A multi-temporal cloud image prediction method and prediction model based on multi-channel satellite data
Patent Information
- Application Number
- CN202410944032.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2026-11-03
- Estimated Expiration
- 2044-07-15
AI Technical Summary
Existing cloud image prediction methods suffer from channel crosstalk in multi-channel scenarios, leading to inaccurate predictions. Furthermore, deep learning methods lack accuracy in long-term series prediction.
A multi-temporal cloud image prediction network is constructed using multi-channel satellite data. By selecting an attention module and a channel separation output module, multi-scale deep features are learned to alleviate the channel crosstalk problem. Long-term and short-term time information is extracted through iterative downsampling and upsampling modules, and the model is optimized by combining the time tilt loss function.
It effectively alleviates channel crosstalk, improves the accuracy of multi-channel cloud image prediction and the prediction capability of long-term series, and generates more realistic satellite cloud image results.
Smart Images
Figure CN118941977B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of meteorological processing technology, and in particular relates to a multi-temporal cloud image prediction method and prediction model based on multi-channel satellite data. Background Technology
[0002] Modern society has multiple ways to acquire meteorological data, including ground-based meteorological observation stations, radar, and geostationary satellites. However, the construction of ground-based meteorological observation stations is limited by terrain conditions, radar has a small coverage area and is easily interfered with by electromagnetic waves. Only geostationary satellites have the advantages of large coverage area, no spatial limitations, and high temporal resolution. Using cloud images taken by geostationary satellites to predict cloud patterns for future time periods is of great significance for studying weather changes and monitoring severe weather.
[0003] Traditional cloud image prediction methods are mostly based on block matching and optical flow. Block matching effectively utilizes the motion information of clouds, but as a linear model, it cannot accurately simulate the nonlinear motion of clouds. Optical flow calculates motion velocity through pixel changes, achieving a nonlinear simulation, but the grayscale of objects changes over time, leading to inaccurate calculation results. Modern methods, particularly deep learning, have been applied to cloud image prediction, achieving superior results compared to traditional methods. However, issues such as prediction fuzziness and short prediction times still exist, especially in multi-channel cloud image prediction where channel crosstalk frequently occurs, affecting the model's ability to generate accurate predictions. Summary of the Invention
[0004] Purpose of the invention: In order to solve the problems existing in the prior art, the present invention provides a multi-temporal cloud image prediction method and prediction model based on multi-channel satellite data.
[0005] Technical Solution: This invention discloses a multi-temporal cloud image prediction method based on multi-channel satellite data, comprising the following steps:
[0006] Step 1: Download satellite cloud images for different bands;
[0007] Step 2: Stitch together satellite cloud images from different bands to form a multi-channel satellite cloud image;
[0008] Step 3: Calculate the brightness temperature value of each pixel in each satellite cloud image, thereby mapping the satellite cloud image to the brightness temperature value form;
[0009] Step 4: Logarithmically normalize the brightness temperature values in each satellite cloud image;
[0010] Step 5: Construct a multi-channel, multi-temporal cloud image prediction network, and use the normalized multi-channel satellite cloud images from Step 4 as the training set to train the multi-channel, multi-temporal cloud image prediction network, thus obtaining the trained multi-channel, multi-temporal cloud image prediction network.
[0011] Step 6: Use the trained multi-channel, multi-temporal cloud image prediction network for prediction.
[0012] Furthermore, in step 3, the brightness temperature value of each pixel in each satellite cloud image is calculated using the following formula:
[0013]
[0014] in, This represents the brightness temperature value of the b-th pixel in the satellite cloud image. This represents the pixel value of the satellite relative to the b-th pixel in the image. and Let represent the coefficients of Planck's function, and It is the spectral response function offset and scale correction term.
[0015] Furthermore, in step 4, a logarithmic function is used for normalization, as shown in the following expression:
[0016]
[0017] in, This represents the normalized brightness temperature value of the b-th pixel in the satellite cloud image. Indicates the minimum brightness temperature value. This represents the brightness temperature value of the b-th pixel in the satellite cloud image. To scale the size.
[0018] Furthermore, the multi-channel multi-temporal cloud image prediction network includes a selection attention module, an iterative downsampling module, an iterative upsampling module, and a channel separation output module;
[0019] The attention selection module includes a first stage and a second stage. The first stage is used to extract multi-scale deep features from the original input image, and these multi-scale deep features are input into the second stage to establish the correlation between features of different scales to obtain channel enhancement features. Then, the channel enhancement features are convolved to the same size as the input features to obtain spatial channel enhancement features. The spatial channel enhancement features are input into the iterative downsampling module to gradually extract the long and short-term temporal information contained in the spatial channel enhancement features, while learning the complex cloud structure and motion patterns hidden in the image. Then, the output features of the iterative downsampling module are input into the iterative upsampling module to restore the detailed features of the image. The output features of the iterative upsampling module are input into the channel separation module. The channel separation module uses C sets of independent filters to predict single-channel satellite cloud images, where C is the number of channels of the original input image input to the multi-channel multi-temporal cloud image prediction network. Then, the predicted single-channel cloud images are stitched together in the channel dimension to obtain the predicted multi-channel satellite cloud images.
[0020] Furthermore, the first stage of the attention selection module includes n pooling layers, n upsampling modules, n element-wise multiplication modules, and a concatenation module. The i-th pooling layer is connected to the i-th upsampling module. The output of the i-th upsampling module is multiplied by the original image input to the attention selection module through the i-th element-wise multiplication module. The outputs of all element-wise multiplication modules are concatenated by the concatenation module to obtain multi-scale deep features. The second stage includes first to fourth reshaping modules, first and second matrix multiplication modules, a softmax module, an addition module, and a first convolution module. The first to third reshaping modules reshape the feature map input to the second stage into feature maps of different dimensions. The features output by the second and third reshaping modules are multiplied by the first matrix multiplication module and then input to the softmax module. The softmax module multiplies the output features of the first reshaping module by the second matrix multiplication module and then inputs them to the fourth reshaping module. The features output by the fourth reshaping module are added to the features input to the second stage to obtain channel enhancement features. The obtained channel enhancement features are input to the first convolution module to obtain spatial channel enhancement features.
[0021] Furthermore, the C-group filters in the channel separation output module have the same structure, each including a second convolution module, a first deconvolution module, a third convolution module, and a fourth convolution module connected in sequence.
[0022] Furthermore, the loss function of the multi-channel, multi-temporal cloud image prediction network... As shown below:
[0023]
[0024] in, Represents a true multi-channel satellite cloud image. express The predicted value, where i represents the i-th sample and j represents the j-th frame. This indicates the batch size, and S represents the total number of frames in the cloud map within a sample.
[0025] A prediction model for multi-temporal cloud images based on multi-channel satellite data includes: a data acquisition module, a data processing module, a model building module, and a model training module;
[0026] The data acquisition module is used to download satellite cloud images of different bands;
[0027] The data processing module is used to execute steps 2 to 4;
[0028] The model building module is used to construct a multi-channel, multi-temporal cloud image prediction network;
[0029] The model training module is used to train the constructed multi-channel, multi-temporal cloud image prediction network.
[0030] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a multi-temporal cloud image prediction method based on multi-channel satellite data.
[0031] A computer-readable storage medium storing a computer program that, when executed by a processor, performs a multi-temporal cloud image prediction method based on multi-channel satellite data.
[0032] Beneficial effects: This invention uses a channel separation output module to generate multi-channel satellite cloud image prediction results, which can effectively alleviate the crosstalk problem that occurs when different channel data are processed in the deep learning network layer. Furthermore, by using a selective attention module, it can learn the interdependence of multi-scale deep features in both spatial and channel dimensions, fully integrate the spatial features provided by different channels, and enable the model to maintain accurate prediction ability in long-term series prediction, generating more realistic satellite cloud image prediction results. Attached Figure Description
[0033] Figure 1 This is a flowchart of the present invention.
[0034] Figure 2 This is a diagram of the multi-channel, multi-temporal cloud image prediction network structure of the present invention.
[0035] Figure 3 This is a comparison chart of the predicted satellite cloud image sequence and the actual satellite cloud image sequence obtained using the method of this invention.
[0036] Figure 4 This is a schematic diagram of the structure of an attention-based multi-channel satellite cloud image prediction device according to the present invention.
[0037] Figure 5 This is a schematic diagram of the hardware structure of an attention-based multi-channel satellite cloud image prediction device according to the present invention. Detailed Implementation
[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0039] Figure 1 A flowchart of an attention-based multi-channel satellite cloud image prediction method provided in this embodiment is shown below. Figure 1 As shown, this embodiment specifically includes the following steps:
[0040] Step S1: Download satellite cloud images of different bands. In this embodiment, GOES-16 ABI L1bRadiance Data is downloaded from the NOAA website. The bands are C08 (6.17μm), C09 (6.93μm), C10 (7.34μm), and C13 (10.33μm).
[0041] Satellite cloud images contain a variety of information closely related to atmospheric precipitation, such as cloud height and thickness, cloud top temperature, and water vapor content, which can help meteorologists predict precipitation in a target area. The satellite cloud image prediction problem can be expressed as: assuming an input sequence... The sequence length is frame, The first in the cloud map sequence Frame, in which , and The satellite cloud image prediction problem can be defined as predicting satellite cloud images based on the input sequence, where width, height, and number of channels are the parameters. The next time length is obtained as follows Future frame sequence This embodiment selects GOES-16, a geostationary environmental observation satellite of the GOES series, as the observation satellite for the fixed observation area. The reflectivity signal of the satellite's advanced reference imager is used as the data source. The water vapor band (C08, C09, C13) and infrared band (C13), which are strongly correlated with precipitation, are selected for accurate prediction. This can further improve the prediction of precipitation and better cope with severe weather.
[0042] Step S2: Refer to the GOES-R satellite product manual to read the variables, parameters, and other information contained in the NetCDF format data.
[0043] The downloaded data source is stored in the NetCDF data format, which is widely used for storing data in meteorological science. NetCDF provides different interfaces for different programming languages. This embodiment uses the netCDF4 library in Python 3 and refers to the GOES-16 satellite product manual to read the metadata containing variables, parameters, and other information.
[0044] Step S3: Use the Planck function and bandpass corrected reflectivity data to obtain the corresponding band brightness temperature data.
[0045] The data downloaded from the website is GOES-16 / ABI L1b Radiance data. This data is used to obtain brightness temperature data suitable for precipitation estimation tasks. The brightness temperature value for the corresponding band can only be obtained after using the Planck function and bandpass correction. The calculation formula is as follows:
[0046]
[0047] in, This represents the brightness temperature value of the b-th pixel in the satellite cloud image. This represents the pixel value of the satellite relative to the b-th pixel in the image. and These represent the coefficients of the Planck function, which can be obtained by reading the variables "planck_fk1" and "planck_fk2" from the NC file in step 2. and These are the spectral response function shift and scale correction terms, which can be obtained by reading the variables "planck_bc1" and "planck_bc2" from the NC file in step 2. and It is the spectral response function offset and scale correction term.
[0048] Step S4: Normalize the data using a logarithmic function-based normalization method.
[0049] In the training process of deep learning, input features often have different dimensions and units. If these are input into the model without processing, it will affect the model's data analysis results. To eliminate the influence of outliers, normalization is often performed to limit the input features to a range of 0 to 1. Since this embodiment uses two different bands, water vapor and infrared, which have significant differences, it is necessary to preserve these differences during normalization. Therefore, this embodiment uses a normalization method based on a logarithmic function, and the specific calculation formula is shown below:
[0050]
[0051] in, This represents the normalized brightness temperature value of the b-th pixel in the satellite cloud image. Indicates the minimum brightness temperature value. This represents the brightness temperature value of the b-th pixel in the satellite cloud image. For scaling step S5: Construct a multi-channel, multi-temporal cloud image prediction network and train the network using the data that has undergone the aforementioned preprocessing operations.
[0052] Multi-channel, multi-temporal cloud image prediction network, such as Figure 2 As shown, it includes an attention selection module, an iterative downsampling module, an iterative upsampling module, and a channel separation output module.
[0053] Existing research using deep learning models to predict satellite cloud images generally uses single-channel satellite cloud images, which contain only data from one satellite band. However, satellite sensors can often capture meteorological data from multiple bands simultaneously. Data from different bands can both verify and complement each other. Therefore, multi-channel satellite cloud images from multiple bands are used for prediction, fusing single-channel satellite cloud images with information from other channels. Stitching other channel images with the single-channel (band) image along the channel dimension can yield more accurate prediction results. However, if multi-channel satellite cloud image data is processed simultaneously, the signal from one channel may leak or couple into other channels, causing serious crosstalk problems. This not only fails to improve prediction accuracy but may also affect the prediction of single-channel satellite cloud images. Therefore, this embodiment builds a selective attention module. In the first stage, the input multi-channel cloud images are pooled at different spatial scales to obtain a set of pooled features with different spatial resolutions. Then, upsampling is used to resample the pooled features at the same resolution. After synthesis along the channel dimension, multi-scale deep features that retain more useful information are obtained, which then proceed to the second stage. Each channel map input to the second stage can be viewed as a mapping at a specific scale. Since feature maps at different scales are correlated, to leverage this correlation, the input features of the second stage are reshaped into different shapes (RESHAPE1, RESHAPE 2, and RESHAPE 3 modules transform the dimensions of the input data to S*4C*HW, S*4C*HW, and S*HW*4C, respectively). Matrix multiplication and softmax operations are then performed between them to establish the correlation between features at different scales. Finally, after being reshaped to the input size, these features are added to the original input features of the second stage to obtain the channel-enhanced features. At this point, each channel in the enhanced features is a weighted sum of all channels, simulating long-distance dependencies between multi-scale feature maps. Finally, the channel-enhanced features are convolved to the same size as the input features to obtain the spatial channel-enhanced features.
[0054] The spatial channel enhancement features are fed into an iterative downsampling module consisting of alternating 3D convolutional layers and 3D-SwinTransformer layers to gradually extract long and short-term temporal information contained in the multi-channel cloud image sequence, while learning the complex cloud structure and motion patterns hidden in the image; then they are fed into an iterative upsampling module consisting of 3D transposed convolutional layers to restore the detailed features of the image.
[0055] If the prediction channels are directly generated through a 3D transposed convolutional layer, meaning all output channels are generated by the same filter, then each transformation will directly consider cross-channel information, disrupting the model's prediction. Therefore, a channel separation module is added, using C sets of independent filters to generate predicted single-channel contour maps. This ensures that cross-channel information is only used in the first transformation, becoming indirect information thereafter, gradually reducing side effects. Finally, the generated single-channel image is stitched together along the channel dimension. C represents the number of channels in the original input image to the multi-channel multi-temporal contour map prediction network; the filter includes one convolutional layer, one deconvolutional layer, and two convolutional layers.
[0056] In the process of predicting time series data, the model has the richest historical information in the earlier data. As the prediction time increases, the amount of information available gradually decreases, and the prediction error increases, thus generating clear preceding frames and blurry subsequent frames. To combat this characteristic, when calculating the error between the predicted and generated images, a logarithmic function is used to dynamically increase the weight of subsequent frames in the total loss. This forces the model to pay more attention to the prediction of subsequent frames, forming a temporal skew loss, and improving the model's overall sequence prediction accuracy.
[0057] This implementation trains a multi-channel, multi-temporal cloud image prediction network using the preprocessed data described above. The temporal tilt loss function guides the model to generate cloud image sequences that are more similar to the real cloud image sequences. The calculation formula is as follows:
[0058]
[0059] in, Represents a true multi-channel satellite cloud image. This represents the multi-channel satellite cloud imagery predicted by the model. Indicates batch size, This represents the number of frames for a single sample. and They represent the first The first sample and the first frame.
[0060] Step S6: Use the optimal parameters obtained from training to encapsulate and create a multi-channel, multi-temporal cloud image prediction model.
[0061] In this embodiment, the ADAM optimizer is selected to train the multi-channel multi-temporal cloud image prediction network by iterating 100 epochs using the above loss function. The parameters that yield the lowest loss on the validation set are selected as the final model parameters and packaged into a directly usable multi-channel multi-temporal cloud image prediction model.
[0062] Step S7: Use a multi-channel, multi-temporal cloud image prediction model to complete real-time multi-channel satellite cloud image prediction.
[0063] By performing the above preprocessing operations on the real-time multi-channel satellite cloud image and inputting it into the multi-channel multi-temporal cloud image prediction model, the predicted multi-channel satellite cloud image can be obtained.
[0064] Depend on Figure 3 It can be seen that the present invention has accurately predicted the long-term satellite cloud image movement trend and cloud cluster changes, providing important references for multiple fields such as precipitation estimation and disaster monitoring.
[0065] Figure 4 This invention provides a functional block diagram of a multi-temporal cloud image prediction device based on multi-channel satellite data. The device mainly includes: an acquisition module 10, a preprocessing module 20, a model training module 30, and a prediction module 40, wherein:
[0066] The acquisition module 10 is used to download GOES-16 ABI L1b Radiance Data from the NOAA website, with the bands being C08 (6.17μm), C09 (6.93μm), C10 (7.34μm), and C13 (10.33μm).
[0067] Preprocessing module 20 is used to preprocess metadata, including reading information such as variables and parameters contained in NetCDF format data by referring to the GOES-R satellite product manual document, obtaining corresponding band brightness temperature data by using the Planck function and band bandpass corrected reflectivity data, and normalizing data by using a logarithmic function-based normalization method.
[0068] Model training module 30 is used to train a multi-channel multi-temporal cloud image prediction network using the data that has undergone the aforementioned preprocessing operations, and to encapsulate the optimal parameters obtained from the training to obtain a multi-channel multi-temporal cloud image prediction model.
[0069] The prediction module 40 is used to complete real-time multi-channel satellite cloud image prediction using a multi-channel multi-temporal cloud image prediction model.
[0070] Figure 5 This is a schematic diagram of the hardware structure of a multi-temporal cloud image prediction device based on multi-channel satellite data provided in an embodiment of the present invention. The electronic device includes: a processor 60, a memory 61, a bus 62 and a communication interface 63. The processor 60, the communication interface 63 and the memory 61 are connected through the bus 62. The processor 60 is used to execute executable modules, such as computer programs, stored in the memory 61.
[0071] The memory 61 includes high-speed random access memory (RAM) and non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 63 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 62 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc.
[0072] The memory 61 stores the program, and the processor 60 executes the program after receiving the execution instruction to implement a multi-temporal cloud image prediction method based on multi-channel satellite data in this embodiment. The processor 60 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor 60 or by instructions in the form of software. The processor 60 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in this embodiment. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this invention can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 61, and processor 60 reads information from memory 61 and, in conjunction with its hardware, completes the steps of the above method.
[0073] This embodiment provides a computer program product for an attention-based multi-channel satellite cloud image prediction method and apparatus. It includes a computer-readable storage medium storing processor-executable non-volatile program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. Specific implementation details can be found in the method embodiments and will not be repeated here. Furthermore, the functional units in the various embodiments of this invention can be integrated into a single processing unit, or each unit can exist physically separately, or two or more units can be integrated into a single unit. If the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-executable non-volatile computer-readable storage medium. Based on this understanding, the technical solution of this invention, 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this embodiment. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media that can store program code.
[0074] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0075] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.
Claims
1. A multi-temporal cloud image prediction method based on multi-channel satellite data, characterized in that, Includes the following steps: Step 1: Download satellite cloud images for different bands; Step 2: Stitch together satellite cloud images from different bands to form a multi-channel satellite cloud image; Step 3: Calculate the brightness temperature value of each pixel in each satellite cloud image, thereby mapping the satellite cloud image to the brightness temperature value form; Step 4: Logarithmically normalize the brightness temperature values in each satellite cloud image; Step 5: Construct a multi-channel, multi-temporal cloud image prediction network, and use the normalized multi-channel satellite cloud images from Step 4 as the training set to train the multi-channel, multi-temporal cloud image prediction network, thus obtaining the trained multi-channel, multi-temporal cloud image prediction network; Step 6: Use the trained multi-channel, multi-temporal cloud image prediction network for prediction. The multi-channel multi-temporal cloud image prediction network includes an attention selection module, an iterative downsampling module, an iterative upsampling module, and a channel separation output module; The attention selection module includes a first stage and a second stage. The first stage is used to extract multi-scale deep features from the original input image, and these multi-scale deep features are input into the second stage to establish the correlation between features of different scales to obtain channel enhancement features. Then, the channel enhancement features are convolved to the same size as the input features to obtain spatial channel enhancement features. The spatial channel enhancement features are input into the iterative downsampling module to gradually extract the long and short-term temporal information contained in the spatial channel enhancement features, while learning the complex cloud structure and motion patterns hidden in the image. Then, the output features of the iterative downsampling module are input into the iterative upsampling module to restore the detailed features of the image. The output features of the iterative upsampling module are input into the channel separation module. The channel separation module uses C sets of independent filters to predict single-channel satellite cloud images, where C is the number of channels of the original input image input to the multi-channel multi-temporal cloud image prediction network. Then, the predicted single-channel cloud images are stitched together in the channel dimension to obtain the predicted multi-channel satellite cloud images. The first stage of the attention selection module includes n pooling layers, n upsampling modules, n element-wise multiplication modules, and a concatenation module. The i-th pooling layer is connected to the i-th upsampling module. The output of the i-th upsampling module is multiplied by the original image input to the attention selection module through the i-th element-wise multiplication module. The outputs of all element-wise multiplication modules are concatenated by the concatenation module to obtain multi-scale deep features. The second stage includes first to fourth reshaping modules, first and second matrix multiplication modules, a softmax module, an addition module, and a first convolution module. The first to third reshaping modules reshape the features input to the second stage into features of different dimensions. The features output by the second and third reshaping modules are multiplied by the first matrix multiplication module and then input to the softmax module. The softmax module multiplies the output features of the first reshaping module by the second matrix multiplication module and then inputs them to the fourth reshaping module. The features output by the fourth reshaping module are added to the features input to the second stage to obtain channel enhancement features. The obtained channel enhancement features are input to the first convolution module to obtain spatial channel enhancement features.
2. The multi-temporal cloud image prediction method based on multi-channel satellite data according to claim 1, characterized in that, In step 3, the brightness temperature value of each pixel in each satellite cloud image is calculated using the following formula: ； in, This represents the brightness temperature value of the b-th pixel in the satellite cloud image. This represents the pixel value of the b-th pixel in the satellite cloud image. and Let represent the coefficients of Planck's function, and It is the spectral response function offset and scale correction term.
3. The multi-temporal cloud image prediction method based on multi-channel satellite data according to claim 1, characterized in that, In step 4, a logarithmic function is used for normalization, as shown in the following expression: ； in, This represents the normalized brightness temperature value of the b-th pixel in the satellite cloud image. Indicates the minimum brightness temperature value. This represents the brightness temperature value of the b-th pixel in the satellite cloud image. To scale the size.
4. The multi-temporal cloud image prediction method based on multi-channel satellite data according to claim 1, characterized in that, The C group filters in the channel separation output module have the same structure, each including a second convolution module, a first deconvolution module, a third convolution module, and a fourth convolution module connected in sequence.
5. The multi-temporal cloud image prediction method based on multi-channel satellite data according to claim 1, characterized in that, The loss function of the multi-channel, multi-temporal cloud image prediction network As shown below: ； in, Represents a true multi-channel satellite cloud image. express The predicted value, where i represents the i-th sample and j represents the j-th frame. This indicates the batch size, and S represents the total number of frames in the cloud map within a sample.
6. A prediction system for a multi-temporal cloud image prediction method based on multi-channel satellite data, applied to the method described in any one of claims 1 to 5, characterized in that, include: The module includes a data acquisition module, a data processing module, a model building module, and a model training module. The data acquisition module is used to download satellite cloud images of different bands; The data processing module is used to execute steps 2 to 4; The model building module is used to construct a multi-channel, multi-temporal cloud image prediction network; The model training module is used to train the constructed multi-channel, multi-temporal cloud image prediction network.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of any one of the multi-temporal cloud image prediction methods based on multi-channel satellite data as claimed in claims 1 to 5.
8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run by the processor, it executes any one of the multi-temporal cloud image prediction methods based on multi-channel satellite data as described in claims 1 to 5.