A high temporal and spatial resolution optical image reconstruction method and system based on multi-source remote sensing data fusion

Abstract

The present invention discloses a method and system for reconstructing optical images with high spatiotemporal resolution based on the fusion of multi-source remote sensing data, aiming to solve the problem of missing spatiotemporal information of optical remote sensing images due to cloud cover, sensor failure and the like in the prior art. The method is as follows: 1. Acquisition, preprocessing and data pair construction of multi-source remote sensing data. 2. Constructing and training an optical image reconstruction model, the model comprising a generator that receives multi-source data as input and a discriminator. 3. Reconstructing optical images of missing phases using the trained generator to generate a complete continuous optical image time series with high spatiotemporal resolution. The present invention can effectively fill in the gaps in optical image data, and the generated images have high spatial and spectral accuracy. It can effectively process data of different resolutions and reduce the influence of geometric registration errors through a multi-input network structure, providing reliable data guarantee for remote sensing applications that rely on continuous optical observations.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing technology and earth observation technology, and specifically relates to a high-temporal and spatial resolution optical image reconstruction method and system based on multi-source remote sensing data fusion. Background Art

[0002] As an important source of Earth observation data, remote sensing images play an irreplaceable role in a wide range of applications, especially time-series optical images, which are of great value in various application scenarios such as surface phenology surveys, disaster prediction, and vegetation monitoring.

[0003] However, current optical imaging technology still has inherent limitations, mainly affected by clouds. Due to the high reflectivity of clouds in the optical band, optical signals have difficulty penetrating clouds, often resulting in image distortion, reduced clarity, and even complete obstruction of ground object information. According to statistics, cloud coverage rates are high globally and over land. Large-scale, long-term cloud cover can lead to a serious lack of temporal information in optical images, especially in cloudy and rainy areas, making it difficult to obtain continuous, cloud-free optical image sequences. In addition, problems such as sensor failures may also lead to image quality degradation or data loss. These problems have seriously hindered applied research that relies on continuous optical remote sensing image time series, such as regional crop disease and pest monitoring and evaluation.

[0004] Therefore, there is an urgent need for a method that can effectively fill in missing optical remote sensing data and reconstruct time-gap images to obtain complete and continuous high-temporal and spatial resolution optical time series data. Summary of the Invention

[0005] The purpose of the present invention is to provide a high-temporal-resolution optical image reconstruction method based on multi-source remote sensing data fusion, aiming to overcome the problem of missing spatiotemporal information of optical remote sensing images caused by cloud obstruction, sensor failure and other reasons in the existing technology, and generate high-temporal-resolution and complete continuous optical time series data, providing high-quality data guarantee for remote sensing applications such as regional-scale vegetation monitoring, disaster warning, and crop disease monitoring and evaluation.

[0006] In a first aspect, the present invention provides a method for reconstructing optical images with high temporal and spatial resolution based on multi-source remote sensing data fusion, comprising:

[0007] The existing cloud-free satellite optical data before and after the target phase are used as reference data, and the synthetic aperture radar data and medium-resolution imaging spectrometer data closest to the target phase are used as the first auxiliary data and the second auxiliary data respectively. They are input into the optical image reconstruction model to generate the satellite optical data of the target phase.

[0008] The optical image reconstruction model adopts a generative adversarial network structure. The reference data and the first auxiliary data are spliced ​​and input into the first encoding branch of the optical image reconstruction model. The first encoding branch performs layer-by-layer downsampling and feature extraction on the input data through residual convolution. The second auxiliary data is input into the second encoding branch of the optical image reconstruction model. The second encoding branch performs feature extraction on the input data through residual convolution. The output features of the first encoding branch and the second encoding branch are spliced ​​and input into the decoder. The decoder performs layer-by-layer upsampling of the input features through residual deconvolution. The outputs of the residual convolution layers of each layer in the first encoding branch are spliced ​​and fused with the inputs of the residual deconvolution layers of each layer in the decoder through jump connections.

[0009] Preferably, the optical image reconstruction model includes a total loss function comprising an adversarial loss, an L1 norm loss, and a spectral angle mapping loss. The spectral angle mapping loss is calculated by taking the mean of the angles between the spectral vectors of each pixel in the generated satellite optical data and the real satellite optical data.

[0010] Preferably, the residual convolution blocks in the first encoding branch include a two-dimensional convolution layer, a batch normalization layer and a leaky linear rectification activation function (Leaky ReLU), and use a residual connection structure to encode the reference data and the first auxiliary data layer by layer.

[0011] Preferably, the output of the last layer of residual deconvolution in the decoder is adjusted through multiple layers of residual convolution and one layer of convolution to adjust the number of feature channels, and then passes through a hyperbolic tangent activation function to obtain the satellite optical data of the target phase.

[0012] Preferably, the reference data, the first auxiliary data and the second auxiliary data are cropped to form image blocks and then input into the optical image reconstruction model; the optical image reconstruction result blocks output by the optical image reconstruction model are spliced ​​to form the satellite optical data of the target phase.

[0013] Preferably, the discriminator in the optical image reconstruction model generates confidence for the input image block, and includes multiple convolution blocks, flattening layers and fully connected layers connected in series.

[0014] Preferably, the optical image reconstruction model is trained using an image dataset. The image dataset is obtained by preprocessing the satellite optical data, synthetic aperture radar data, and medium-resolution imaging spectrometer data. The preprocessing process includes: truncating pixel values ​​in the image data that exceed the valid range; performing data normalization; cropping the image data using a sliding window to obtain image blocks that constitute the dataset; and expanding the number of samples in the dataset through data augmentation.

[0015] In a second aspect, the present invention provides a high-temporal-resolution optical image reconstruction system for performing the aforementioned high-temporal-resolution optical image reconstruction method. The high-temporal-resolution optical image reconstruction system includes a data acquisition module, a preprocessing module, and a reconstruction module. The data acquisition module is used to acquire synthetic aperture radar data, medium-resolution imaging spectrometer data, and cloud-free satellite optical data before and after the target phase. The preprocessing module is used to preprocess the satellite optical data, synthetic aperture radar data, and medium-resolution imaging spectrometer data. The reconstruction module includes a generator and a discriminator. The generator includes an encoder and a decoder. The encoder includes two encoding branches. The first encoding branch includes multiple residual convolution blocks connected in series. The second encoding branch also includes a residual convolution block. The output features of the first and second encoding branches are concatenated and input to the decoder. The decoder includes multiple residual deconvolution blocks connected in series, multiple residual convolution blocks, a convolution layer, and an activation function. The output of the residual convolution block in the first encoding branch of the corresponding layer is jump-connected to the input of the residual deconvolution block in the encoder.

[0016] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory stores the computer program; and the processor executes the aforementioned high spatiotemporal resolution optical image reconstruction method.

[0017] In a fourth aspect, the present invention provides a readable storage medium storing a computer program; when the computer program is executed by a processor, it is used to implement the aforementioned high temporal and spatial resolution optical image reconstruction method.

[0018] The present invention has the following beneficial effects:

[0019] 1. The present invention can effectively reconstruct optical remote sensing images that are missing due to cloud cover, sensor failure, etc., fill data gaps, and generate complete and continuous optical time series data with high temporal and spatial resolution, providing reliable data support for various remote sensing applications that rely on continuous optical observations.

[0020] 2. The present invention integrates the advantages of multi-source remote sensing data (including optical and SAR), adopts a specially designed optical image reconstruction model, and introduces spectral angle mapping loss into the model's loss function, effectively processing multi-source heterogeneous data and significantly improving the spatial and spectral accuracy of reconstructed images, especially in complex or data-missing areas.

[0021] 3. The multi-input network structure design adopted in the present invention can effectively process data of different resolutions, avoid information loss or distortion caused by resampling in traditional methods, reduce the impact of geometric alignment errors on model performance, and enhance the robustness and applicability of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flowchart of Example 1 of the present invention.

[0023] Figure 2 It is a structural diagram of the optical image reconstruction model in Example 1 of the present invention.

[0024] Figure 3 These are some of the generated image blocks and real image blocks in the Ningbo experimental area in Example 1 of the present invention.

[0025] Figure 4 This is a high temporal and spatial resolution optical time series data result diagram generated in the Ningbo experimental area from May 1 to November 25, 2021 in Example 1 of the present invention. DETAILED DESCRIPTION

[0026] A specific embodiment of the present invention is further described in detail below with reference to the accompanying drawings.

[0027] Example 1

[0028] like Figure 1 As shown, a high-temporal-resolution optical image reconstruction method based on multi-source remote sensing data fusion includes data acquisition, preprocessing and data pair construction, optical image reconstruction model construction and training, and optical image reconstruction and time series generation. This embodiment takes the Ningbo experimental area as an example, and uses the acquired Sentinel-2 (Sentinel-2) optical data, Sentinel-1 (Sentinel-1) SAR (Synthetic Aperture Radar) data and MODIS (Moderate-Resolution Imaging Spectroradiometer) optical data to build and train an optical image reconstruction model to achieve reconstruction of missing optical images and generate complete and continuous high-temporal-resolution optical time series data. The specific steps are as follows:

[0029] Step 1: Acquire, preprocess, and construct multi-source remote sensing data. Multi-source remote sensing image data within the Ningbo experimental area were obtained, including Sentinel-2 optical data, Sentinel-1 SAR data, and MODIS optical data. The corresponding selected band information for each data is shown in Table 1.

[0030] Table 1

[0031]

[0032] The acquired raw data is preprocessed. The specific process of preprocessing is as follows:

[0033] (1) Outlier processing: To ensure the training stability of the model, pixel values ​​in the original image data that are outside the valid range are truncated. Specifically, the pixel values ​​of Sentinel-2 images are truncated to the range of 0–10,000; the pixel values ​​of Sentinel-1 SAR images are truncated to the range of -50–1; and the pixel values ​​of MODIS images are truncated to the range of -100–16,000.

[0034] (2) Perform data standardization and standardize all image data after outlier processing so that their mean is 0 and their standard deviation is 1 to facilitate model convergence and training.

[0035] (3) Use a sliding window to crop the image to construct image patches suitable for model input. For high-resolution Sentinel-2 and Sentinel-1 SAR images, a 250 × 250 pixel sliding window is used for cropping, with a step size of 200 pixels to increase the amount of data and overlap between windows. For MODIS images, a 5 × 5 pixel sliding window is used to crop the image to match its resolution, with a step size of 4 pixels.

[0036] (4) A data augmentation strategy is adopted. To improve the robustness and generalization ability of the model, the cropped data is randomly flipped upside down and left to right, and the dataset is expanded to four times the original data.

[0037] Through the above preprocessing process, time-matched data pairs are finally constructed for model training, validation, and testing. These time-matched data pairs include SAR image patches and MODIS optical image patches as auxiliary model inputs, cloud-free Sentinel-2 optical image patches that are close in time to the target phase (cloud-free Sentinel-2 images that are adjacent to and before the target phase) as model reference inputs, and cloud-free Sentinel-2 optical image patches corresponding to the target phase as model outputs.

[0038] Step 2: Construct and train an optical image reconstruction model; the optical image reconstruction model uses a generative adversarial network model with MSF-ECGAN as the main body, such as Figure 2 As shown, the optical image reconstruction model includes a generator and a discriminator.

[0039] The generator receives reference Sentinel-2 image blocks, auxiliary SAR image blocks, and MODIS image blocks as input and outputs corresponding optical image reconstruction result blocks. The generator employs an improved encoder-decoder architecture, comprising an encoder and a decoder. The encoder includes two encoding branches. The first encoding branch consists of six residual convolution blocks (ResConvBlocks) connected in series. The second encoding branch consists of one residual convolution block. The input of the first encoding branch consists of the concatenated features of the reference Sentinel-2 image blocks and the auxiliary SAR image blocks. The input of the second encoding branch consists of the auxiliary MODIS image blocks. The decoder consists of six residual deconvolution blocks (ResDeconvBlocks) connected in series, three residual convolution blocks (ResConvBlocks), a 1×1 convolution layer, and a Tanh activation function. The output of the residual convolution block in the first encoding branch of the corresponding layer is skip-connected to the input of the residual deconvolution block in the encoder. The output features of the first encoding branch and the second encoding branch are concatenated and input into the decoder.

[0040] In the first encoding branch of the encoder, the input reference Sentinel-2 data and auxiliary SAR data (total input channels: 18) are first downsampled layer by layer through six serially connected residual convolutional blocks (ResConvBlocks) for feature extraction. Each residual convolutional block includes a 2D convolution layer, a batch normalization layer (BatchNorm), and an activation function (LeakyReLU). Using a residual connection structure, the features of the concatenated Sentinel-2 image block and SAR image block are encoded layer by layer to extract multi-scale deep features.

[0041] In the second encoding branch of the encoder, the auxiliary data MODIS image block (input channel 6) is extracted through an independent residual convolution block (input and output channels 6→6, convolution kernel size 3, stride 1, padding 1). Before entering the decoder, it is spliced ​​with the features output by the first encoding branch (1152 channels), and the number of feature channels after splicing is 1158.

[0042] The decoder consists of six residual deconvolution blocks (ResDeconvBlocks). Each ResDeconvBlock uses deconvolution to upsample the high-dimensional features extracted by the encoder layer layer by layer, gradually restoring spatial resolution. Furthermore, skip connections are used to concatenate and fuse the features of each encoder layer with the upsampled results of the corresponding decoder layer. This combines feature information at different scales and preserves more effective details and spatial structure. At the end of the decoder, three residual convolution blocks are serially connected, followed by a final 1×1 convolution layer to adjust the number of feature channels. After passing through the Tanh activation function, the final output is the optical image reconstruction block.

[0043] The optical image reconstruction result block output by the generator and the real target Sentinel-2 optical image block (8 bands) are respectively stitched with the corresponding SAR image block (2 bands) in the channel dimension to generate the corresponding stitched image block. The stitched image block generated based on the optical image reconstruction result block is a fake stitched block; the stitched image block generated based on the real target Sentinel-2 optical image block is a real stitched block.

[0044] The discriminator receives the stitched image patches as input, with a total of 10 input channels. It consists of seven convolutional blocks connected in series. Each convolutional block includes a two-dimensional convolutional layer (Conv2d), a batch normalization layer (BatchNorm), and an activation function (LeakyReLU). The input and output channels of the convolutional blocks are: 10→20, 20→40, 40→80, 80→40, 40→20, 20→10, 10→5, respectively. The corresponding convolution kernel sizes, strides, and padding are: (3, 2, 1), (3, 2, 1), (3, 2, 1), (2, 2, 0), (2, 2, 0), (2, 2, 0), (2, 2, 0). The extracted features are then flattened through a flattening layer and fed into a fully connected layer, which ultimately outputs a confidence score indicating whether the input image patch is “fake” (generated by the generator) or “real” (real data).

[0045] During the training process, a variety of loss functions are used to perform end-to-end optimization training on the optical image reconstruction model. It is composed of a weighted combination of adversarial loss, L1 loss and spectral angle (SAM) mapping loss, and its form is shown in formula (1):

[0046] (1)

[0047] in, is the adversarial loss, using the least squares GAN loss, as shown in formula (2):

[0048] L LSCGAN = 1 2 [ D ( real , anc ) − 1 ] 2 + 1 2 [ D ( G ( ref , anc ), anc )] 2 (2)

[0049] is the pixel loss, using the L1 norm, as shown in formula (3):

[0050] (3)

[0051] is the spectral angle mapping loss, which is used to measure the spectral similarity between the generated image and the real image, as shown in formula (4):

[0052] (4)

[0053] D represents the discriminator, G represents the generator; Represents a real image, Indicates the generation of images, Indicates auxiliary image, represents the reference image; H, W, and C represent the height, width, and number of channels of the image block, respectively; and are the weight coefficients of L1 loss and SAM loss respectively. The optimal weight coefficient is determined by repeated experiments and is set to =100, =10.

[0054] During the model training phase, the constructed time-matched data pairs are divided into training, validation, and test sets. In this embodiment, 9432 data points of the total data set (a total of 11394 data points) are used for the training set, 1038 data points are used for the validation set, and 924 data points are used for the test set. During the training process, the data batch size is set to 8. The Adam optimizer is used for parameter optimization of both the generator and the discriminator, with an initial learning rate of 0.0002 and a momentum parameter of 0. 、 .

[0055] Step 3: Using the trained optical image reconstruction model, the generator takes the corresponding auxiliary data image blocks (SAR and MODIS) and reference data image blocks (Sentinel-2 for the preceding and following phases) for the time period to be reconstructed. The generator then outputs the reconstructed optical image blocks for that phase. This process is repeated for all missing phases in the time series, and the reconstructed image blocks are stitched together to ultimately produce a complete, continuous optical image time series with high spatial and temporal resolution.

[0056] Test results at the Ningbo experimental site show that the reconstruction method provided by this embodiment achieves good accuracy. Using commonly used image quality assessment metrics for evaluation, the obtained mean absolute error (MAE) is 0.0131, the root mean square error (RMSE) is 0.0005, the peak signal-to-noise ratio (PSNR) is 36.2754 dB, the spectral angle mapping (SAM) is 0.0518, and the structural similarity index (SSIM) is 0.9431. Figure 3 The following figure shows some generated image blocks and real image blocks in the Ningbo experimental area. The generated image results are highly consistent with the real image in terms of spatial structure, spatial texture and ground object spectrum. Figure 4 Shown are the results of continuous high-temporal and spatial resolution optical time series data obtained at the Ningbo experimental area from May 1 to November 25, 2021. These results demonstrate that the reconstruction method provided in this embodiment can generate optical images that are highly similar to real images, with high spatial and spectral fidelity.

[0057] Example 2

[0058] A high-temporal-resolution optical image reconstruction system is configured to execute the high-temporal-resolution optical image reconstruction method provided in Example 1. The high-temporal-resolution optical image reconstruction system comprises a data acquisition module, a preprocessing module, and a reconstruction module. The data acquisition module is configured to acquire synthetic aperture radar data, medium-resolution imaging spectrometer data, and satellite optical data with missing target phases.

[0059] The preprocessing module is used to preprocess satellite optical data, aperture radar data, and medium-resolution imaging spectrometer data. The reconstruction module includes a generator and a discriminator. The generator includes an encoder and a decoder. The encoder includes two encoding branches. The first encoding branch includes six residual convolution blocks connected in series.

[0060] The input and output channels of the six residual convolution blocks (ResConvBlock) in the first encoding branch are 18→36, 36→72, 72→144, 144→288, 288→576, 576→1152, respectively; the corresponding convolution kernel sizes, strides, and padding are (3, 2, 1), (3, 2, 2), (3, 2, 1), (2, 2, 0), (2, 2, 0), (2,2, 1), respectively.

[0061] The second encoding branch includes a residual convolution block. The output features of the first and second encoding branches are concatenated and input to the decoder. The decoder includes six residual deconvolution blocks connected in series, three residual convolution blocks, a 1×1 convolution layer, and an activation function.

[0062] The input and output channels of the six residual deconvolution blocks (ResDeconvBlock) are: 1158→579, 866→578, 577→433, 360→288, 126→90, 32→16; the corresponding deconvolution kernel sizes, strides, and padding are: (4, 1, 0), (2, 2, 0), (2, 2, 0), (2, 2, 0), (3, 2, 2), (2, 2, 0).

[0063] The input and output channels of the three serial residual convolution blocks are: 50→24, 24→16, 16→8, the convolution kernel size is 3, the stride is 1, and the padding is 1.

[0064] The input and output channels of the 1×1 convolutional layer are 8→8, the convolution kernel size is 1, the stride is 1, and the padding is 0.

[0065] The output of the residual convolution block in the first encoding branch of the corresponding layer is skip-connected to the input of the residual deconvolution block in the encoder.

[0066] Example 3

[0067] An electronic device, specifically, the electronic device includes a memory and a processor, the memory stores executable code, and when the processor executes the executable code, the method described in embodiment 1 is implemented.

[0068] The memory may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface (which may be wired or wireless), such as the Internet, a wide area network (WAN), a local area network (LAN), or a metropolitan area network (MAN).

[0069] The bus may be an ISA bus, a PCI bus or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc.

[0070] The memory is used to store the program, and the processor executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any of the aforementioned embodiments of the present invention can be applied to the processor or implemented by the processor.

[0071] The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor 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. The methods, steps, and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or the like. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0072] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be referred to the previous method embodiment and will not be repeated here.

[0073] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion 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, server, or 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 media include various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0074] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A high spatiotemporal resolution optical image reconstruction method based on multi-source remote sensing data fusion, characterized in that: The method comprises: The existing cloud-free satellite optical data before and after the target phase is used as reference data. The synthetic aperture radar data and moderate resolution imaging spectrometer data closest to the target phase are used as the first auxiliary data and the second auxiliary data, respectively. These data are input into the optical image reconstruction model to generate the satellite optical data of the target phase. Generate satellite optical data for all missing phases in the time series to obtain a complete continuous optical image time series with high temporal and spatial resolution; The optical image reconstruction model adopts a generative adversarial network structure; the reference data and the first auxiliary data are spliced ​​and input into the first encoding branch of the optical image reconstruction model; the first encoding branch performs layer-by-layer downsampling feature extraction on the input data through multiple residual convolution blocks connected in series; the residual convolution blocks each include a two-dimensional convolution layer, a batch normalization layer, and a leaky linear rectification activation function, and utilize a residual connection structure to encode the reference data and the first auxiliary data layer by layer; the second auxiliary data is input into the second encoding branch of the optical image reconstruction model; the second encoding branch performs feature extraction on the input data through an independent residual convolution block; the output features of the first encoding branch and the second encoding branch are spliced ​​and input into the decoder; the decoder performs layer-by-layer upsampling on the input features through residual deconvolution; the output of each layer of residual convolution in the first encoding branch is spliced ​​and fused with the input of each layer of residual deconvolution in the decoder through jump connections; The input data of the discriminator in the optical image reconstruction model includes synthetic aperture radar data; the total loss function of the optical image reconstruction model The total loss function is a weighted combination of adversarial loss, L1 loss and spectral angle mapping loss. The expression is: ;in, It's about fighting losses. is the pixel loss, is the spectral angle mapping loss; the spectral angle mapping loss is the mean of the angles between the spectral vectors of each pixel in the generated satellite optical data and the real satellite optical data.

2. The high spatiotemporal resolution optical image reconstruction method according to claim 1, wherein: The output of the last residual deconvolution layer in the decoder is subjected to multi-layer residual convolution and one-layer convolution to adjust the number of feature channels, and then passes through a hyperbolic tangent activation function to obtain satellite optical data of the target phase; instance normalization is added to the residual deconvolution.

3. The high spatiotemporal resolution optical image reconstruction method according to claim 1, wherein: The reference data, the first auxiliary data and the second auxiliary data are cut to form image blocks and then input into the optical image reconstruction model; the optical image reconstruction result blocks output by the optical image reconstruction model are spliced ​​to form satellite optical data of the target phase.

4. The high spatiotemporal resolution optical image reconstruction method according to claim 1, wherein: The discriminator in the optical image reconstruction model generates confidence for the input image block, and includes multiple convolution blocks, flattening layers and fully connected layers connected in series.

5. The high spatiotemporal resolution optical image reconstruction method according to claim 1, wherein: The optical image reconstruction model is trained using an image dataset; the image dataset is obtained by preprocessing the satellite optical data, synthetic aperture radar data, and medium resolution imaging spectrometer data; The preprocessing process includes: truncating pixel values ​​in the image data that exceed the valid range; data standardization; using a sliding window to crop the image data to obtain image blocks that constitute the data set; and expanding the number of samples in the data set through data enhancement.

6. A high temporal and spatial resolution optical image reconstruction system, characterized by: Used to execute the high spatiotemporal resolution optical image reconstruction method as described in claim 1; the high spatiotemporal resolution optical image reconstruction system includes a data acquisition module, a preprocessing module and a reconstruction module; the data acquisition module is used to obtain synthetic aperture radar data, medium resolution imaging spectrometer data, and cloud-free satellite optical data before and after the target phase; the preprocessing module is used to preprocess the satellite optical data, synthetic aperture radar data, and medium resolution imaging spectrometer data; the reconstruction module includes a generator and a discriminator; the generator includes an encoder and a decoder; the encoder includes two encoding branches; the first encoding branch includes a plurality of residual convolution blocks connected in series; the second encoding branch includes a residual convolution block; the output features of the first encoding branch and the second encoding branch are spliced ​​and input into the decoder; the decoder includes a plurality of residual deconvolution blocks connected in series, a plurality of residual convolution blocks, a convolution layer and an activation function; the output of the residual convolution block in the first encoding branch of the corresponding layer is jump-connected to the input of the residual deconvolution block in the encoder.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The memory stores a computer program; the processor executes the high temporal and spatial resolution optical image reconstruction method according to any one of claims 1 to 6.

8. A readable storage medium storing a computer program; characterized in that: When the computer program is executed by a processor, it is used to implement the high spatiotemporal resolution optical image reconstruction method according to any one of claims 1 to 5.

Images