Joint inversion method for atmosphere and sea color parameters under sparse remote sensing satellite data

Abstract

The invention discloses an atmosphere and sea color parameter joint inversion method under sparse remote sensing satellite data. The method comprises the following steps: firstly, acquiring various satellite remote sensing and sea color data in a unified format; secondly, using the obtained satellite remote sensing and sea color data to construct a sparse atmosphere sea color data set through pixel mask and combining observation geometric information carried by the satellite remote sensing data; then, based on remote sensing waveband data processing, a sparse remote sensing depth inversion network comprising a sparse processing network, an external depth estimation network and a scale fusion network is constructed; and finally, training the sparse remote sensing depth inversion network through the sparse atmospheric sea color data set, outputting an inversion result image, and performing model evaluation. According to the method, joint estimation of atmosphere and sea color multiple parameters can be realized, the inversion precision in a complex ocean and atmosphere interaction scene is improved, and reliable technical support is provided for high-precision atmosphere correction and ocean carbon flux estimation.

Description

Technical Field

[0001] This method belongs to the fields of deep learning and atmospheric remote sensing, and specifically relates to a method for jointly retrieving atmospheric and ocean color parameters from sparse remote sensing satellite data. Background Art

[0002] As the core data source for global ocean ecological parameter monitoring, the multi-source ocean color satellite collaborative observation system provides crucial technical support for improving the understanding of the ocean ecosystem, optimizing the sustainable utilization efficiency of ocean resources, formulating climate response strategies, and implementing marine environmental protection, and has important scientific value and practical significance.

[0003] Traditional ocean color remote sensing atmospheric correction algorithms are based on the assumption that the water-leaving reflectance approaches zero in the near-infrared and short-wave infrared bands, and construct an aerosol scattering radiation lookup table through a radiative transfer model to retrieve the aerosol optical properties. This method shows good applicability in waters with non-absorbing and weakly absorbing aerosols, but has significant limitations in the face of complex atmospheric environments dominated by strongly absorbing aerosols, restricting the universal application of the algorithm.

[0004] Currently, satellite remote sensing technology has become an effective means of obtaining aerosol vertical distribution characteristics and ocean water component information. However, limited by the insufficient spatial coverage and time resolution defects of the observation system, existing high-vertical-resolution aerosol products are difficult to meet the refined requirements of ocean color remote sensing atmospheric correction. It is worth noting that although advanced sensors such as the Tropospheric Monitoring Instrument and the Multi-Angle Polarization Radiometer introduce Gaussian-type vertical distribution assumptions in aerosol inversion, such prior information has not been systematically integrated into the atmospheric correction algorithm framework.

[0005] Satellite remote sensing data is restricted by cloud interference and revisit cycles, and generally has defects such as low effective observation rate and insufficient time continuity, seriously affecting the quantitative inversion accuracy of ocean water components. As the core preprocessing link of ocean color remote sensing, atmospheric correction aims to eliminate the influence of the atmosphere on the apparent reflectance, and decouple the atmosphere-ocean coupling process through physical modeling to accurately obtain key parameters such as water-leaving radiance. Traditional methods face the dual challenges of difficult decoupling of complex physical processes and insufficient computational stability when satellite data is severely lacking. Summary of the Invention

[0006] To solve the problems of difficult decoupling of complex physical processes and insufficient computational stability faced when satellite data is severely lacking, this method provides a method for jointly retrieving atmospheric and ocean color parameters from sparse remote sensing satellite data. The present invention constructs a deep learning model that couples aerosol vertical distribution characteristics, breaks through the parameterization limitations of traditional algorithms, and aims to improve the inversion accuracy and algorithm robustness of ocean color elements under complex atmospheric conditions. The present invention includes the following steps:

[0007] 1. Obtain a variety of satellite remote sensing and ocean color data in a unified format;

[0008] 2. Use the satellite remote sensing and ocean color data obtained in step 1 to construct a sparse atmospheric ocean color dataset through pixel masking and combined with the observation geometry information of the satellite remote sensing data;

[0009] 3. Taking remote sensing band data processing as the main line, construct a sparse remote sensing depth inversion network, including sparse processing network, external depth estimation network and scale fusion network;

[0010] 4. The sparse remote sensing depth inversion network is trained through the sparse atmospheric and ocean color data set, and the inversion result image is output. After the training is completed, a single satellite image at a specific time is obtained for model evaluation.

[0011] Preferably, the features of the sparse atmospheric ocean color dataset in step 2 include observation geometry, land surface information and remote sensing band information after 95% pixel masking, and the label is the ocean color data in step 1. The observation geometry includes solar azimuth angle (SAA), solar zenith angle (SZA), observation azimuth angle (VAA), and observation zenith angle (VZA). The land surface information includes normalized difference vegetation index (NDVI). The remote sensing band information consists of 7 bands, and the central wavelengths are respectively in the following nanometer (nm) ranges: [0.45, 0.52], [0.52, 0.60], [0.63, 0.69], [0.77, 0.90], [1.55, 1.75], [2.08, 2.35], and [0.52, 0.90].

[0012] Preferably, the sparse remote sensing depth inversion network in step 3 includes a sparse processing network (I), an external depth estimation network (II) and a scale fusion network (III). The sparse processing network is used to extract sparse remote sensing band information. The external depth estimation network jointly encodes the land surface information and the observation geometry. The two networks perform parallel calculations to obtain sparse remote sensing band information and joint encoding guidance information respectively. The scale fusion network guides the reconstruction of the sparse remote sensing band information according to the joint encoding guidance information and finally obtains the inversion result image. The three parts are introduced respectively as follows:

[0013] (I) The sparse processing network extracts and completes features from sparse satellite remote sensing data, and shortens the mapping distance from sparse features to labels through a pre-reconstruction method. Its structure is a discrete encoder-decoder network. The encoder is a Vision Transformer (ViT) network with a single-layer multi-layer perceptron network (MLP) as the output head, and the decoder is a convolutional neural network (CNN network) with a U-Net architecture; the discrete encoder-decoder network uses a discrete training method, that is, for the encoder network, the remote sensing band information and dataset labels of the sparse atmospheric ocean color dataset proposed by this method are used as training data for independent training. The output of the trained encoder network and the labels of the sparse atmospheric ocean color dataset are used as training data to train the decoder network. Connecting the trained encoder and decoder in series gives the pre-trained discrete encoder-decoder network; a network consisting of n layers of residual networks containing convolutional layers and normalization and a single convolutional layer is connected after the discrete encoder-decoder network to obtain the to-be-reconstructed features; the discrete encoder-decoder network is pre-trained and the parameters are frozen after training.

[0014] (II) The external depth estimation network jointly encodes land surface information and observation geometry to obtain joint encoded information and helps in pixel reconstruction and mapping for the atmospheric correction and ocean color prediction tasks in Network III; for land surface information, the same discrete encoder-decoder network as in (I) is used, and for observation geometry, a non-discrete encoder-decoder with the same architecture as in (I) without using the discrete training method is used. The queries and keys of the ViT in the non-discrete encoder-decoder come from Network I; the output of the land surface information passing through the discrete encoder-decoder network and the output of the observation geometry passing through the non-discrete encoder-decoder network are concatenated in channels, passed through n layers of joint external information encoding modules, and then passed through a single convolutional layer for channel dimensionality reduction to obtain the joint encoded guidance information.

[0015] The joint external information encoding module includes a self-attention module, specifically: the input passes through a single convolutional layer and normalization, and uses itself as the query, key, and value respectively through a single convolutional layer. After multiplying the query and key values and weightifying them through Softmax to obtain the attention weights, multiplying them with the value passing through the convolutional layer to obtain the output. After residual connection between this output and the input passing through a single convolutional layer, the output of this layer module is obtained.

[0016] (III) The scale fusion network performs scale fusion on the to-be-reconstructed features obtained by Network I and the joint coding guidance information obtained by Network II to obtain the final atmospheric or ocean color component data, i.e., the inversion result image. The scale fusion network includes several layers of semantic prior fusion (SPF) and guidance reconstruction layers. The first guidance reconstruction layer receives the to-be-reconstructed features, the nth guidance reconstruction layer receives the output of the (n - 1)th SPF layer (6 ≥ n ≥ 2), the first SPF layer receives the joint coding guidance information output in (II) and the output of the first guidance reconstruction layer, the nth SPF layer receives the output of the (n - 1)th SPF layer and the output of the nth guidance reconstruction layer (5 ≥ n ≥ 2), and the last guidance reconstruction layer receives the output of the last SPF layer to obtain the final atmospheric correction or ocean color component data; the SPF uses the same self-attention module as in Network II, and the guidance reconstruction layer is a U-Net network;

[0017] Preferably, 90% of the data in the dataset is selected as the training dataset and 10% of the data is selected as the test dataset in step 4; first, the separate encoder-decoder network is trained: for the ViT-based encoder network, the satellite image patch_size is 8, the number of layers is 5, the number of self-attention heads is 8, dropout is 0.1, the batch size is 4, the learning rate is 1e-4, and the number of iterations is 50 times, and it is trained on 2 RTX4070super; for the CNN-based decoder network, the number of input channels and output channels is the same as the number of label channels, the batch size is 12, and the other parameters are the same as the encoder training parameters, and the output of the encoder network and the label are used as training data to train the decoder network; the sparse remote sensing depth inversion network freezes the parameters of the trained separate network and trains it on the sparse atmospheric-ocean color dataset proposed in this method, and the training parameters and loss function are the same as those of the ViT network training; the trained network is used to perform feasibility detection on real single-satellite images.

[0018] The beneficial effects of the present invention:

[0019] (1) The present invention proposes a method for joint inversion of atmospheric-ocean color parameters from sparse remote sensing satellite data. By constructing a cross-modal coupling architecture that combines physical constraints and deep learning, it effectively solves the problem of radiation transfer decoupling caused by insufficient spatial coverage of satellite data; this method can achieve joint estimation of multiple atmospheric-ocean color parameters, improve the inversion accuracy in complex ocean-atmosphere interaction scenarios, and provide reliable technical support for high-precision atmospheric correction and ocean carbon flux estimation.

[0020] (2)The present invention proposes a sparse remote sensing depth inversion network with multi-modal feature interaction. By constructing a hybrid structure of sparse feature extraction, external prior encoding, and scale fusion, it can effectively solve the ill-posed radiative transfer inversion problem under sparse observation conditions and achieve high-precision joint reconstruction of atmosphere-ocean color parameters. Description of the Drawings

[0021] Figure 1 It is a flow chart of the depth atmospheric correction and ocean color inversion method for sparse remote sensing satellite data;

[0022] Figure 2 It is the network architecture diagram proposed by the present invention;

[0023] Figure 3 It is a display diagram of remote sensing band information at the central wavelength of [0.45, 0.52] nm passing through a 95% pixel mask;

[0024] Figure 4 It is the 550 nm aerosol inversion result of the discrete encoder-decoder network of the present invention in atmospheric correction;

[0025] Figure 5 It is the scatter density diagram of the 550 nm aerosol inversion of the discrete encoder-decoder network of the present invention in atmospheric correction;

[0026] Figure 6 It is the 405 - 420 nm water-leaving radiance inversion result of the discrete encoder-decoder network of the present invention;

[0027] Figure 7 It is the scatter density diagram of the 405 - 420 nm water-leaving radiance inversion of the discrete encoder-decoder network of the present invention. Detailed Embodiments

[0028] To better illustrate the invention and advantages of this project, the following further explains the invention content in combination with the drawings and examples.

[0029] Example 1:

[0030] This method provides a method for jointly inverting atmosphere and ocean color parameters from sparse remote sensing satellite data, as Figure 1 shown, including the following steps:

[0031] 1. Use a predetermined JavaScript language algorithm to obtain various satellite remote sensing and ocean color data in a unified format on the Google Earth Engine platform;

[0032] 2. Use the satellite remote sensing and ocean color data obtained in step 1 to simulate the serious lack of satellite remote sensing data through 95% pixel mask, and combine the observation geometry information of satellite remote sensing data to construct a sparse atmospheric ocean color dataset;

[0033] 3. Based on remote sensing band data processing, a sparse remote sensing depth inversion network is constructed, including a sparse processing network, an external depth estimation network, and a scale fusion network;

[0034] 4. The sparse remote sensing depth inversion network is trained using a sparse atmospheric and ocean color dataset to obtain an inversion result image. After the training is completed, a single satellite image at a specific time is obtained for model evaluation.

[0035] The step 1 uses the planetary-level geospatial analysis platform Google Earth Engine platform (GEE platform) to obtain ocean color data, wherein the ocean color data includes apparent reflectance data, water-off radiance data, and atmospheric aerosol data, and takes the median of the effective values ​​in the study area within one acquisition cycle, and the study area is the Bohai Sea area of ​​​​China; the GEE platform is used to collect ocean color data for a total of 18 years from 2002 to 2020, with an acquisition cycle of 8 days and a downsampling resolution of 1000m; the images collected using GEE contain valid parts, missing parts, and invalid parts, wherein the valid part is the pixel with a value greater than 0 in the study area, the missing part is the part with a value less than or equal to 0 in the study area, and the invalid part is the part outside the study area in the image.

[0036] Remote sensing band information contains the most important and abundant data for remote sensing processing. Land surface information provides water body conditions, which helps to guide atmospheric correction and ocean color prediction research when there are changes in the geographical and water body environments. Observation geometry is an important part of the basic data of remote sensing images and is one of the essential parameters for atmospheric and ocean color inversion; the characteristics of the sparse atmospheric and ocean color dataset described in step 2 include the observation geometry, land surface information, and remote sensing band information that have passed through 95% pixel masking, with the label being the ocean color data described in step 1. The observation geometry includes the solar azimuth angle (SAA), solar zenith angle (SZA), viewing azimuth angle (VAA), and viewing zenith angle (VZA). The land surface information includes the normalized difference vegetation index (NDVI). The remote sensing band information consists of 7 bands, and their central wavelengths are respectively located in the following nanometer (nm) magnitude intervals: [0.45, 0.52], [0.52, 0.60], [0.63, 0.69], [0.77, 0.90], [1.55, 1.75], [2.08, 2.35], [0.52, 0.90]; the 95% pixel masking constitutes an interpolation ineffectiveness and a non-obvious meaningful remote sensing inversion task, which can greatly eliminate the domain gap between the data collected by the collection algorithm and the real world. The remote sensing band information with a central wavelength of [0.45, 0.52] nm through the 95% pixel masking is as Figure 3 shown.

[0037] For the atmospheric correction and ocean color inversion tasks from the aspects of remote sensing band data, land surface information, and observation geometry, as Figure 2 shown, this method proposes the sparse remote sensing depth inversion network described in step 3, including a sparse processing network (Ⅰ), an external depth estimation network (Ⅱ), and a scale fusion network (Ⅲ). The sparse processing network is used to extract sparse remote sensing band information. The external depth estimation network jointly encodes the land surface information and the observation geometry. The two networks perform parallel calculations to obtain sparse remote sensing band information and joint encoding guidance information respectively. The scale fusion network guides the reconstruction of the sparse remote sensing band information according to the joint encoding guidance information and finally obtains the inversion components. The following will introduce the three parts separately:

[0038] (I) Atmospheric correction and ocean color inversion require as much relevant and effective information as possible. The sparse processing network performs feature extraction and completion tasks on sparse remote sensing satellite data, shortening the mapping distance from sparse features to labels through a pre-reconstruction method. Transformer has a natural advantage in sparse processing tasks due to its powerful global modeling ability and broader inductive bias. Convolutional neural network (CNN) has better pixel-level output ability, which can further improve the pixel prediction accuracy. However, directly cascading and training the two simply will cause the CNN network to affect the modeling ability of the Transformer network. Therefore, this method proposes a sparse processing network. Specifically, the sparse processing network is a discrete encoder-decoder network. The encoder is a Vision Transformer (ViT) network with a multi-layer perceptron network (MLP) as the output head, and the decoder is a CNN network with a U-Net architecture. The discrete encoder-decoder network represents a discrete training method, that is, for the encoder network, the remote sensing band information and dataset labels of the sparse atmospheric ocean color dataset proposed by this method are used as training data for independent training. The output of the trained encoder network and the labels of the sparse atmospheric ocean color dataset are used as training data to train the decoder network. Connecting the trained encoder and decoder in series gives the pre-trained discrete encoder-decoder network. The discrete network is pre-trained and the parameters are frozen after training. This network can effectively combine the global modeling ability of the Transformer and the pixel-level output ability of the CNN network, providing effective prior information for the final atmospheric correction and ocean color prediction.

[0039] A network consisting of n layers of residual networks containing convolutional layers and normalization and one convolutional layer in series follows the discrete encoder-decoder network to obtain the to-be-reconstructed features.

[0040] (II) The external depth estimation network aims to jointly encode land surface information and observation geometry to obtain high-quality guiding information and assist in the pixel reconstruction and mapping of the atmospheric correction and ocean color prediction tasks in Network III. Sparse land surface information requires feature extraction and completion. Therefore, the same discrete encoder-decoder network as in (I) is used for land surface information. Observation geometry usually does not require high precision. Only a non-discrete encoder-decoder with the same architecture is used to maintain the semantic space consistency, and its queries and keys in the ViT come from Network I. The outputs of the two are concatenated in channels, passed through n layers of joint external information encoding modules, and then passed through one convolutional layer for channel dimensionality reduction to obtain the jointly encoded guiding information.

[0041] The combined external information encoding module includes a self-attention module, specifically: the input passes through a convolutional layer and normalization, and then passes through a convolutional layer respectively with itself as the query, key, and value. After multiplying the query and key values and weighting them with Softmax to obtain the attention weights, multiplying the obtained attention weights with the value passed through the convolutional layer to obtain the output, and then performing a residual connection between this output and the input passed through a convolutional layer to obtain the output of this layer module.

[0042] (III) The scale fusion network performs scale fusion on the to-be-reconstructed features obtained by Network I and the joint coding guidance information obtained by Network II to obtain the final atmospheric correction or ocean color component data; specifically, it includes 5 layers of semantic prior fusion (SPF) and 6 layers of guidance reconstruction layers. The first layer of the guidance reconstruction layer receives the to-be-reconstructed features, the nth layer of the guidance reconstruction layer receives the output of the (n - 1)th layer of SPF (6 ≥ n ≥ 2), the first layer of the SPF layer receives the joint coding guidance information output in (II) and the output of the first layer of the guidance reconstruction layer, the nth layer of the SPF receives the output of the (n - 1)th layer of the SPF layer and the output of the nth layer of the guidance reconstruction layer (5 ≥ n ≥ 2), and finally the sixth layer of the guidance reconstruction layer receives the output of the fifth layer of the SPF to obtain the final atmospheric correction or ocean color component data; the SPF uses the same self-attention module as in Network II, and the guidance reconstruction layer is a U-Net network.

[0043] According to the method of the above embodiments, the performance analysis of the deep atmospheric correction and ocean color inversion method for sparse remote sensing satellite data is carried out according to the following step 4:

[0044] Example 2:

[0045] 1. Training and validation of the separate encoder-decoder network

[0046] Select 90% of the data in the dataset as the training dataset and 10% of the data as the test dataset;

[0047] For the ViT-based encoder network, the satellite image patch_size is 8, the number of layers is 5, the number of self-attention heads is 8, dropout is 0.1, the batch size is 4, the learning rate is 1e-4, and the number of iterations is 50 times. It is trained on 2 RTX4070super, and the loss function is as follows:

[0048]

[0049] where represents the model parameters, represents the network input, represents various dataset labels, represents the image mask, 1 represents the valid part, and 0 represents the invalid part and the missing part, Take 1.0.

[0050] For the CNN-based decoder network, the number of input channels and output channels is the same as the number of label channels. For example, the number of label channels for the atmospheric correction task is 9, the number of label channels for aerosol optical depth inversion is 1, the batch size is 12, and the other parameters are the same as those for the encoder training. The output of the encoder network and the labels are used as training data to train the decoder network, and the training is carried out on the loss function in formula (1).

[0051] Figure 4 Shows the 550nm aerosol inversion results of the proposed separate encoder-decoder network for atmospheric correction. Figure 6 Shows the remote water-leaving radiance inversion results at 405 - 420nm of the proposed separate encoder-decoder network, which respectively include sparse feature input, ViT encoder output, CNN decoder output, label ground truth, and pixel scatter density map. Experiments show that the ViT encoder can extract relatively complete and accurate target information from sparse information. However, affected by the cutting of image patches (Patches), it shows a strong "split block feeling" macroscopically. The CNN decoder network can refine and reconstruct the output of the encoder at the pixel level, and finally obtain a predicted image highly similar to the label. Figure 5 and Figure 7 Show the corresponding Figure 4 and Figure 6 scatter density maps, recording the correlation coefficient (R 2 ), root mean square error (RMSE), mean absolute percentage error (MAPE), and the number of test points (Number). Experiments show that the network with a separate encoder-decoder can invert approximate real atmospheric correction information or ocean color component information from sparse satellite data, and can greatly reduce the mapping distance from sparse features to labels.

[0052] 2. Training method for sparse remote sensing depth inversion network

[0053] The sparse remote sensing depth inversion network freezes the parameters of the separate network in Example 2.1, reducing the computing power required by the model while maintaining the feature extraction ability of the separate network. It is trained on the sparse atmospheric and ocean color dataset proposed in this method, and the training parameters are the same as those of the ViT network in Example 2.1. Formula (1) is used as the loss function for training.

[0054] The proposed method for depth atmospheric correction and ocean color inversion from sparse remote sensing satellite data introduces external auxiliary data of land surface information and observation geometry, which can reflect the changing state of the real-world environment. With the assistance of external data of land surface information and observation geometry, the inversion ability from sparse satellite data can be further improved, realizing efficient atmospheric correction and ocean color inversion capabilities.

Claims

1. A joint inversion method for atmosphere and ocean color parameters based on sparse remote sensing satellite data, characterized in that: The following steps are involved: Step 1. Obtain a variety of satellite remote sensing and ocean color data; Step 2. Use satellite remote sensing and ocean color data to construct a sparse atmospheric ocean color dataset through pixel masking and combined with the observation geometry information of satellite remote sensing data; Step 3. Based on remote sensing band data processing, a sparse remote sensing depth inversion network including a sparse processing network, an external depth estimation network and a scale fusion network is constructed; Step 4. Train the sparse remote sensing depth inversion network using the sparse atmospheric and ocean color dataset, output the inversion result image, and evaluate it.

2. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 1, characterized in that: The features of the sparse atmospheric and ocean color data set include pixel-masked observation geometry, land surface information, and remote sensing band information, and the label is the ocean color data described in step 1; the observation geometry includes solar azimuth, solar zenith angle, observation azimuth, and observation zenith angle, and the land surface information includes the normalized difference vegetation index NDVI.

3. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 2, characterized in that: The sparse remote sensing depth inversion network includes a sparse processing network, an external depth estimation network and a scale fusion network. The sparse processing network is used to extract sparse remote sensing band information. The external depth estimation network jointly encodes land surface information and observation geometry. The two networks perform parallel calculations to obtain sparse remote sensing band information and joint encoding guidance information respectively. The scale fusion network guides the reconstruction of the sparse remote sensing band information according to the joint encoding guidance information and finally obtains an inversion result image.

4. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 3, characterized in that: The specific implementation process of the sparse processing network is as follows: The sparse processing network extracts and completes features of sparse satellite remote sensing data, shortens the mapping distance from sparse features to labels by a pre-reconstruction method, and has a structure of a pre-trained discrete encoder-decoder network, wherein the encoder is a ViT network with a multi-layer perceptron network as an output head, and the decoder is a convolutional neural network with a U-Net architecture; the discrete encoder-decoder network uses a discrete training method, i.e., the encoder network uses remote sensing band information and dataset labels of a sparse atmospheric and ocean color data set as training data for independent training, the decoder network is trained using the output of the trained encoder network and the labels of the sparse atmospheric and ocean color data set as training data, and the pre-trained discrete encoder-decoder network is obtained by connecting the trained encoder and decoder in series; a network including n layers of convolutional layers and a normalized residual network and a layer of convolutional layers in series is used to follow the discrete encoder-decoder network to obtain pending reconstruction features.

5. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 4, characterized in that: The specific implementation process of the external depth estimation network is as follows: The external depth estimation network uses the same discrete encoder-decoder network as the sparse processing network for land surface information, and uses a non-discrete encoder-decoder with the same architecture as the sparse processing network for observation geometry without using a discrete training method, wherein the query and key of ViT in the non-discrete encoder-decoder come from the sparse processing network; The output of the land surface information through the discrete encoder-decoder network and the output of the observed geometry through the non-discrete encoder-decoder network are concatenated on the channel, and the joint coding guidance information is obtained after the channel dimension reduction through one convolutional layer after the n-layer joint external information encoding module.

6. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 5, characterized in that: The joint external information encoding module is a self-attention module, which is specifically implemented as follows: the input passes through a convolution layer and normalization, and passes through a convolution layer with itself as the query, key and value respectively. The query and key values ​​are multiplied and then weighted by Softmax to obtain the attention weight, which is multiplied with the value passed through the convolution layer to obtain the output. This output is residually connected with the input passed through a convolution layer to obtain the output of the self-attention module of this layer.

7. The method for joint inversion of atmosphere and ocean color parameters based on sparse remote sensing satellite data according to claim 6, characterized in that: The specific implementation process of the scale fusion network is as follows: The scale fusion network scale-fuses the pending reconstruction features obtained by the sparse processing network and the joint coding guidance information obtained by the external depth estimation network to obtain a final inversion result image. The scale fusion network includes several layers of semantic prior fusion SPF and guidance reconstruction layers. The first guidance reconstruction layer receives the pending reconstruction features, the nth guidance reconstruction layer receives the output of the n-1th SPF layer, the first SPF layer receives the joint coding guidance information and the output of the first guidance reconstruction layer, the nth SPF layer receives the output of the n-1th SPF layer and the output of the nth guidance reconstruction layer, and the last guidance reconstruction layer receives the output of the last SPF layer to obtain the final atmospheric correction or ocean color component data; the SPF uses the same self-attention module as in the external depth estimation network, and the guidance reconstruction layer is a U-Net network.

Images