CN-122016056-A - Calibration reference source data simulation method, device, electronic equipment and program product

CN122016056ACN 122016056 ACN122016056 ACN 122016056ACN-122016056-A

Abstract

The embodiment of the disclosure discloses a calibration reference source data simulation method, a device, electronic equipment and a program product, wherein the method comprises the steps of preprocessing first calibration related data of an infrared instrument to be calibrated and calibration reference source data of a reference infrared instrument to generate a model training data set; the infrared instrument to be calibrated and the reference infrared instrument are respectively loaded on different satellites; the model training data set comprises a plurality of data matching pairs, wherein the data matching pairs comprise sample input features obtained by the first calibration related data and corresponding label features obtained by calibration reference source data, a mixed attention multi-layer perceptron model is built based on the model training data set, an end-to-end mapping relation between the first calibration related data and the calibration reference source data is learned in a data driving mode, and second calibration related data of an infrared instrument to be calibrated is inverted into observation data of the reference infrared instrument by utilizing the built mixed attention multi-layer perceptron model.

Inventors

QIN PING
AN HONGDA
QU JIANHUA
LV YUANYANG
YUAN MINGGE

Assignees

北京华云星地通科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (10)

1. A method of scaling reference source data simulation, comprising: the method comprises the steps of preprocessing first calibration related data of an infrared instrument to be calibrated and calibration reference source data of a reference infrared instrument to generate a model training data set, wherein the infrared instrument to be calibrated and the reference infrared instrument are infrared instruments respectively loaded on different satellites; Constructing a mixed attention multi-layer perceptron model based on the model training data set, and learning an end-to-end mapping relation between the first scaling related data and the scaling reference source data in a data driving mode; And inverting the second calibration related data of the infrared instrument to be calibrated into the observation data of the reference infrared instrument by using the constructed mixed attention multi-layer perceptron model.
2. The method of claim 1, wherein the mixed-attention multi-layer perceptron model mixes a channel attention mechanism, a self-attention mechanism, and a cross-attention mechanism.
3. The method of claim 1, wherein the mixed-attention multi-layer perceptron model includes a feature expansion module, a balancing layer, and a feature extraction module, the feature expansion module and feature extraction module being of symmetrical construction; The characteristic expansion module is used for carrying out linear transformation and characteristic expansion on the sample input characteristics, and improving characteristic dimension layer by layer to generate multi-scale characteristics; The balance layer is used for mapping the last layer output of the feature expansion module to a potential feature vector with higher dimensionality; And the feature extraction module takes the potential space vector as an initial input, performs dimension reduction and reconstruction layer by combining the multi-scale features output by the feature expansion module, and generates dense features.
4. The method of claim 3, wherein the feature extension module comprises N layers, each layer comprising a channel attention sub-module, a full connection transformation sub-module, and a self attention sub-module connected in sequence; the channel attention submodule learns the correlation between the current data matching pair and each channel in the sample input characteristics through self-adaptive channel weight weighting; The fully-connected transformation submodule is used for performing basic linear transformation and feature expansion operation on the features weighted by the channel attention submodule; The self-attention submodule is used for regarding the output of the fully-connected transformation submodule as a sequence and modeling the dependency relationship between the sequences.
5. A method according to claim 3, wherein the feature extraction module comprises N layers, each layer comprising a cross-attention sub-module, a residual connection sub-module and a full connection transformation sub-module connected sequentially in sequence; The cross attention sub-module is used for selectively retrieving related information from the multi-scale features extracted by the feature expansion module, so that the bottom-layer detailed information output by the feature expansion module and the high-level comprehensive information output by the feature extraction module are focused simultaneously; the residual connection submodule is used for ensuring that the multi-scale characteristics obtained by the characteristic expansion module can be transmitted to the decoding process of the characteristic extraction module in a lossless manner; the full-connection submodule is used for performing dimension reduction and linear transformation on the output of the residual connection submodule so as to keep the characteristic dimension identical to the output of the corresponding layer of the characteristic expansion module.
6. The method of claim 1, wherein inverting the second scaled related data of the infrared instrument to be scaled into the observed data of the reference infrared instrument using the constructed hybrid attention multi-layer perceptron model comprises: preprocessing the second calibration related data to obtain multi-channel input characteristics; Inputting the input characteristics obtained after pretreatment into the constructed mixed attention multi-layer perceptron model to obtain a normalized brightness temperature value; and carrying out inverse standardization operation on the normalized bright temperature value output by the mixed attention multi-layer perceptron model, so as to restore the bright temperature value into a bright temperature value with practical physical meaning and unit.
7. A scaled reference source data simulation apparatus, comprising: The system comprises a preprocessing module, a model training data set, a calibration data acquisition module and a calibration data acquisition module, wherein the preprocessing module is used for preprocessing first calibration related data of an infrared instrument to be calibrated and calibration reference source data of a reference infrared instrument to generate the model training data set; A building module configured to build a mixed-attention multi-layer perceptron model based on the model training dataset, learning an end-to-end mapping relationship between the first scaled related data to the scaled reference source data in a data-driven manner; And the inversion module is configured to invert the second calibration related data of the infrared instrument to be calibrated into the observation data of the reference infrared instrument by using the constructed mixed attention multi-layer perceptron model.
8. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the method of any of claims 1-6.
9. A computer readable storage medium having stored thereon computer instructions, wherein the computer instructions, when executed by a processor, implement the method of any of claims 1-6.
10. A computer program product comprising computer instructions which, when executed by a processor, implement the method of any of claims 1-6.

Description

Calibration reference source data simulation method, device, electronic equipment and program product Technical Field The disclosure relates to the technical field of satellite remote sensing, in particular to a calibration reference source data simulation method, a calibration reference source data simulation device, electronic equipment and a program product. Background The on-orbit radiometric calibration of the satellite remote sensing instrument is a basic stone for ensuring the quantitative application precision of the observation data. The cross calibration is used as an important on-orbit calibration means, and the calibration precision of the reference infrared instrument is transferred to the infrared instrument to be calibrated by carrying out cooperative observation on the same earth target at the same time by the infrared instrument to be calibrated and the reference infrared instrument with high precision and high stability. GMI (GPM microwave imager) instruments of GPM (global precipitation observation planning) satellites are often used as calibration reference sources for other microwave radiometers due to their extremely high calibration accuracy and stability. However, realizing high-precision cross calibration faces many challenges including firstly that two satellites are difficult to realize strict space-time and observation geometric matching, obvious errors are introduced by small differences, secondly that the coverage of observation data of a reference infrared instrument (such as GMI) is limited, the requirement of high-frequency calibration of the infrared instrument to be calibrated (such as FY-3D/MWRI) cannot be met, and furthermore, the direct use of the observation matching data for analysis is limited by the quantity and quality of matching samples, so that a steady and universal calibration relation is difficult to establish. The prior art generally relies on building physical models or simplified empirical statistical models based on radiation transmission theory to correlate observations of both instruments. Although the mechanism of the physical model is clear, the physical model is complex in calculation, and a large number of atmospheric and surface parameters need to be accurately input, and the parameters have uncertainty. Empirical statistical models (e.g., multiple linear regression), while simple, have difficulty capturing complex nonlinear relationships between instrument observation signals and final radiated light temperatures, and the combined effects of numerous instrument internal state parameters (e.g., temperature) on the observations, resulting in reduced model accuracy and robustness as instrument state changes or long-term operational performance deteriorates. In recent years, deep learning techniques, particularly deep neural networks, have demonstrated great capabilities in nonlinear relational modeling and complex pattern recognition. A multi-layer perceptron (MLP) is used as a basic deep neural network, and has strong general function approximation capability. However, standard MLPs tend to be underrepresented when dealing with multi-feature, high-dimensional data such as satellite instrument parameters, with internal strong correlations and dependencies. Standard MLPs treat all input features equally, cannot adaptively focus on features critical to the output, and also have difficulty in efficiently modeling complex interaction sequences between features. Therefore, a solution is needed that can overcome the above drawbacks, deeply fuse the instrument physical state information, adaptively learn the complex nonlinear mapping, and thereby perform high-fidelity simulation to scale the reference source data. Disclosure of Invention Embodiments of the present disclosure provide a calibration reference source data simulation method, apparatus, electronic device, and program product. In a first aspect, embodiments of the present disclosure provide a calibration reference source data simulation method, including: the method comprises the steps of preprocessing first calibration related data of an infrared instrument to be calibrated and calibration reference source data of a reference infrared instrument to generate a model training data set, wherein the infrared instrument to be calibrated and the reference infrared instrument are infrared instruments respectively loaded on different satellites; Constructing a mixed attention multi-layer perceptron model based on the model training data set, and learning an end-to-end mapping relation between the first scaling related data and the scaling reference source data in a data driving mode; And inverting the second calibration related data of the infrared instrument to be calibrated into the observation data of the reference infrared instrument by using the constructed mixed attention multi-layer perceptron model. Wherein the mixed-attention multi-layer perceptron model mixes a channel attention mechanism, a se