High-precision retrieval method of sea surface chlorophyll concentration by using cross-attention mechanism and fusion of hyperspectral and optical data

Abstract

The application proposes a high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and space spectrum fusion, aiming at solving the trade-off problem between spatial resolution and temporal resolution of sea surface chlorophyll concentration in satellite remote sensing. The method can fully mine and effectively fuse spatial and spectral information of different resolutions by introducing cross-attention mechanism through space spectrum fusion technology, so as to realize the production of high spatial resolution and high temporal resolution sea surface reflectance data; and uses random forest fitting inversion method to improve the precision and robustness of chlorophyll concentration inversion. These methods overcome the application limitations of traditional sea surface chlorophyll concentration production methods in complex marine environments. The application can significantly improve the spatial resolution and temporal resolution of sea surface chlorophyll concentration, and has strong adaptability and operability.

Description

Technical Field

[0001] This invention belongs to the field of ocean color remote sensing, and in particular relates to a high-precision method for inverting sea surface chlorophyll concentration by applying cross-attention mechanism and spatial-spectral fusion. Background Technology

[0002] Due to the physical and technological limitations of sensors, a trade-off between spatial resolution, temporal resolution, and spectral resolution is essential in satellite remote sensing. Increasing spatial resolution typically means the sensor needs to capture more detailed information, which reduces its coverage area and increases the time interval required to cover the same area twice, resulting in reduced temporal resolution. Furthermore, for sea surface reflectance data, obtaining high-resolution data across multiple spectral bands simultaneously requires complex sensing technologies and significant data processing capabilities, further limiting the possibility of maintaining high spatial or temporal resolution while simultaneously maintaining high spectral resolution. Therefore, depending on different application requirements and mission characteristics, specific technical and design choices must be made among these three factors, making it impossible to simultaneously achieve high spatial, temporal, and spectral resolution. High-precision sea surface color product inversion often relies on high-spectral-resolution sea surface reflectance data, and its spatial and temporal resolution are also limited by the sea surface reflectance data, resulting in a scarcity of high-quality sea surface color product images. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a high-precision method for retrieving sea surface chlorophyll concentration by applying a cross-attention mechanism and spatial-spectral fusion. The specific technical solution is as follows:

[0004] A high-precision method for retrieving sea surface chlorophyll concentration using cross-attention mechanism and spatial-spectral fusion includes the following steps:

[0005] Step 1: To obtain the spatial resolution SR of the location p2 to be predicted at the time t3 to be predicted. H The time resolution is TR H Based on the chlorophyll concentration data, a spatial resolution SR was constructed. H The time resolution is TR L The multi-band remote sensing image dataset is X1, and it has a spatial resolution of SR. L The time resolution is TR H The multi-band remote sensing image dataset is X2; where the spatial resolution is SR H >SR L Time resolution TR L <TR H ;

[0006] Select multi-band remote sensing image X11 from X1 that is adjacent to or contains the location p2 to be predicted, and date t1 of the location p2 to be predicted.

[0007] Select multi-band remote sensing image X10 from X1 that is adjacent to or contains the location p2 to be predicted, and is close to t1 at time t0.

[0008] Select multi-band remote sensing image X21 from X2 that is adjacent to or contains the location p1 to be predicted location p2 on date t1 for training the spatial-spectral fusion model;

[0009] Select multi-band remote sensing image X20 from X2 that is adjacent to or contains the location p1 of the location to be predicted p2 and is close to t1 at time t0.

[0010] Select the multi-band remote sensing image X23 located at the location p2 to be predicted at the time t3 to be predicted from X2;

[0011] Select the multi-band remote sensing image X22 located at the location p2 to be predicted from X2, which is close to the time t3 to be predicted.

[0012] Select multi-band remote sensing image X12 located at location p2 of the target location at t2, which is close to the target time t3;

[0013] Step 2: Take the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21 and X20, and create a training dataset for the spatial-spectral fusion model; construct a spatial-spectral fusion model based on the cross-attention mechanism and complete the training of the spatial-spectral fusion model;

[0014] Step 3: Obtain sea surface reflectance data images corresponding to multi-band remote sensing images X12, X22, and X23, create a prediction dataset, input it into the spatial-spectral fusion model trained in Step 2, and predict the spatial resolution at time t3 as SR. H The time resolution is TR H X3 multi-band remote sensing images;

[0015] Step 4: Select several sea surface reflectance data images from the multi-band remote sensing image dataset X1, and the known chlorophyll concentrations in the same time and region as the selected sea surface reflectance data images, and perform latitude and longitude matching to obtain the training dataset for the chlorophyll concentration inversion model.

[0016] Step 5: Construct a chlorophyll concentration inversion model based on random forest fitting. Use the training dataset of the chlorophyll concentration inversion model in Step 4 to train the chlorophyll concentration inversion model and obtain the trained chlorophyll concentration inversion model.

[0017] Step Six: Set the spatial resolution of time t3 obtained in Step Three to SR. H The time resolution is TR H In the multi-band remote sensing image X3 input step five, the spatial resolution of the specified region p2 at time t3 in the chlorophyll concentration inversion model is SR. H The time resolution is TR H chlorophyll concentration.

[0018] Furthermore, the multi-band remote sensing images X11, X10, X21, X20 and X12, X22, X23 all require the selection of blue, green, red, and near-infrared bands, and the multi-band remote sensing images X11, X10, X12 and X21, X23 also require the selection of blue, green, red, and near-infrared bands. 0、 X22 and X23 have wavelength ranges that are as close as possible in their corresponding bands.

[0019] Furthermore, the location p2 to be predicted is included in location p1.

[0020] Furthermore, t0, t1, t2, and t3 are close; at the same time, the time interval between t0 and t1 is the same as the time interval between t2 and t3.

[0021] Furthermore, step two includes the following sub-steps:

[0022] S2.1: For the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21 and X20, spatial cutting and land masking are performed on all locations p1 to remove all land areas; then missing values ​​and outliers are filled by interpolation; then the filled X21 and X20 are upsampled to the same spatial resolution as the multi-band remote sensing images X11 and X10;

[0023] S2.2: Cut the four sea surface reflectance data images with the same spatial resolution obtained after processing in step S2.1 into three-dimensional image blocks of the same size and number. Use the sea surface reflectance data images corresponding to X10, X20 and X21 as the input of the spatial-spectral fusion model, and use the sea surface reflectance data image corresponding to X11 as the output of the spatial-spectral fusion model to create a training dataset.

[0024] S2.3: Construct a spatial-spectral fusion model based on a convolutional neural network, implement a cross-attention mechanism, and train the spatial-spectral fusion model using the training dataset from S2.2 to obtain the trained spatial-spectral fusion model.

[0025] Furthermore, step three includes the following sub-steps:

[0026] S3.1: Spatially cut and land mask all land areas of the sea surface reflectance data images corresponding to X12, X22 and X23 at the predicted location p2; then fill in missing and outlier values ​​by interpolation; then upsample the sea surface reflectance data images corresponding to X22 and X23 to the same spatial resolution as the multi-band remote sensing image X12.

[0027] S3.2: Cut the three sea surface reflectance data images with the same spatial resolution obtained after processing in S3.1 into three-dimensional image patches of the same size as the training dataset for the spatial-spectral fusion model made in step two. Use the sea surface reflectance data images corresponding to X12, X22 and X23 as the input of the spatial-spectral fusion model trained in step two to create a prediction dataset.

[0028] S3.3: Input the prediction dataset created in S3.2 into the spatial-spectral fusion model trained in step two to obtain the spatial resolution of prediction time t3 as SR. H The time resolution is TR H X3 multi-band remote sensing image.

[0029] Furthermore, in step four, when performing latitude and longitude matching, if the known chlorophyll concentration in the same time and area is a grid image, and the spatial resolution of the grid image is different from the spatial resolution of the multi-band remote sensing image dataset X1, it is necessary to downsample the image with the higher spatial resolution to ensure that the latitude and longitude matching is performed at the same spatial resolution.

[0030] The beneficial effects of this invention are as follows:

[0031] The high-precision sea surface chlorophyll concentration inversion method provided by this invention applies cross-attention mechanism and spatial-spectral fusion. It uses machine learning for fitting, which effectively overcomes the nonlinear relationship problem encountered by traditional methods when dealing with the spatiotemporal variation of chlorophyll concentration in complex water bodies. It learns complex nonlinear patterns from a large amount of data and establishes a more accurate prediction model, thereby improving the accuracy and robustness of chlorophyll concentration prediction. The image generated by this method can have higher spatial and temporal resolution at the same time, which can make up for the limitations of a single sensor. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the overall process of a high-precision sea surface chlorophyll concentration inversion method that utilizes cross-attention mechanism and spatial-spectral fusion, as described in an embodiment of the present invention.

[0033] Figure 2 This is a flowchart of the spatial spectrum fusion part of an embodiment of the present invention.

[0034] Figure 3 This is a flowchart of the inversion fitting part of an embodiment of the present invention.

[0035] Figure 4 Figure 1 shows the chlorophyll concentration after applying the present invention and the comparison method in this embodiment. Figure 2 shows the chlorophyll concentration inversion result based on NN with 10-meter resolution after applying the method of the present invention; Figure 3 shows the chlorophyll concentration inversion result based on NN with 300-meter resolution in the WFR dataset; Figure 4 shows the chlorophyll concentration inversion result based on OC4M with 10-meter resolution after applying the method of the present invention; Figure 5 shows the chlorophyll concentration inversion result based on OC4ME with 300-meter resolution in the WFR dataset. Detailed Implementation

[0036] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0037] Explanation of technical terms:

[0038] The 6S model (Second Simulation of the Satellite Sigal in the Solar Spectrum) is a widely used atmospheric radiative transfer model used to simulate the propagation of solar radiation in the atmosphere, especially in satellite remote sensing applications.

[0039] like Figure 1 As shown, the high-precision sea surface chlorophyll concentration inversion method of the present invention, which applies cross-attention mechanism and spatial-spectral fusion, includes the following steps:

[0040] Step 1: To obtain the spatial resolution SR of the location p2 to be predicted at the time t3 to be predicted. H The time resolution is TR H Based on the chlorophyll concentration data, a spatial resolution SR was constructed. H The time resolution is TR L The multi-band remote sensing image dataset is X1, and it has a spatial resolution of SR. L The time resolution is TR H The multi-band remote sensing image dataset is X2; where the spatial resolution is SR H >SR L Time resolution TR L <TR H .

[0041] Select multi-band remote sensing image X11 from X1 that is adjacent to or contains the location p2 to be predicted, and date t1 of the location p2 to be predicted.

[0042] Select multi-band remote sensing image X10 from X1 that is adjacent to or contains the location p2 to be predicted, and is close to t1 at time t0.

[0043] Select multi-band remote sensing image X21 from X2 that is adjacent to or contains the location p1 to be predicted location p2 on date t1 for training the spatial-spectral fusion model;

[0044] Select multi-band remote sensing image X20 from X2 that is adjacent to or contains the location p1 of the location to be predicted p2 and is close to t1 at time t0.

[0045] Select the multi-band remote sensing image X23 located at the location p2 to be predicted at the time t3 to be predicted from X2;

[0046] Select the multi-band remote sensing image X22 located at the location p2 to be predicted from X2, which is close to the time t3 to be predicted.

[0047] Select multi-band remote sensing image X12 from X1, which is close to the predicted time t3 and located at the predicted location p2.

[0048] The multi-band remote sensing images X11, X10, X21, X20 and X12, X22, X23 here all require the selection of blue, green, red, and near-infrared bands. Furthermore, the multi-band remote sensing images X11, X10, X12 and X21, X23... 0、 X22 and X23 have wavelength ranges that are as close as possible in their corresponding bands.

[0049] The inclusion relationship between locations p1 and p2 is preferably such that the location p2 to be predicted is included in the known location p1.

[0050] Because chlorophyll concentration varies considerably at different times, t0, t1, t2, and t3 should be as close as possible; at the same time, the time interval between t0 and t1 should be as similar as the time interval between t2 and t3.

[0051] In this embodiment, the selected reference date t0 is December 3rd, the adjacent prediction date t1 is December 8th, the designated sea area p1 is the South China Sea with latitude from 20.807°N to 21.363° and longitude from 109.241°E to 109.651°E, the multi-band remote sensing images X11 and X10 are 10-meter spatial resolution multi-band Sentinel-2L1C image pairs for the designated sea area on December 8th and December 3rd, respectively, and the multi-band remote sensing images X21 and X20 are Sentinel-3 300-meter resolution multi-band image pairs for the designated sea area on December 8th and December 3rd, respectively, from OLCI WFR. The Sentinel-2 corresponding file processing is performed on the GEE platform, and the Sentinel-3 data files are: S3A_OL_2_WFR____20201203T025356_20201203T025656_20201204T104820_0179_065_360_2520_MAR_O_NT_002.SEN3 and S3A_OL_2_WFR____20201208T022401_20201208T022701_20201209T130731_0179_066_046_2520_MAR_O_NT_002.SEN3.

[0052] In this embodiment, the selected bands for the Sentinel-2 sea surface reflectance data image are: B2, B3, B4, B5, B6, and B8; the selected bands for the Sentinel-3 surface reflectance data in the WFR dataset are: Oa04, Oa06, Oa08, Oa11, Oa12, and Oa17.

[0053] In this embodiment, to verify the model output, t2 = t0, t3 = t1, p2 is a region contained within p1, and X12 and X2 are selected. 2、 X23 can be derived from X10 and X2 respectively. 0、 Extracted from X21 remote sensing imagery.

[0054] Step 2: Take the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21, and X20 of t1 and t0, and create a training dataset for the spatial-spectral fusion model; construct a spatial-spectral fusion model based on the cross-attention mechanism and complete the training of the spatial-spectral fusion model. For example... Figure 2 As shown, step two specifically includes the following sub-steps:

[0055] S2.1: For the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21 and X20 of t1 and t0, spatial cutting and land masking are performed on location p1 to remove all land areas; then missing values ​​and outliers are filled by interpolation; then the filled X21 and X20 are upsampled to the same spatial resolution as the multi-band remote sensing images X11 and X10.

[0056] S2.2: Cut the four sea surface reflectance data images with the same spatial resolution obtained after processing in step S2.1 into three-dimensional image blocks of the same size and number. Use the sea surface reflectance data images corresponding to X10, X20 and X21 as the input of the spatial-spectral fusion model, and use the sea surface reflectance data image corresponding to X11 as the output of the spatial-spectral fusion model to create a training dataset.

[0057] S2.3: Construct a spatial-spectral fusion model based on a convolutional neural network, implement a cross-attention mechanism, and train the spatial-spectral fusion model using the training dataset in S2.2 to obtain the trained spatial-spectral fusion model. At this time, the spatial-spectral fusion model can use the convolution method to simultaneously examine the temporal changes of sea surface reflectance in a specified area on adjacent dates and the channel differences between remote sensing images from different sensors in the specified area and time.

[0058] The image fusion model based on convolutional neural networks uses the following method to examine the temporal differences of the same region at different times:

[0059] F SPM =(W a *F H +b a )⊙σ(W b *F L +b b )

[0060] The following methods were also used to examine the differences between different sensor channels:

[0061]

[0062] Among them, F SPM F SRM F represents the feature maps output by the SPM (Spatial Perception Module) and SRM (Spectral Reconstruction Module), respectively. H High spatial resolution imagery (SR) H Corresponding feature map, F L Represents low spatial resolution imagery (i.e., SR). L Corresponding feature map f represents the feature map obtained by the SRM module for processing the i-th band. concat For cascading operations, ⊙ represents element-wise multiplication, σ represents the Sigmoid function, W represents the weight parameter, and b represents the bias parameter.

[0063] In this embodiment, because the multi-band Sentinel-2 L1C imagery is used, it is necessary to invert the multi-band Sentinel-2 L1C imagery into sea surface reflectance data imagery using the 6S atmospheric radiative transfer model. The OLCI WFR Sentinel-3 300-meter resolution multi-band imagery itself is sea surface reflectance data imagery and has already undergone land masking. Therefore, the OLCI WFR Sentinel-3 300-meter resolution multi-band imagery is directly spatially segmented, and missing and outlier values ​​are interpolated and filled. The Sentinel-3 300-meter resolution multi-band imagery is then upsampled to obtain a 10-meter resolution sea surface reflectance data imagery. In this embodiment, the processing of the Sentinel-2 imagery is performed on the GEE platform.

[0064] Step 3: Obtain sea surface reflectance data images corresponding to multi-band remote sensing images X12, X22, and X23, create a prediction dataset, input it into the spatial-spectral fusion model trained in Step 2, and predict the spatial resolution at time t3 as SR. H The time resolution is TR H Multiband remote sensing image X3. For example... Figure 2 As shown, step three specifically includes the following sub-steps:

[0065] S3.1: Spatially cut and land mask all land areas of the sea surface reflectance data images corresponding to X12, X22 and X23 at the predicted location p2; then fill in missing and outlier values ​​by interpolation; then upsample the sea surface reflectance data images corresponding to X22 and X23 to the same spatial resolution as the multi-band remote sensing image X12.

[0066] S3.2: Cut the three sea surface reflectance data images with the same spatial resolution obtained after processing in S3.1 into three-dimensional image patches of the same size as the training dataset for the spatial-spectral fusion model made in step two. Use the sea surface reflectance data images corresponding to X12, X22 and X23 as the input of the spatial-spectral fusion model trained in step two to create a prediction dataset.

[0067] S3.3: Input the prediction dataset created in S3.2 into the spatial-spectral fusion model trained in step two to obtain the multi-band sea surface reflectance data image X3 at prediction time t3. At this time, X3 possesses the same high spatial resolution SR as the X1 dataset. HFurthermore, since X23 can be predicted as long as it is available, X3 also has the same high temporal resolution TR as X2. H .

[0068] Step 4: Select several sea surface reflectance data images from the multi-band remote sensing image dataset X1, and the known chlorophyll concentrations in the same time and region as the selected sea surface reflectance data images, and perform latitude and longitude matching to obtain the training dataset for the chlorophyll concentration inversion model.

[0069] The selected sea surface reflectance data images from the multi-band remote sensing image dataset X1 should have the same band selection as X10, X11 and X12, and the temporal and spatial selection should also be as close as possible.

[0070] When the known chlorophyll concentration in the same time and region is a grid image, and the spatial resolution of the grid image is different from that of the multi-band remote sensing image dataset X1, it is necessary to downsample the image with the higher spatial resolution to ensure that the latitude and longitude are matched at the same spatial resolution.

[0071] In this embodiment, the selected sea surface reflectance data images from the multi-band remote sensing image dataset X1 are X10 and X11. The known chlorophyll concentrations for the same time and region are from the Sentinel-3 WFR dataset. Since the known chlorophyll concentrations are from Sentinel-3, which has a spatial resolution of 300m, while the spatial resolution of the Sentinel-2 sea surface reflectance data is 10m, the 10m spatial resolution Sentinel-2 sea surface reflectance data is first downsampled to a 300m resolution. Then, based on time and latitude / longitude, it is matched with the chlorophyll concentrations inverted from the WFR using the OC4ME and NN methods to obtain two inversion fitting datasets.

[0072] Step 5: Construct a chlorophyll concentration inversion model based on random forest fitting. Use the training dataset of the chlorophyll concentration inversion model in Step 4 to train the chlorophyll concentration inversion model and obtain the trained chlorophyll concentration inversion model.

[0073] In this embodiment, chlorophyll concentration inversion models were trained using two inversion fitting datasets, resulting in two chlorophyll concentration inversion models.

[0074] Step Six: Set the spatial resolution of time t3 obtained in Step Three to SR. H The time resolution is TR H In the multi-band remote sensing image X3 input step five, the spatial resolution of the specified region p2 at time t3 in the chlorophyll concentration inversion model is SR.H The time resolution is TR H chlorophyll concentration.

[0075] The inversion fitting process in steps four, five, and six is ​​as follows: Figure 3 As shown.

[0076] The results obtained in this embodiment are as follows: Figure 4 As shown, (a) represents the chlorophyll concentration retrieved at a spatial resolution of 10 meters based on NN after applying the method of the present invention, (b) represents the chlorophyll concentration at a spatial resolution of 300 meters based on NN in the WFR dataset, (c) represents the chlorophyll concentration retrieved at a spatial resolution of 10 meters based on OC4M after applying the method of the present invention, and (d) represents the chlorophyll concentration retrieved at a spatial resolution of 300 meters based on OC4ME in the WFR dataset.

[0077] By comparing (a) and (b), and (c) and (d), it can be seen that the chlorophyll concentration is highly consistent, and the image texture details of the chlorophyll concentration obtained by the method of this invention are richer. By comparing (c) and (d), the missing data areas in Figure (d) are filled in Figure (c).

[0078] In the method of the present invention, as long as there is available sea surface reflectance data in the set area for remote sensing image data X2 with high temporal resolution on a date close to the prediction date, the high spatial resolution sea surface chlorophyll concentration of this set area on the prediction date can be retrieved. Therefore, the sea surface chlorophyll concentration obtained by the method of the present invention has both the same spatial resolution as the X1 dataset and the same temporal resolution as the X2 dataset, thereby generating grid data of sea surface chlorophyll concentration with both high spatial resolution and high temporal resolution.

[0079] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A high-precision method for retrieving sea surface chlorophyll concentration using cross-attention mechanism and spatial-spectral fusion, characterized in that, Includes the following steps: Step 1: To obtain the spatial resolution SR of the location p2 to be predicted at the time t3 to be predicted. H The time resolution is TR H Based on the chlorophyll concentration data, a spatial resolution SR was constructed. H The time resolution is TR L The multi-band remote sensing image dataset is X1, and it has a spatial resolution of SR. L The time resolution is TR H The multi-band remote sensing image dataset is X2; where the spatial resolution is SR H >SR L Time resolution (TR) L <TR H ; Select multi-band remote sensing image X11 from X1 that is adjacent to or contains the location p2 to be predicted, and date t1 of the location p2 to be predicted. Select multi-band remote sensing image X10 from X1 that is adjacent to or contains the location p2 to be predicted, and is close to t1 at time t0. Select multi-band remote sensing image X21 from X2 that is adjacent to or contains the location p1 to be predicted location p2 on date t1 for training the spatial-spectral fusion model; Select multi-band remote sensing image X20 from X2 that is adjacent to or contains the location p1 of the location to be predicted p2 and is close to t1 at time t0. Select the multi-band remote sensing image X23 located at the location p2 to be predicted at the time t3 to be predicted from X2; Select the multi-band remote sensing image X22 located at the location p2 to be predicted from X2, which is close to the time t3 to be predicted. Select multi-band remote sensing image X12 located at location p2 of the target location at t2, which is close to the target time t3; Step 2: Take the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21 and X20, and create a training dataset for the spatial-spectral fusion model; construct a spatial-spectral fusion model based on the cross-attention mechanism and complete the training of the spatial-spectral fusion model; at this time, the spatial-spectral fusion model can use the convolution method to simultaneously examine the temporal changes of sea surface reflectance in a specified area on adjacent dates and the channel differences between the specified area and time from remote sensing images from different sensors; Step 3: Obtain sea surface reflectance data images corresponding to multi-band remote sensing images X12, X22, and X23, create a prediction dataset, input it into the spatial-spectral fusion model trained in Step 2, and predict the spatial resolution at time t3 as SR. H The time resolution is TR H X3 multi-band remote sensing images; Step 4: Select several sea surface reflectance data images from the multi-band remote sensing image dataset X1, and the known chlorophyll concentrations in the same time and region as the selected sea surface reflectance data images, and perform latitude and longitude matching to obtain the training dataset for the chlorophyll concentration inversion model. Step 5: Construct a chlorophyll concentration inversion model based on random forest fitting. Use the training dataset of the chlorophyll concentration inversion model in Step 4 to train the chlorophyll concentration inversion model and obtain the trained chlorophyll concentration inversion model. Step Six: Set the spatial resolution of time t3 obtained in Step Three to SR. H The time resolution is TR H In the multi-band remote sensing image X3 input step five, the spatial resolution of the specified region p2 at time t3 in the chlorophyll concentration inversion model is SR. H The time resolution is TR H chlorophyll concentration.

2. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 1, characterized in that, The multi-band remote sensing images X11, X10, X21, X20 and X12, X22, X23 all require the selection of blue, green, red, and near-infrared bands. Furthermore, the multi-band remote sensing images X11, X10, X12 and X21, X23... 0、 X22 and X23 have wavelength ranges that are as close as possible in their corresponding bands.

3. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 1, characterized in that, The location p2 to be predicted is included in location p1.

4. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 1, characterized in that, t0, t1, t2, and t3 are close; at the same time, the time interval between t0 and t1 is the same as the time interval between t2 and t3.

5. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 1, characterized in that, Step two includes the following sub-steps: S2.1: For the sea surface reflectance data images corresponding to the multi-band remote sensing images X11, X10, X21 and X20, spatial cutting and land masking are performed on all locations p1 to remove all land areas; then missing values ​​and outliers are filled by interpolation; then the filled X21 and X20 are upsampled to the same spatial resolution as the multi-band remote sensing images X11 and X10. S2.2: Cut the four sea surface reflectance data images with the same spatial resolution obtained after processing in step S2.1 into three-dimensional image blocks of the same size and number. Use the sea surface reflectance data images corresponding to X10, X20 and X21 as the input of the spatial-spectral fusion model, and use the sea surface reflectance data image corresponding to X11 as the output of the spatial-spectral fusion model to create a training dataset. S2.3: Construct a spatial-spectral fusion model based on a convolutional neural network, implement a cross-attention mechanism, and train the spatial-spectral fusion model using the training dataset from S2.2 to obtain the trained spatial-spectral fusion model.

6. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 5, is characterized in that, Step three includes the following sub-steps: S3.1: Spatially cut and land mask all land areas of the sea surface reflectance data images corresponding to X12, X22 and X23 at the predicted location p2; then fill in missing and outlier values ​​by interpolation; then upsample the sea surface reflectance data images corresponding to X22 and X23 to the same spatial resolution as the multi-band remote sensing image X12. S3.2: Cut the three sea surface reflectance data images with the same spatial resolution obtained after processing in S3.1 into three-dimensional image patches of the same size as the training dataset for the spatial-spectral fusion model made in step two. Use the sea surface reflectance data images corresponding to X12, X22 and X23 as the input of the spatial-spectral fusion model trained in step two to create a prediction dataset. S3.3: Input the prediction dataset created in S3.2 into the spatial-spectral fusion model trained in step two to obtain the spatial resolution of prediction time t3 as SR. H The time resolution is TR H X3 multi-band remote sensing image.

7. The high-precision sea surface chlorophyll concentration inversion method applying cross-attention mechanism and spatial-spectral fusion as described in claim 5, is characterized in that, When performing latitude and longitude matching in step four, if the known chlorophyll concentration in the same time and area is a grid image, and the spatial resolution of the grid image is different from the spatial resolution of the multi-band remote sensing image dataset X1, it is necessary to downsample the image with the higher spatial resolution to ensure that the latitude and longitude matching is performed at the same spatial resolution.

Images