of implementing the above technical solution, the output value of the infrared imager is determined according to the energy of the received infrared light. However, in the related technology, calibration resources (including calibration devices, calibration operation time, etc.) are required to complete the calibration process. The existing effective radiation calibration of space infrared remote sensing instruments is generally based on the Global Space-based Cross-calibration System (GSICS).

[0003] The Global Space-based Cross-calibration System (GSICS) is an international program initiated by the World Meteorological Organization (WMO) and the Coordination Group for Meteorological Satellites (CGMS). Its purpose is to cross-calibrate satellite remote sensing instruments in international business operations and unify them to a common reference standard to ensure that observations made by different remote sensing instruments on different satellites at different times and locations are comparable. The core of cross-calibration is to obtain the correction coefficients of the two remote sensing instruments in terms of radiation through the correlation between the observation data of remote sensing instruments at small time intervals and similar spatial locations. The correction coefficients reflect the difference in radiation response when the two instruments observe the same target, and the observation values ​​of the satellite to be calibrated can be corrected by the correction coefficients.

[0004] From the development of GSICS, it can be seen that the purpose of cross-calibration is to correct the radiation benchmarks of different instruments. The working principle is to eliminate random errors through large samples and find out the systematic deviation of the radiation benchmarks of the two instruments. Due to sample limitations, it is difficult to provide short-term, single-time calibration coefficients.

[0005] Cross-calibration is mainly divided into three steps: 1) Sample matching and screening, that is, matching and screening the data of the cross-observation area of ​​the two satellites to make them as consistent as possible in time, space and observation geometry. This is an important step in the cross-calibration process and directly affects the calculation efficiency and calibration accuracy; 2) Data conversion, that is, converting the matching sample data of the two instruments into a directly comparable form, mainly including unit conversion, spectral convolution or spectral compensation processing; 3) Calibration calculation, calibrating the matching sample data after data conversion and filtering according to the empirical fitting relationship. Conventionally, fitting the samples using polynomials is the recommended method. Moreover, due to the design linearity requirements of infrared remote sensing instruments, general fitting usually uses first-order fitting and second-order fitting, that is, the fitting formula is

[0006]

[0007] Wherein, N in the formula is generally 1 or 2. For the cross calibration coefficient {a i}, we can use a large sample to j ,y j The sample pair can be obtained by solving the equation shown in the above formula. It can be a pair of sample points of brightness temperature (physical unit is Kelvin or other equivalent physical units), a pair of sample points of radiance (physical unit is W / m2·μm·steradian or other equivalent physical units), or a data count value (Digital Number) of the instrument to be calibrated and the radiance or brightness temperature of the reference instrument.

[0008] From a mathematical point of view, the current cross-calibration data fitting method is a mathematical fitting based on data statistics, which deviates from the physical basis of infrared remote sensing instruments. Due to statistical principles and sample limitations, GSICS is difficult to provide short-term, single-time calibration coefficients.

[0009] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0010] The object of the present invention is to provide a cross-radiation calibration method and device with artificial intelligence correction to solve some or all of the above problems.

[0011] To achieve the above object, the present invention provides an artificial intelligence corrected cross radiation calibration method, comprising the following steps:

[0012] S1: Obtain sample pairs through the time-space matching method, fit the sample data, and obtain the fitting curve;

[0013] S2: Select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values;

[0014] S3: using the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of the neural network, using the cross-calibration result value as the output of the neural network, and training the neural network;

[0015] S4: Input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through the network output.

[0016] In one embodiment of the present invention, the sample pairs are brightness temperature sample point pairs, radiance sample point pairs, data count values ​​of the instrument to be calibrated, and the radiance or brightness temperature of the reference instrument.

[0017] In one embodiment of the present invention, step S3 specifically includes the following steps:

[0018] S301: The output value x of the remote sensing instrument to be calibrated at a certain time point j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j as a label;

[0019] S302: Normalize all inputs and labels;

[0020] S303: Create a fully connected type including an input layer, at least one hidden layer and an output layer as a neural network model to be trained;

[0021] S304: Train the neural network model using the normalized input and labels.

[0022] In one embodiment of the present invention, step S304 specifically includes the following steps: during the training process, using forward propagation to calculate the predicted value y output by the neural network model pred , according to the predicted value y pred and the true label y to calculate the loss value of the current model; according to the loss value in each round of training, calculate and record the gradient, then clear the gradient and execute the back propagation algorithm according to the loss value to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold and stop the training process.

[0023] In one embodiment of the present invention, step S304 specifically includes the following steps: the loss function adopts In the form of pred represents the predicted value of the neural network output, and y represents the cross-calibration result value obtained in step S2.

[0024] The present invention also provides an artificial intelligence corrected cross radiation calibration device, comprising: a fitting curve acquisition module, a data processing module, a training module and a data acquisition module;

[0025] The fitting curve acquisition module is used to acquire sample pairs through a time-space matching method, fit the sample data, and obtain a fitting curve;

[0026] The data processing module is used to select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values;

[0027] The training module is used to use the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of the neural network, and the cross calibration result value as the output of the neural network to train the neural network;

[0028] The data acquisition module is used to input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through network output.

[0029] In one embodiment of the present invention, the sample pairs are brightness temperature sample point pairs, radiance sample point pairs, data count values ​​of the instrument to be calibrated, and the radiance or brightness temperature of the reference instrument.

[0030] In one embodiment of the present invention, the training module includes: an input submodule, a processing submodule, a model creation submodule and a training submodule.

[0031] The input submodule is used to convert the output value x of the remote sensing instrument to be calibrated at a certain time point into j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j as a label;

[0032] The processing submodule is used to normalize all inputs and labels;

[0033] The model creation submodule is used to create a fully connected type neural network model to be trained, which includes an input layer, at least one hidden layer and an output layer;

[0034] The training submodule is used to train the neural network model using the normalized input and labels.

[0035] In one embodiment of the present invention, the training submodule is specifically used to: during the training process, use forward propagation to calculate the predicted value y output by the neural network model pred , according to the predicted value y predand the true label y to calculate the loss value of the current model; according to the loss value in each round of training, calculate and record the gradient, then clear the gradient and execute the back propagation algorithm according to the loss value to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold and stop the training process.

[0036] In one embodiment of the present invention, the loss function is In the form of pred represents the predicted value of the neural network output, and y represents the cross-calibration result value obtained in step S2.

[0037] Compared with the prior art, the cross-radiation calibration method and device corrected by artificial intelligence according to the present invention automatically learns the complex relationship between the output value of the remote sensing instrument to be calibrated and the optical path temperature measurement value and the calibration result by utilizing the powerful learning ability of the neural network. Normalization processing can improve the stability and convergence speed of training, and the fully connected neural network can capture various features and patterns in the input data, thereby improving the accuracy and reliability of the calibration result. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flow chart of a cross-radiation calibration method corrected by artificial intelligence according to an embodiment of the present invention;

[0039] Figure 2 It is a structural schematic diagram of a cross-radiation calibration device corrected by artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation modes.

[0041] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0042] like Figure 1 As shown, an artificial intelligence corrected cross-radiation calibration method according to a preferred embodiment of the present invention is mainly used in a calibration device of an infrared imager, that is, a device that executes the calibration method of an infrared imager.

[0043] The present invention calibrates the infrared imager based on the infrared radiation principle. In terms of the basic framework, the method of obtaining sample pairs of two infrared remote sensing instruments recommended by the standard GSICS method is still adopted, that is, a large number of sample pairs are obtained by the time-space matching method, and the above fitting formula is fitted to obtain a fitting curve. The conventional GSICS method completes the operation by obtaining the fitting curve, and then the instrument to be calibrated is corrected by the fitting curve.

[0044] Specifically, the present invention comprises the following steps:

[0045] S1: Obtain sample pairs through the time-space matching method, fit the sample data, and obtain the fitting curve.

[0046] Specifically, the formula for the fitting curve is:

[0047]

[0048] Where N is 1 or 2, representing a first-order fit or a second-order fit; i} is the cross calibration coefficient; x is the output value of the remote sensing instrument to be calibrated, and y is the cross calibration result value.

[0049] The sample pairs can be brightness temperature (physical unit is Kelvin or other equivalent physical units) sample point pairs, radiance (physical unit is watt / square meter·micrometer·steradian or other equivalent physical units) sample point pairs, or the data count value (Digital Number) of the instrument to be calibrated and the radiance or brightness temperature of the reference instrument.

[0050] S2: Select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values.

[0051] Specifically, after obtaining the fitting curve in step S1, a series of time points are selected, here [t1, t2, …, t K ], where K represents the number of time points that can meet the method requirements, where meeting the method requirements means that K samples are sufficient to train a GSICS-AICNN network with satisfactory accuracy. The temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at these time points are represented by [T1, T2, …, T M ], where M represents the number of temperature values ​​of the optical path components that can be measured. For the number of samples, cross-radiation calibration can generally be performed. Depending on the satellite working mode, the sample volume will be sufficient.

[0052] After selecting the above time points, the output values ​​{xj |j=1,2,…,K} is input into the fitting curve of step S1, and the corresponding cross calibration result value {y j |j=1,2,…,K}.

[0053] S3: The output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument are used as the input of the neural network, and the cross-calibration result value is used as the output of the neural network to train the neural network.

[0054] In step S3, the output value x of the remote sensing instrument to be calibrated at each time point is j and the temperature measurement values ​​of the infrared remote sensing instrument optical path components at that time [T1, T2, …, T M ] together form an M+1 dimensional vector as the input of the neural network, and then the output value x of the remote sensing instrument to be calibrated at that moment is j The corresponding cross calibration result value y j As the output of the neural network; at all K time points that meet the requirements of the method, (M+1)×K dimensional input and K dimensional output are formed, and then the neural network is trained, and the present invention names it as an artificial intelligence corrected cross-radiation calibration network (GSICS based AICorrected Neural Network, GSICS-AICNN).

[0055] The neural network may be selected from (but not limited to) a fully connected neural network, and the optimizer (Adam) may be used (but not limited to).

[0056] When training a neural network, the following steps are performed:

[0057] S301: The output value x of the remote sensing instrument to be calibrated at a certain time point j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j As a label.

[0058] S302: Normalize all inputs and labels:

[0059]

[0060] Among them, d i represents the i-th input, d i-max and d i-min is the maximum and minimum value of the input vector, d i ′ is the value after normalization. Label normalization uses the same method as above.

[0061] S303: Create a fully connected neural network including an input layer, at least one hidden layer and an output layer as a model to be trained.

[0062] The fully connected neural network in the present invention obtains the output (calibration result) by subjecting the input data (the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path) to a series of linear transformations and activation functions. In the training stage, the parameters of the model are adjusted using the known input and the corresponding labels (normalized results) so that the model can predict the calibration result of the new input data as accurately as possible.

[0063] In the present invention, the activation function used is the relu function. Other activation functions can also be used if they meet the accuracy requirements after testing.

[0064] The artificial intelligence corrected cross-radiation calibration network (GSICS-AICNN) adopts the following code (using Python code as an example, but not limited to Python code):

[0065]

[0066]

[0067] S304: Train the model using the normalized input and labels.

[0068] During the training process, the GSICS-AICNN forward propagation is used to calculate the predicted value y of the neural network output pred , according to the predicted value y pred And the true label y calculates the loss value y of the current model loss ; According to the loss value y in each round of training loss , calculate and record the gradient, then clear the gradient and calculate the loss value y loss Execute the back propagation algorithm to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold value, and stop the training process. In the present invention, the threshold value is set to 0.1, which is used to adjust the required radiation calibration accuracy. The smaller the value, the higher the required radiation calibration accuracy. When the required radiation calibration accuracy is reduced, the value can be increased.

[0069] During the training process of the model, an optimizer is also provided to update the parameters of the model according to the calculated gradient, so that the model is adjusted in the direction of reducing the loss. The parameters of the model are the network and training parameters in the training process.

[0070] Among them, the loss function adopts In the form of pred represents the predicted value of the neural network output, y represents the label, that is, the cross-calibration result value obtained in step S2. This formula only shows the form of the loss function.

[0071] During the training process, through multiple cycles, the model parameters are continuously adjusted to minimize the loss function, so that the model can better fit the given training data.

[0072] The logical process of training is in the form of the following code (using Python code as an example, but not limited to Python code):

[0073]

[0074] By repeating forward propagation, calculating loss, back propagation and parameter updating, the model parameters are gradually adjusted so that the model can better fit the training data. During the training process, the training effect of the model can be monitored by observing the change of loss value, using the validation set for evaluation, etc., and the hyperparameters (such as learning rate, number of layers, number of neurons, etc.) can be adjusted as needed to improve the performance of the model.

[0075] S4: Input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through the network output.

[0076] The calibration result is:

[0077] y=GSICS-AICNN.out(T1,T2,…,T M ,x).

[0078] The present invention is based on the high-precision fitting ability of neural networks to functions, and takes the physical factors that affect the calibration accuracy (i.e., temperature) as part of the input, ensuring that the method can obtain high-quality calibration data within the entire radiation dynamic range of the remote sensing image under any working condition (i.e., the state represented by the temperature of the optical path components of the remote sensing instrument); and can provide any short-time, single-shot radiation calibration results.

[0079] like Figure 2 As shown, an artificial intelligence corrected cross radiation calibration device according to a preferred embodiment of the present invention includes: a fitting curve acquisition module 1, a data processing module 2, a training module 3 and a data acquisition module 4.

[0080] The fitting curve acquisition module 1 is used to acquire sample pairs through a time-space matching method, fit the sample data, and obtain a fitting curve.

[0081] Specifically, the formula for the fitting curve is:

[0082]

[0083] Where N is 1 or 2, representing a first-order fit or a second-order fit; i} is the cross calibration coefficient; x is the output value of the remote sensing instrument to be calibrated, and y is the cross calibration result value.

[0084] The sample pairs can be brightness temperature (physical unit is Kelvin or other equivalent physical units) sample point pairs, radiance (physical unit is watt / square meter·micrometer·steradian or other equivalent physical units) sample point pairs, or the data count value (Digital Number) of the instrument to be calibrated and the radiance or brightness temperature of the reference instrument.

[0085] The data processing module 2 is used to select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values.

[0086] Specifically, after obtaining the fitting curve, a series of time points are selected, here [t1, t2, …, t K ], where K represents the number of time points that can meet the method requirements, where meeting the method requirements means that K samples are sufficient to train a GSICS-AICNN network with satisfactory accuracy. The temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at these time points are represented by [T1, T2, …, T M ], where M represents the number of temperature values ​​of the optical path components that can be measured. For the number of samples, cross-radiation calibration can generally be performed. Depending on the satellite working mode, the sample volume will be sufficient.

[0087] After selecting the above time points, the output values ​​{x j |j=1,2,…,K} is input into the fitting curve of step S1, and the corresponding cross calibration result value {y j |j=1,2,…,K}.

[0088] The training module 3 is used to train the neural network by taking the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of the neural network and taking the cross calibration result value as the output of the neural network.

[0089] In training module 3, the output value x of the remote sensing instrument to be calibrated at each time point is j and the temperature measurement values ​​of the infrared remote sensing instrument optical path components at that time [T1, T2, …, T M] together form an M+1 dimensional vector as the input of the neural network, and then the output value x of the remote sensing instrument to be calibrated at that moment is j The corresponding cross calibration result value y j As the output of the neural network; at all K time points that meet the requirements of the method, (M+1)×K dimensional input and K dimensional output are formed, and then the neural network is trained, and the present invention names it as the artificial intelligence corrected cross-radiation calibration network (GSICS based AI Corrected Neural Network, GSICS-AICNN).

[0090] The neural network may be selected from (but not limited to) a fully connected neural network, and the optimizer (Adam) may be used (but not limited to).

[0091] When training the neural network, the training module 3 specifically includes:

[0092] The input submodule 301 is used to convert the output value x of the remote sensing instrument to be calibrated at a certain time point into j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j as a label;

[0093] The processing submodule 302 is used to normalize all inputs and labels:

[0094]

[0095] Among them, d i represents the i-th input, d i-max and d i-min is the maximum and minimum value of the input vector, d i ′ is the value after normalization. Label normalization uses the same method as above.

[0096] The model creation submodule 303 is used to create a fully connected neural network including an input layer, at least one hidden layer and an output layer as a model to be trained.

[0097] The fully connected neural network in the present invention obtains the output (calibration result) by subjecting the input data (the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path) to a series of linear transformations and activation functions. In the training phase, the parameters of the model are adjusted using the known input and the corresponding labels (normalized calibration results) so that the model can predict the calibration results of new input data as accurately as possible.

[0098] In the present invention, the activation function used is the relu function. Other activation functions can also be used if they meet the accuracy requirements after testing.

[0099] The artificial intelligence corrected cross-radiation calibration network (GSICS-AICNN) adopts the following code (using Python code as an example, but not limited to Python code):

[0100]

[0101] The training submodule 304 is used to train the model using the normalized input and labels.

[0102] During the training process, the training submodule 304 is used to calculate the predicted value y output by the neural network using the GSICS-AICNN forward propagation. pred , according to the predicted value y pred And the true label y calculates the loss value y of the current model loss ; According to the loss value y in each round of training loss , calculate and record the gradient, then clear the gradient and calculate the loss value y loss Execute the back propagation algorithm to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold value, and stop the training process. In the present invention, the threshold value is set to 0.1, which is used to adjust the required radiation calibration accuracy. The smaller the value, the higher the required radiation calibration accuracy. When the required radiation calibration accuracy is reduced, the value can be increased.

[0103] During the training process of the model, an optimizer is also provided to update the parameters of the model according to the calculated gradient, so that the model is adjusted in the direction of reducing the loss. The parameters of the model are the network and training parameters in the training process.

[0104] Among them, the loss function adopts In the form of pred represents the predicted value of the neural network output, y represents the label, that is, the cross-calibration result value obtained in step S2. This formula only shows the form of the loss function.

[0105] During the training process, through multiple cycles, the model parameters are continuously adjusted to minimize the loss function, so that the model can better fit the given training data.

[0106] The logical process of training is in the form of the following code (using Python code as an example, but not limited to Python code):

[0107]

[0108]

[0109] By repeating forward propagation, calculating loss, back propagation and parameter updating, the model parameters are gradually adjusted so that the model can better fit the training data. During the training process, the training effect of the model can be monitored by observing the change of loss value, using the validation set for evaluation, etc., and the hyperparameters (such as learning rate, number of layers, number of neurons, etc.) can be adjusted as needed to improve the performance of the model.

[0110] The data acquisition module 4 is used to input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through the network output.

[0111] The calibration result is:

[0112] y=GSICS-AICNN.out(T1,T2,…,T M ,x).

[0113] The foregoing description of specific exemplary embodiments of the present invention is for the purpose of illustration and demonstration. These descriptions are not intended to limit the present invention to the precise form disclosed, and it is clear that many changes and variations can be made based on the above teachings. The purpose of selecting and describing the exemplary embodiments is to explain the specific principles of the present invention and its practical application, so that those skilled in the art can realize and utilize various different exemplary embodiments of the present invention and various different selections and changes. The scope of the present invention is intended to be limited by the claims and their equivalents.

Claims

1. An artificial intelligence corrected cross radiation calibration method, characterized in that: The following steps are involved: S1: Obtain sample pairs through the time-space matching method, fit the sample data, and obtain the fitting curve; S2: Select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values; S3: using the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of the neural network, using the cross-calibration result value as the output of the neural network, and training the neural network; S4: Input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through the network output.

2. The cross-radiation calibration method corrected by artificial intelligence as claimed in claim 1, characterized in that: The sample pairs are brightness temperature sample point pairs, radiance sample point pairs, data count values ​​of the instrument to be calibrated, and the radiance or brightness temperature of the reference instrument.

3. The cross-radiation calibration method corrected by artificial intelligence as claimed in claim 1, characterized in that: Step S3 specifically includes the following steps: S301: The output value x of the remote sensing instrument to be calibrated at a certain time point j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j as a label; S302: Normalize all inputs and labels; S303: Create a fully connected type including an input layer, at least one hidden layer and an output layer as a neural network model to be trained; S304: Train the neural network model using the normalized input and labels.

4. The cross-radiation calibration method corrected by artificial intelligence as claimed in claim 3, characterized in that: Step S304 specifically includes the following steps: During the training process, use forward propagation to calculate the predicted value y output by the neural network model pred , according to the predicted value y pred and the true label y to calculate the loss value of the current model; according to the loss value in each round of training, calculate and record the gradient, then clear the gradient and execute the back propagation algorithm according to the loss value to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold and stop the training process.

5. The cross-radiation calibration method corrected by artificial intelligence as claimed in claim 4, characterized in that: Step S304 specifically includes the following steps: the loss function is adopted In the form of pred represents the predicted value of the neural network output, and y represents the cross-calibration result value obtained in step S2.

6. An artificial intelligence corrected cross radiation calibration device, characterized in that: include: Fitting curve acquisition module, data processing module, training module and data acquisition module; The fitting curve acquisition module is used to acquire sample pairs through a time-space matching method, fit the sample data, and obtain a fitting curve; The data processing module is used to select a series of time points to obtain the temperature measurement values ​​of the optical path components of the infrared remote sensing instrument at the time points; and input the output values ​​of the remote sensing instrument to be calibrated corresponding to the selected series of time points into the fitting curve to obtain the corresponding cross-calibration result values; The training module is used to use the output value of the remote sensing instrument to be calibrated and the temperature measurement value of the optical path component of the infrared remote sensing instrument as the input of the neural network, and the cross calibration result value as the output of the neural network to train the neural network; The data acquisition module is used to input the observation data value of the remote sensing instrument at a certain moment and the optical path temperature measurement value at that moment into the trained model, and obtain the calibration result of the remote sensing instrument at any moment through network output.

7. The cross-radiation calibration device corrected by artificial intelligence as claimed in claim 6, characterized in that: The sample pairs are brightness temperature sample point pairs, radiance sample point pairs, data count values ​​of the instrument to be calibrated, and the radiance or brightness temperature of the reference instrument.

8. The cross-radiation calibration device corrected by artificial intelligence as claimed in claim 6, characterized in that: The training module includes: an input submodule, a processing submodule, a model creation submodule and a training submodule The input submodule is used to convert the output value x of the remote sensing instrument to be calibrated at a certain time point into j and the optical path temperature measurement value at that time [T1, T2, …, T M ] as input, and the calibration result y obtained by fitting the curve j as a label; The processing submodule is used to normalize all inputs and labels; The model creation submodule is used to create a fully connected type neural network model to be trained, which includes an input layer, at least one hidden layer and an output layer; The training submodule is used to train the neural network model using the normalized input and labels.

9. The cross-radiation calibration device corrected by artificial intelligence as claimed in claim 8, characterized in that: The training submodule is specifically used to: during the training process, use forward propagation to calculate the predicted value y output by the neural network model pred , according to the predicted value y pred and the true label y to calculate the loss value of the current model; according to the loss value in each round of training, calculate and record the gradient, then clear the gradient and execute the back propagation algorithm according to the loss value to calculate the new gradient value of the model parameters, and adjust the parameters of the model according to the calculated new gradient value; repeat the above operation until the loss value is less than the set threshold and stop the training process.

10. The cross-radiation calibration device corrected by artificial intelligence as claimed in claim 9, characterized in that: The loss function is In the form of pred represents the predicted value of the neural network output, and y represents the cross-calibration result value obtained in step S2.

Images