Mid-infrared Band Generation Method during AGRI's Exposure to Stray Light Pollution

Abstract

The present invention discloses a method for generating mid-infrared band during the period when AGRI is contaminated by stray light, which includes: collecting the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocessing the collected brightness temperature observation data of the mid-infrared band and the far-infrared band; constructing and training a MWIR generation model for generating the brightness temperature data of the mid-infrared band from the brightness temperature observation data of the far-infrared band of AGRI according to the mapping relationship between the marked brightness temperature observation data of the mid-infrared band and the far-infrared band; inputting the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light according to the trained MWIR generation model to generate the brightness temperature data of the mid-infrared band at the same moment. The present invention can generate the brightness temperature data of the MWIR band during the period of stray light contamination more accurately, and can provide a reliable and replicable technical solution for improving the quality of geostationary satellite MWIR data.

Description

Technical Field

[0001] The present invention relates to the field of satellite remote sensing technology, and particularly to a method for generating mid-infrared bands during the period when AGRI is contaminated by stray light. Background Art

[0002] Geostationary meteorological satellites, such as FY-4A and FY-4B, have achieved three-dimensional atmospheric profile detection, high spatial resolution, high-frequency observation, and lightning spatio-temporal distribution detection in geostationary orbit. The main imaging instrument on geostationary meteorological satellites is the Advanced Geostationary Radiation Imager (AGRI), which can provide observation data at various wavelengths from visible light to infrared. The observation data of AGRI are widely used in the fields of weather and climate, such as atmospheric and surface parameter inversion, meteorological disaster monitoring, and numerical weather forecasting, etc.

[0003] Research shows that the observation deviations of FY-4A and FY-4B AGRI in the infrared band are within a reasonable range. However, recent research shows that the brightness temperature (BT) observed in the mid-infrared (MWIR) band (3.75 μm) of FY-4A AGRI sometimes shows anomalies at midnight (i.e., 16:00–18:00 UTC). These BT anomalies have an obvious seasonal dependence and usually occur 1–2 months before or after the spring or autumn equinox. Under normal circumstances, the nocturnal MWIR BT is usually lower than 300K, while when anomalies occur, the MWIR BT is often 10–20K higher than the normal value. It is analyzed that the BT anomalies in the MWIR band of AGRI are caused by stray light pollution, which is also a problem encountered by the Visible and Infrared Spin Scan Radiometer on the FY-2 series of satellites. The MWIR band measures the radiant energy from the atmosphere and the Earth, as well as the energy reflected by the sun, but the solar reflection component can be ignored at night. Therefore, the overall BT of the MWIR band at night is colder than that during the day. If the optical system is contaminated by direct or scattered solar radiation (i.e., scattered light) at night, it may cause the observed value of MWIR BT to increase abnormally.

[0004] Currently, due to the non-negligible stray light pollution, FY-4B AGRI stops observing at midnight for about 120 days around the spring and autumn equinoxes. However, the MWIR band is very valuable and necessary in various scientific applications. For example, the emissivity of fog and low-level clouds in the MWIR band is much lower than that in the far-infrared (LWIR) band, and this characteristic helps to detect these phenomena at night. The response of the MWIR band to pixel heat sources is different from that of the LWIR band, which is particularly beneficial for the early detection of forest fires. In addition, the MWIR band is also helpful for the identification of cirrus clouds, fragmented clouds, and overlapping clouds. Therefore, it is crucial to obtain reliable AGRI MWIR observations during the period of stray light pollution.

[0005] For FY-2, various methods have been adopted to correct the MWIR measurement anomalies of the FY-2 visible and infrared spin-scanning radiometer, including hardware and software solutions. However, FY-2 and FY-4 imagers differ in spectral, spatial, and temporal resolutions, as well as the patterns and scopes of scattered light contamination. The correction method for the MWIR anomaly observations of the FY-2 visible and infrared spin-scanning radiometer cannot be applied to FY-4 AGRI. It is necessary to develop a brand-new correction method for the FY-4 imager to provide reliable and reproducible technical means for improving the data quality of the MWIR band of FY-4 AGRI. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the related art to some extent.

[0007] An object of the present invention is to provide a method for generating the mid-infrared band during the period when AGRI is contaminated by stray light, constructing a MWIR generation model for removing stray light contamination during the period of stray light contamination, and generating mid-infrared band data during the period when AGRI is contaminated by stray light with the MWIR generation model, giving play to the self-learning ability and non-linear processing ability of the deep learning method, and providing reliable and reproducible technical means for improving the data quality of the MWIR band of AGRI.

[0008] Another object of the present invention is to provide a mid-infrared band generation system during the period when AGRI is contaminated by stray light.

[0009] To achieve the above object, on the one hand, the present invention provides a method for generating the mid-infrared band during the period when AGRI is contaminated by stray light, including:

[0010] Collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band;

[0011] According to the mapping relationship marked between the brightness temperature observation data of the mid-infrared band and the far-infrared band, construct and train a MWIR generation model for generating mid-infrared band brightness temperature data from the brightness temperature observation data of the far-infrared band of AGRI;

[0012] According to the trained MWIR generation model, input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light, and generate the mid-infrared band brightness temperature data at the same moment.

[0013] A further preferred technical solution of the present invention is that the preprocessing of the collected brightness temperature observation data of the mid-infrared band and the far-infrared band includes:

[0014] Interpolate and standardize the collected brightness temperature observation data in the mid-infrared band and far-infrared band to obtain samples;

[0015] Divide the samples at one moment into multiple small samples with equal regional sizes, and process each moment in turn to obtain a dataset for training the MWIR generation model;

[0016] Divide the dataset into a training set, a validation set, and a test set.

[0017] Preferably, the interpolation and standardization processing of the collected brightness temperature observation data in the mid-infrared band and far-infrared band includes:

[0018] Define a region and the longitude and latitude grids within the region. The parameters of the grids include longitude, latitude, and spatial resolution;

[0019] Using the nearest neighbor interpolation method, interpolate the full-disk observations of the mid-infrared band and far-infrared band collected by AGRI onto the defined longitude and latitude grids, and the boundary value filling strategy uses the extrapolation method;

[0020] Respectively count the maximum and minimum values of the brightness temperature observation data of the mid-infrared band and far-infrared band of AGRI, and perform maximum-minimum normalization processing on all the brightness temperature observation data.

[0021] Preferably, for the maximum-minimum normalization processing of all the brightness temperature observation data, the calculation formula used for maximum-minimum normalization is:

[0022]

[0023] Where X is the original brightness temperature observation data, and X scaled is the transformed data; X min is the minimum value in the brightness temperature observation data, and X max is the maximum value in the brightness temperature observation data.

[0024] Preferably, the MWIR generation model uses the U-Net deep learning framework;

[0025] The U-Net deep learning framework includes four encoder modules and four decoder modules that make up the U-shaped network structure. Each encoder module and decoder module consists of two depthwise separable convolutional layers, batch normalization processing, and ReLU activation functions;

[0026] After each encoder module, connect a max pooling layer, and pass the output to the attention module, and then connect to the next encoder module;

[0027] Each decoder module first performs upsampling, then combines with the output of the corresponding encoder module through skip connection, and then performs convolution operations;

[0028] The output of the last decoder module undergoes a 1×1 convolution to generate the final MWIR result.

[0029] Preferably, a training set is used to train the MWIR generation model, and the mean absolute error loss is used as the training objective to adjust the network weights of the model.

[0030] Preferably, during the training of the MWIR generation model, a validation set is used to finely tune the hyperparameters of the MWIR generation model;

[0031] After the training of the MWIR generation model is completed, MAE, RMSE, and correlation coefficient scores are used as regression evaluation indicators to evaluate the model.

[0032] On the other hand, the present invention provides a mid-infrared band generation system during AGRI's stray light contamination, including:

[0033] A data acquisition module, configured to collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band;

[0034] A model construction module, configured to construct and train a MWIR generation model for generating mid-infrared band brightness temperature data from the far-infrared band brightness temperature observation data of AGRI according to the mapping relationship between the labeled mid-infrared band and far-infrared band brightness temperature observation data;

[0035] A MWIR generation module, configured to input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light according to the trained MWIR generation model, and generate the brightness temperature data of the mid-infrared band at the same moment.

[0036] On another aspect, the present invention further provides a readable storage medium, on which computer instructions are stored, and the computer instructions cause the computer to execute the above-mentioned mid-infrared band generation method during AGRI's stray light contamination.

[0037] On yet another aspect, the present invention further provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned mid-infrared band generation method during AGRI's stray light contamination.

[0038] In another aspect, the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned method for generating the mid-infrared band during the period when AGRI is contaminated by stray light.

[0039] Beneficial effects: Aiming at the problem that the MWIR of the existing geostationary satellite AGRI is contaminated by stray light during midnight, the present invention collects the LWIR and MWIR brightness temperature observation data observed during the period without stray light contamination, interpolates and normalizes the full-disk observation data, and constructs a set of models for MWIR generation based on the deep learning network framework;

[0040] Using this model, the brightness temperature data of the MWIR band during the period of stray light contamination can be generated more accurately, which can provide a reliable and replicable technical solution for improving the quality of geostationary satellite MWIR data. Description of the Drawings

[0041] Figure 1 It is a flowchart of the method for generating the mid-infrared band during the period when AGRI is contaminated by stray light in the embodiment of the present invention.

[0042] Figure 2 It is a comparison result graph of the MWIR brightness temperature generated during the normal observation stage by the generation method of the present invention and the MWIR brightness temperature observed by AGRI.

[0043] Figure 3 It is a comparison result graph of the MWIR brightness temperature generated during the period of stray light contamination by the generation method of the present invention and the MWIR brightness temperature observed at normal times.

[0044] Figure 4 It is a comparison result graph of the MWIR brightness temperature generated during the period of stray light contamination by the generation method of the present invention and the observed MWIR brightness temperature for a specific case. Detailed Embodiments

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention, and they should not be construed as limitations on the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be construed as indicating or implying relative importance.

[0046] The problem that the MWIR anomaly observation correction method based on the previous FY-2 visible and infrared spin-scanning radiometer cannot be directly applied to FY-4A GRI. With the development of artificial intelligence (AI), numerous studies have shown that deep learning methods can automatically learn appropriate features from datasets and capture the statistical relationship between input variables and targets. The application of deep learning methods in the meteorological field has achieved success. In particular, the U-Net model has received attention due to its simplicity, efficiency, ease of understanding, and customizability. The U-Net model can fuse features at different scales and effectively improve the model accuracy. Therefore, the U-Net structure or U-Net-based models are widely used in satellite remote sensing tasks, such as cloud physical property inversion and satellite image generation. Therefore, the U-Net algorithm is expected to be used to obtain the measurement values of the AGRI MWIR band.

[0047] Therefore, the purpose of the present invention is to develop a MWIR generation model that removes stray light pollution during stray light pollution, giving play to the autonomous learning ability and non-linear processing ability of deep learning methods, and providing a reliable and replicable technical solution and beneficial attempt for improving the data quality of the MWIR band of FY-4A GRI.

[0048] The following combines Figures 1 - 4 Specifically describe the mid-infrared band generation method and system provided by the present invention during the period when AGRI is affected by stray light pollution.

[0049] Example 1: This example provides a mid-infrared band generation method during the period when AGRI is affected by stray light pollution, as Figure 1 shown, including:

[0050] S100. Collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not affected by stray light pollution, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band.

[0051] In this example, the brightness temperature observation data of the mid-infrared band and the far-infrared band collected are respectively: the brightness temperature observation data of the MWIR band and the LWIR band of FY-4A AGRI, GK-2A AMI, and Himawari-8A HSI.

[0052] Himawari-8, GK-2A, and FY-4A are the same and all belong to geostationary meteorological satellites. The observation data of Himawari-8A HSI and GK-2A AMI are also similar to the observation data of FY-4A AGRI. Based on the needs of testing and verification, in this example, the brightness temperature observation data of the MWIR band and the LWIR band of GK-2A AMI and Himawari-8A HSI are collected simultaneously as a supplement to the observation data of FY-4A AGRI.

[0053] Among them, the MWIR and LWIR band data of FY-4A AGRI are the data of Channel 8 (central wavelength of 3.75 μm) and Channels 9–16 (central wavelengths of 6.25 μm, 7.10 μm, 8.55 μm, 10.80 μm, 12.00 μm, and 13.5 μm respectively) observed by FY-4A AGRI. During the stray light pollution period (i.e., from early February to mid-May, and from late August to mid-November), the brightness temperature observation data at 15:00 UTC and 19:00 UTC are selected. During the period without stray light pollution, the brightness temperature observation data at 16:00 UTC, 17:00 UTC, and 18:00 UTC are selected. The central wavelengths of the brightness temperatures of the MWIR and LWIR bands of GK-2A AMI are 3.83 μm and 12.36 μm respectively, and the central wavelengths of the brightness temperatures of the MWIR and LWIR bands of Himawari-8A HI are 3.85 μm and 12.35 μm respectively.

[0054] Preprocess the collected brightness temperature observation data in the mid-infrared band and far-infrared band. The specific method is as follows:

[0055] S110. Interpolate and standardize the collected brightness temperature observation data in the mid-infrared band and far-infrared band to obtain samples, including:

[0056] S111. Define a region and the longitude and latitude grids within this region. The parameters of the grids include longitude, latitude, and spatial resolution. In this embodiment, an equi-longitude and equi-latitude grid with a spatial resolution of 0.04 degrees (about 4 km) is generated within the range of longitude 44.00°E–166.88°E and latitude 62.88°S–60.00°N.

[0057] S112. Use the nearest neighbor interpolation method to interpolate the full-disk observation values of the mid-infrared band and far-infrared band of AGRI and AMI collected onto the defined longitude and latitude grids, and adopt the extrapolation method for the boundary value filling strategy. Himawari-8A HI is in the NC format of grid data and can be directly read and plotted, so no processing is performed.

[0058] S113. Respectively count the maximum and minimum values of the brightness temperature observation data in the mid-infrared band and far-infrared band of AGRI, and perform maximum-minimum normalization processing on all the brightness temperature observation data. The calculation formula for maximum-minimum normalization is:

[0059]

[0060] Among them, X is the original brightness temperature observation data, X scaled is the transformed data; X min is the minimum value in the brightness temperature observation data, X maxis the maximum value in the brightness temperature observation data. The maximum and minimum value normalization method helps to convert the data into a unified scale, which is crucial for deep learning algorithms.

[0061] S120. For the samples obtained in step S110, perform spatiotemporal matching on the MWIR and LWIR observation data at the same time. In order to save computing resources, the processed MWIR and LWIR regional data at each time are cropped into 36 small areas of 6×6, each of which includes 512×512 grid points, and each small area is a sample data. For each sample data, determine whether it contains missing or invalid values. If so, remove the sample. Process the MWIR and LWIR observation data at each time in turn to form a data set.

[0062] S130. In order to establish a MWIR generation model during the AGRI stray light pollution period based on the U-Net deep learning algorithm, this embodiment uses the FY-4A observation LWIR multi-band observation data as the input factor for model training. In order to ensure that the model can achieve the best performance at each stage, this embodiment splits the data set into a training set, a validation set, and a test set in chronological order. Specifically, the following strategy is used to split the data set:

[0063] (1) Training set: data used to optimize model weights. This example uses MWIR and LWIR observation data from January 2019 to August 2020 as the training set. The U-Net model is trained with these data to capture the spatial variation characteristics of MWIR during normal periods.

[0064] (2) Validation set: used to evaluate the performance of the model during training and adjust hyperparameters to avoid overfitting. This example uses MWIR and LWIR observation data from September and October 2020 as the validation set. During model training, the performance on the validation set is monitored and adjusted as needed to ensure the generalization ability of the model on new data.

[0065] (3) Test set 1: data used to evaluate model performance. This example uses MWIR and LWIR observation data from November and December 2020 as test set 1. After completing model training and validation, the model is evaluated on the test set to verify its effectiveness in generating MWIR in practice.

[0066] (4) Test set 2: Data used for the final evaluation of the model performance. In this embodiment, the MWIR and LWIR observation data from January to December 2021 are used as test set 2. After completing model training, validation, and testing, it is applied on test set 2. The trained model is used to generate MWIR data during the stray light pollution period, and the generated data is compared with the MWIR data in the normal period at the adjacent moment. At the same time, the MWIR and LWIR data of Himawari-8 AHI and GK2A AMI are used as references to verify its final effect in practical applications.

[0067] S200. According to the mapping relationship between the labeled brightness temperature observation data in the mid-infrared band and the far-infrared band, construct and train a MWIR generation model that generates mid-infrared band brightness temperature data from the far-infrared band brightness temperature observation data of AGRI.

[0068] In this embodiment, the MWIR generation model adopts the U-Net deep learning framework. The input feature channels of the model are 6, and the feature variables are the data of 6 bands in channels 9 - 16 (with central wavelengths of 6.25μm, 7.10μm, 8.55μm, 10.80μm, 12.00μm, and 13.5μm respectively) observed by FY-4A AGRI. The output feature channel is 1, and the feature variable is the data of the band in channel 8 (with a central wavelength of 3.75μm) observed by FY-4A AGRI. To ensure the accuracy of the training data, in this embodiment, the MWIR and LWIR observation data are interpolated and normalized in step S100, and the MWIR not contaminated by stray light is used as the true value for training. The mean absolute error loss is used as the training objective. Through the training and evaluation of the model, it is found that the U-Net algorithm performs excellently in generating MWIR band data, has a good correspondence with the MWIR observed brightness temperature of Himawari-8 AHI and GK2A AMI, and has high accuracy and reliability.

[0069] The U-Net network structure adopted in this embodiment is similar to SmaAt-UNet and is an improved convolutional neural network model. The model introduces an attention module and depthwise separable convolution to improve the efficiency of the model and reduce the number of parameters. The U-Net deep learning framework includes four encoder modules and four decoder modules that form a U-shaped network structure. Each encoder module and decoder module consists of two depthwise separable convolutional layers, batch normalization processing, and ReLU activation functions;

[0070] After each encoder module, a max-pooling layer is connected, and the output is passed to the attention module, and then connected to the next encoder module; the encoder part is responsible for extracting the features of the LWIR brightness temperature observed by FY-4A AGRI input, gradually reducing the spatial resolution, and increasing the depth of the feature map at the same time. The specific operations are as follows:

[0071] At the input layer, the input of the model is the LWIR brightness temperature data of 6 channels in the 9-14 bands with a size of 512×512, and then it enters two consecutive depthwise separable convolutional layers. The convolutional operation is used to extract the spatial features of the LWIR brightness temperature data, and batch normalization processing and ReLU activation function are applied after each convolution. The depthwise separable convolutional layer here can be decomposed into two steps: depthwise convolution (performing convolution independently on each input channel) and pointwise convolution (combining the output of the depthwise convolution with a 1×1 convolutional kernel), which significantly reduces the number of parameters and the amount of computation. After the convolutional operation, a 2×2 max-pooling layer is used to reduce the size of the feature map, halving the width and height of the image while doubling the number of feature channels.

[0072] The convolutional block attention module is introduced after each encoder module in the U-Net model. These modules include two sub-modules: a) Channel attention module, which emphasizes the feature channels that are more important for generating the MWIR brightness temperature by weighting the channel dimension of the input feature map. b) Spatial attention module, which highlights the more important regions in the brightness temperature image by weighting the spatial dimension of the input feature map. The output of the attention module is used together with the input of the next layer to enhance the model's attention to key features.

[0073] Each decoder module first performs upsampling, then combines with the output of the corresponding encoder module through a skip connection, and then performs a convolutional operation; the decoder part is used to gradually restore the spatial resolution of the feature image while reducing the depth of the feature map. The specific operations are as follows:

[0074] The input of the decoder module comes not only from the upsampled feature map but also from the feature map of the corresponding encoder module (transmitted through the skip connection). The skip connection ensures that high-resolution information is not lost during the decoding process, enabling the model to better restore the details of the input. The upsampling layer in the decoder uses bilinear interpolation to double the size of the feature map. Compared with transposed convolution, bilinear interpolation has a lower computational cost and does not produce artifacts. Similar to the encoder part, two consecutive depthwise separable convolutional layers are also used in the decoder module, and batch normalization processing and ReLU activation function are applied after each convolution.

[0075] The last layer of the decoder is a 1×1 convolutional layer, which is used to map the output of the decoder to the required number of output channels. The final output is single-channel MWIR brightness temperature data with a size of 512×512.

[0076] While maintaining the prediction performance, U-Net significantly reduces the complexity of the model, making it suitable for environments with low computing resources. The structural design of U-Net reflects the efficient utilization of computing resources. By integrating the attention mechanism and depthwise separable convolution, the model significantly reduces the number of parameters and the amount of computation while maintaining the prediction accuracy. Its typical UNet architecture ensures efficient information transmission and restoration through the encoder-decoder module and skip connections, making it suitable for real-time MWIR generation scenarios.

[0077] The training of the MWIR generation model mainly includes the following aspects:

[0078] (1) Data preparation, which has been completed in step S100.

[0079] (2) Model definition: Use the U-Net algorithm, set the input feature channels to 6, the output feature channels to 1, and the input and output data sizes to 512×512;

[0080] (3) Training settings: Before training the model, some training hyperparameters need to be set. Among them, the batch size is 16, the max epoch is 100, and the learning rate is 0.0001.

[0081] (4) Model training: Use 1 Nvidia RTX A6000 GPU to train the model. During the training process, use the mean absolute error loss as the training objective and continuously adjust the parameters to make the model obtain the best performance on the training data.

[0082] (5) Model evaluation and verification: After the training is completed, the model needs to be evaluated to assess its performance on the training data. First, use the data in the test set for evaluation. In this embodiment, based on the model results and the observed data, use MAE, RMSE, and the correlation coefficient (R) to conduct overall tests respectively to evaluate the performance of the model in generating MWIR band data. If the performance of the model is not ideal, the performance of the model can be improved by adjusting the model parameters or changing the training data. Second, use the data during the stray light pollution period in 2021 to apply and verify the model, and compare it with the MWIR band brightness temperature at the nearest normal observation time to evaluate the effect of the model in actual applications.

[0083] S300. According to the trained MWIR generation model, input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light, and generate the brightness temperature data of the mid-infrared band at the same time.

[0084] This step process is equivalent to the work of evaluating the trained MWIR generation model. In this embodiment, the evaluation of the trained MWIR generation model is mainly carried out on test set 1 (the period of normal observation without stray light contamination) and test set 2 (the period of stray light contamination in 2021). First, we evaluated the performance of the U-Net model in generating the MWIR band using test set 1. Figure 2 Figure for the comparison of the MWIR brightness temperature generated by the U-Net model and the FY-4A AGRI conventional observation results. Obviously, the brightness temperature generated by the U-Net model is highly consistent with the FY-4A AGRI observation values, as Figure 2 shown in (a). The brightness temperature generated by the U-Net model is densely distributed near the 1:1 line, indicating that the generation error is very small (MAE = 1.47K, RMSE = 2.39K, R = 0.99). In addition, as Figure 2 shown in the probability density function graph in (b), the frequency distribution of the brightness temperature generated by the U-Net model is almost exactly the same as the frequency distribution of the FY-4A AGRI MWIR observation brightness temperature values. This consistency is particularly obvious in the brightness temperature range of 260 - 290K, and the brightness temperature generated by the U-Net model is almost exactly the same as the actual observation values. This shows that during the normal observation period, the brightness temperature generated by the U-Net model can well represent the MWIR band data. Based on this model, a large number of relatively accurate MWIR band data can be generated for the FY-4 satellite during the stray light contamination period.

[0085] To further test the actual application effect of the MWIR generation model, this embodiment applies the model to the stray light contamination period (test set 2). As Figure 3 shown, it is the comparison of the MWIR (16:00–18:00 UTC) generated by the U-Net model with the MWIR in the adjacent normal periods (15:00 UTC and 19:00 UTC) using the data during the periods of stray light contamination (16:00 - 18:00 UTC) in March, April, September, and October 2021. The results show that during the period of 16:00–18:00 UTC, the brightness temperature observed by FY-4A is overestimated to varying degrees and cannot accurately represent the lower brightness temperature. Since the overestimation degree is different for each month, we analyzed each month separately. In March, as Figure 3As shown in (a), the observed brightness temperatures at 16:00 UTC and 18:00 UTC are overestimated to varying degrees, and the brightness temperatures exceed 300 K in both cases. In the range of 260–280 K, the observed brightness temperature distribution is wider than that at 15:00 UTC and 19:00 UTC. In particular, the overestimation of the brightness temperature at 18:00 UTC is the most significant, and the lowest observed brightness temperature (about 230 K) is significantly higher than the lowest brightness temperatures at 15:00 UTC and 19:00 UTC (about 200 K). Compared with the observed values, the brightness temperatures generated by U-Net at 16:00 and 18:00 UTC are closer to the observed values at 15:00 UTC and 19:00 UTC, effectively reducing the brightness temperature below 300 K. The observed brightness temperature at 17:00 UTC shows reasonable consistency with the brightness temperatures at 15:00 UTC and 19:00 UTC, without obvious overestimation. In April, as Figure 3 shown in (b), the lowest observed brightness temperature from 16:00 to 18:00 UTC exceeds 220 K, and the highest brightness temperature exceeds 300 K. It is worth noting that the overestimation of the observed brightness temperature at 17:00 UTC is the most serious, and the brightness temperature values exceeding 300 K are widely distributed. However, compared with the observed values, the brightness temperatures generated by U-Net from 16:00 to 18:00 UTC have a good correspondence with the brightness temperatures at 15:00 UTC and 19:00 UTC. Similar to March, as Figure 3 shown in (c), in September, only the observed brightness temperatures at 16:00 UTC and 18:00 UTC show overestimation, and the brightness temperatures generated by U-Net are basically consistent with the brightness temperature distribution during the normal observation period. In October, as Figure 3 shown in (d), the AGRI observed brightness temperatures at 16:00 UTC and 17:00 UTC are overestimated to varying degrees, and the overestimated temperature range between 260 and 300 K at 16:00 UTC is too large. At 17:00 UTC, the overestimation of the AGRI observed brightness temperature is more serious than that at 16:00 UTC. The U-Net generation results at 16:00 and 17:00 UTC have good consistency with the brightness temperatures during the normal observation period.

[0086] To more intuitively evaluate the effectiveness of the U-Net model, as Figure 4As shown, this embodiment gives an example during the stray light pollution for specific illustration. Since there is a lack of real observations during the stray light pollution, therefore, the observation data of the MWIR bands of GK-2A AMI (3.83 μm) and Himawari-8 AHI (3.85 μm) are used as references to qualitatively evaluate the performance of the U-Net model. To minimize the difference in observation time between satellites, the starting observation times of the data of GK-2A AMI and Himawari-8 AHI are consistent with those of the FY-4A AGRI data. It should be noted that the MWIR band is less affected by carbon dioxide in the lower troposphere, which means that when the sensor zenith angle is large, the radiance will show obvious transmission attenuation. The sub-satellite longitudes of the FY-4A, GK-2A, and Himawari-8 satellites are 104.7°E, 128.2°E, and 140.7°E respectively. Therefore, there are differences in the observation geometric parameters (such as the sensor zenith angle) of the three satellites. In addition, the central wavelengths and spectral response functions of the MWIR bands of the three satellites are also different. Therefore, the MWIR brightness temperatures of GK-2A AMI or Himawari-8 AHI cannot be regarded as the true values of the MWIR brightness temperatures generated by the U-Net model. In addition, this embodiment also provides the brightness temperature differences (BTDs, the former minus the latter) between the MWIR and LWIR bands of GK-2A AMI / Himawari-8 AHI for evaluation.

[0087] Figure 4 Example results at 17:00 UTC on October 31, 2021 are shown. From Figure 4 it can be seen that the observed brightness temperatures of FY-4A in most regions exceed 250 K, especially in the region from 20°S to 20°N as shown in Figure 4 (a). For the analysis of the observations of GK-2A AMI in Figure 4 (c) and Himawari-8 AHI in Figure 4 (d), it shows that between 20°S and 20°N, the brightness temperatures generally do not exceed 300 K, and there are regions where the brightness temperatures are lower than 210 K. On the contrary, in the ocean regions with low brightness temperatures, the observed values of FY-4A AGRI MWIR are abnormally high, up to about 40 K. In the land regions, the observed values also show overestimations to varying degrees. In contrast, the brightness temperatures generated by the U-Net model as shown in Figure 4 (b) are in most regions consistent with those of GK-2A AMI shown in Figure 4 (c) and Figure 4The observations of Himawari-8 AHI shown in (d) are very close. The U-Net model effectively eliminates the high-brightness temperature in ocean and land areas. The brightness temperature generated by the U-Net model is closer to the MWIR observations of GK-2A AMI. Similarly, it is difficult for the U-Net model to generate MWIR brightness temperatures exceeding 300K. Therefore, the brightness temperatures generated by the U-Net model are inconsistent with the observations of GK-2A AMI and Himawari-8 AHI in some areas. As Figure 4 The BTDs results shown in also show certain differences, such as Figure 4 The brightness temperature difference of FY-4A in (e) is on the high side. However, as Figure 4 As shown in (f), the results generated by the U-Net model more accurately reflect the BTDs, and are closer to Figure 4 the observations of GK-2A AMI in (g) and Figure 4 the observations of Himawari-8 AHI in (h), thus more accurately presenting the performance of high clouds. Generally speaking, during the stray light pollution period (midnight period of the spring and autumn equinoxes, 16:00-18:00 UTC), the U-Net model has a good effect on removing stray light and can further improve the MWIR brightness temperature data of FY-4A AGRI.

[0088] It can be seen that the MWIR generation model based on U-Net deep learning can accurately generate the brightness temperature in the MWIR band of FY-4A AGRI during the stray light pollution period. This is also the first time to solve the problem of stray light pollution in the MWIR band of FY-4A AGRI, and it can provide a reliable and replicable solution for FY-4B.

[0089] Embodiment 2: This embodiment provides a mid-infrared band generation system during the period of stray light pollution of AGRI, including:

[0090] A data acquisition module, configured to collect the brightness temperature observation data in the mid-infrared band and far-infrared band of AGRI during the period without stray light pollution, and preprocess the collected brightness temperature observation data in the mid-infrared band and far-infrared band;

[0091] A model construction module, configured to construct and train a MWIR generation model for generating the brightness temperature data in the mid-infrared band from the brightness temperature observation data in the far-infrared band of AGRI according to the mapping relationship between the labeled brightness temperature observation data in the mid-infrared band and far-infrared band;

[0092] A MWIR generation module, configured to input the brightness temperature observation data in the far-infrared band of AGRI during the period of stray light pollution according to the trained MWIR generation model, and generate the brightness temperature data in the mid-infrared band at the same moment.

[0093] Embodiment 3: This embodiment provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions cause the computer to execute a method for generating mid-infrared band data during the period when AGRI is contaminated by stray light. The method includes the following steps:

[0094] Collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band;

[0095] According to the mapping relationship between the marked brightness temperature observation data of the mid-infrared band and the far-infrared band, construct and train a MWIR generation model for generating mid-infrared band brightness temperature data from the brightness temperature observation data of the far-infrared band of AGRI;

[0096] According to the trained MWIR generation model, input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light to generate the mid-infrared band brightness temperature data at the same time.

[0097] Embodiment 4: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus. The processor can call the logical instructions in the memory to execute the method for generating mid-infrared band data during the period when AGRI is contaminated by stray light. The method includes the following steps:

[0098] Collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band;

[0099] According to the mapping relationship between the marked brightness temperature observation data of the mid-infrared band and the far-infrared band, construct and train a MWIR generation model for generating mid-infrared band brightness temperature data from the brightness temperature observation data of the far-infrared band of AGRI;

[0100] According to the trained MWIR generation model, input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light to generate the mid-infrared band brightness temperature data at the same time.

[0101] In addition, when the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0102] Embodiment 5: The computer program product provided in this embodiment includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for generating the mid-infrared band during the period when AGRI is contaminated by stray light. The method includes the following steps:

[0103] Collect the brightness temperature observation data of the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data of the mid-infrared band and the far-infrared band.

[0104] According to the mapping relationship between the marked brightness temperature observation data of the mid-infrared band and the far-infrared band, construct and train an MWIR generation model for generating the brightness temperature data of the mid-infrared band from the brightness temperature observation data of the far-infrared band of AGRI.

[0105] According to the trained MWIR generation model, input the brightness temperature observation data of the far-infrared band during the period when AGRI is contaminated by stray light to generate the brightness temperature data of the mid-infrared band at the same moment.

[0106] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0107] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a mid-infrared band during AGRI being contaminated by stray light, characterized in that, Including: Collect the brightness temperature observation data in the mid-infrared band and far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data in the mid-infrared band and far-infrared band; According to the mapping relationship between the brightness temperature observation data in the mid-infrared band and far-infrared band of the corresponding channels at the same time, construct and train a MWIR generation model that generates mid-infrared band brightness temperature data from the far-infrared band brightness temperature observation data of AGRI; the MWIR generation model adopts a U-Net deep learning framework; The U-Net deep learning framework includes four encoder modules and four decoder modules that form a U-shaped network structure. Each encoder module and decoder module consists of two depthwise separable convolutional layers, batch normalization processing, and a ReLU activation function; A max-pooling layer is connected after each encoder module, and the output is passed to an attention module and then connected to the next encoder module; Each decoder module first performs upsampling, then combines with the output of the corresponding encoder module through a skip connection, and then performs a convolution operation; The output of the last decoder module passes through a 1×1 convolution to generate the final MWIR result; According to the trained MWIR generation model, input the brightness temperature observation data in the far-infrared band during the period when AGRI is contaminated by stray light to generate the brightness temperature data in the mid-infrared band at the same time.

2. The method for generating the mid-infrared band during the period when the AGRI is contaminated by stray light according to claim 1, wherein The preprocessing of the collected brightness temperature observation data in the mid-infrared band and far-infrared band includes: Perform interpolation and normalization processing on the collected brightness temperature observation data in the mid-infrared band and far-infrared band to obtain samples; Divide the samples at one moment into multiple small samples with equal area sizes, and process the samples at each moment in turn to obtain a dataset for training the MWIR generation model; Divide the dataset into a training set, a validation set, and a test set.

3. The method for generating the mid-infrared band during AGRI being contaminated by stray light according to claim 2, wherein, The interpolation and normalization processing of the collected brightness temperature observation data in the mid-infrared band and far-infrared band includes: Define a region and the longitude and latitude grids within the region. The parameters of the grids include longitude, latitude, and spatial resolution; Using the method of nearest neighbor interpolation, interpolate the full-disk observation values of the mid-infrared band and far-infrared band collected by AGRI onto the defined longitude and latitude grids, and the boundary value filling strategy adopts an extrapolation method; Respectively calculate the maximum and minimum values of the brightness temperature observation data in the mid-infrared band and far-infrared band of AGRI, and perform maximum-minimum normalization processing on all the brightness temperature observation data.

4. The method for generating the mid-infrared band during the AGRI being contaminated by stray light according to claim 3, wherein, The calculation formula for the maximum-minimum normalization processing of all the brightness temperature observation data is: Among them, X is the original observed brightness temperature data, X scaled is the transformed data; X min is the minimum value in the observed brightness temperature data, X max is the maximum value in the observed brightness temperature data.

5. The method for generating the mid-infrared band during the AGRI being contaminated by stray light according to claim 2, wherein Use the training set to train the MWIR generation model, adopt the mean absolute error loss as the training target, and adjust the network weights of the model.

6. The method for generating the mid-infrared band during the AGRI being contaminated by stray light according to claim 5, wherein During the training of the MWIR generation model, use the validation set to fine-tune the hyperparameters of the MWIR generation model; After the MWIR generation model is trained, use MAE, RMSE, and correlation coefficient scores as regression evaluation indicators to evaluate the model.

7. A mid-infrared band generation system during AGRI being contaminated by stray light, characterized in that, Including: A data acquisition module, configured to collect the brightness temperature observation data in the mid-infrared band and the far-infrared band during the period when AGRI is not contaminated by stray light, and preprocess the collected brightness temperature observation data in the mid-infrared band and the far-infrared band; A model construction module, configured to construct and train a MWIR generation model for generating mid-infrared band brightness temperature data from the far-infrared band brightness temperature observation data of AGRI according to the mapping relationship between the brightness temperature observation data in the mid-infrared band and the far-infrared band corresponding to the same moment; the MWIR generation model adopts a U-Net deep learning framework; The U-Net deep learning framework includes four encoder modules and four decoder modules that form a U-shaped network structure, and each encoder module and decoder module are composed of two depthwise separable convolutional layers, batch normalization processing, and ReLU activation functions; A max-pooling layer is connected after each encoder module, and the output is passed to an attention module and then connected to the next encoder module; Each decoder module first performs upsampling, then combines with the output of the corresponding encoder module through a skip connection, and then performs a convolution operation; The output of the last decoder module passes through a 1×1 convolution to generate the final MWIR result; A MWIR generation module, configured to input the brightness temperature observation data in the far-infrared band during the period when AGRI is contaminated by stray light according to the trained MWIR generation model, and generate the brightness temperature data in the mid-infrared band at the same moment.

8. A non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions cause the computer to execute the method for generating the mid-infrared band during the period when AGRI is contaminated by stray light according to any one of claims 1-6.

9. An electronic device, comprising: A processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus, and the processor calls the logic instructions in the memory to execute the method for generating the mid-infrared band during the period when AGRI is contaminated by stray light according to any one of claims 1-6.

Images