Passive microwave and SAR (Synthetic Aperture Radar) data fused sea ice concentration inversion method and device

Abstract

The invention discloses a passive microwave and SAR data fused sea ice concentration inversion method and device, and relates to the field of remote sensing image processing, and the method comprises the steps: carrying out the geographic projection and resampling of preprocessed SAR data; performing calculation to obtain sea ice concentration inversion parameters, aligning the re-sampled AMSR2 data with the re-sampled SAR data, and cutting an image overlapping region to obtain an image pair; training a Pix2PixGAN network, and inputting a part of sea ice concentration inversion parameters to the Pix2PixGAN network to obtain an ice water classification result; and removing a seawater region in the remaining sea ice concentration inversion parameters according to the obtained ice water classification result, inputting the remaining sea ice concentration inversion parameters after the removal operation into the Pix2PixGAN network, and calculating an output result by adopting an ASI algorithm to obtain the sea ice concentration. According to the invention, high-precision sea ice concentration inversion can be realized.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing, specifically to a method and apparatus for inverting sea ice concentration by fusing passive microwave and SAR data. Background Technology

[0002] Polar sea ice is a crucial component of the Earth's climate system, directly participating in the regulation of global energy balance, ocean circulation, and climate feedback mechanisms. The presence of sea ice significantly increases surface albedo, reflecting most shortwave radiation and limiting heat exchange and transfer between the ocean and atmosphere, thus helping to maintain low temperatures in polar regions and having a profound impact on the polar and global climate systems. As one of the most critical indicators describing sea ice distribution, sea ice concentration (SIC) is defined as the percentage of sea ice area within a unit of ocean area, usually expressed as a percentage. Sea ice concentration can be used to monitor seasonal variations, multi-year trends, marginal sea ice movement, and open water formation, and is of great significance for assessing polar ocean navigation capabilities, resource exploration, and climate model input. With global warming, Arctic sea ice is showing a continuous decreasing trend, and sea ice growth and melting are changing rapidly, especially in summer. Against this backdrop, the scientific and engineering communities have an increasingly urgent need for high-precision, long-term, and high-resolution sea ice concentration data. Therefore, Arctic sea ice monitoring has significant scientific and economic value.

[0003] In sea ice classification, high-resolution SAR (Synthetic Aperture Radar) data is used to obtain detailed sea ice distribution information. Combined with traditional sea ice concentration products, richer multi-scale ice condition maps can be constructed, providing support for polar sea ice monitoring and route optimization. With the rise of machine learning, many scholars have used SAR image texture features and machine learning methods to classify sea ice and seawater, such as Support Vector Machine (SVM), Iterative Region Growing using Semantics (IRGS), Conditional Random Field (CRF), Random Forest (RF), Bayesian algorithms, and Principal Component Analysis (PCA). However, their performance largely depends on the selection of texture features. Among them, Support Vector Machine (SVM) combined with texture feature analysis uses gray-level co-occurrence matrix to extract feature values, improving the accuracy and reliability of classification; Bayesian algorithm and neural network combined use Bayesian sea ice detection algorithm and backpropagation (BP) neural network to improve the accuracy and efficiency of classification; Principal Component Analysis (PCA) combines spectral-spatial-joint features and uses gray-level co-occurrence matrix for feature extraction, enhancing the depth and accuracy of image classification. Compared with traditional machine learning algorithms, deep learning algorithms have higher sea ice classification accuracy and perform well in sea ice edge areas. Currently, there are three main deep learning models applied to SAR image sea ice concentration inversion: Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Fully Convolutional Network (FCN).

[0004] Current research on sea ice concentration inversion mainly includes passive microwave concentration inversion algorithms, optical data inversion algorithms, and multi-source data fusion methods. These include single-frequency inversion methods based on polarization ratio, such as the Bootstrap method which uses polarization ratio information of vertical and horizontal brightness temperatures at 19 GHz or 37 GHz for concentration inversion; single-polarization methods based on frequency gradients, which describe the distribution of sea ice and seawater within a resolution cell by the electromagnetic scattering differences of microwaves at different frequencies, such as the calibration-validation (Cal / Val) and NORSEX algorithms; concentration inversion algorithms that combine polarization ratio and frequency gradient, such as the ASI algorithm and the NASA Team (NT) algorithm; and sea ice concentration inversion algorithms based on multi-source data fusion, such as those based on neural network models. These models use passive microwave polarization ratio, SAR backscattering, and other features to build a neural network model, construct hidden layer units based on feature fusion, iteratively update the model through error backpropagation, and obtain the time-series concentration inversion results using the maximum likelihood constraint criterion. In addition, there are some commonly used inversion algorithms, as well as some hybrid algorithms that use multiple of the above algorithms simultaneously.

[0005] However, current sea ice concentration inversion algorithms are insufficient in terms of accuracy. Therefore, how to achieve high-precision sea ice concentration inversion has become an urgent problem to be solved. Summary of the Invention

[0006] This application provides a method and apparatus for sea ice concentration inversion by fusing passive microwave and SAR data, which can achieve high-precision sea ice concentration inversion.

[0007] In a first aspect, embodiments of this application provide a method for retrieving sea ice concentration by fusing passive microwave and SAR data, the method comprising: Acquire Sentinel-1 GRD data to obtain SAR data and preprocess the SAR data, then perform geographic projection and resampling on the preprocessed SAR data; The sea ice concentration inversion parameters were calculated, the AMSR2 data were resampled, and the resampled AMSR2 data and the resampled SAR data were aligned. The overlapping areas of the images were cropped to obtain image pairs. The Pix2PixGAN network is trained based on the obtained image pairs. Some sea ice concentration inversion parameters are input into the Pix2PixGAN network to obtain ice and water classification results. Based on the obtained ice-water classification results, the seawater region is removed from the remaining sea ice concentration inversion parameters. The remaining sea ice concentration inversion parameters after the removal operation are then input into the Pix2PixGAN network, and the ASI algorithm is used to calculate the output results to obtain the sea ice concentration.

[0008] In conjunction with the first aspect, in one embodiment, the preprocessing includes orbit correction, thermal noise removal, radiometric calibration, and speckle filtering.

[0009] In conjunction with the first aspect, in one implementation, the geographic projection and resampling of the preprocessed SAR data specifically includes: The GRD data was projected onto the NSIDC Sea Ice PolarStereographic North projection with EPSG code 3411 using the SNAP platform. Resample SAR data from its original resolution to a preset distance resolution.

[0010] In conjunction with the first aspect, in one implementation, the resampling of AMSR2 data and alignment of the resampling AMSR2 data and resampling SAR data, followed by cropping the overlapping image region to obtain an image pair, specifically includes: The acquired AMSR2 data is resampled from the original resolution to a preset distance resolution using a bicubic convolution method. The resampled AMSR2 data and resampled SAR data are aligned, and the overlapping areas of the AMSR2 and SAR data on the same date are cropped and reassembled into an image pair.

[0011] In conjunction with the first aspect, in one implementation, the calculation of sea ice concentration inversion parameters specifically includes:

[0012]

[0013]

[0014] in, , , Indicates the parameters for sea ice concentration inversion. This represents the brightness temperature value for vertical polarization at 18.7 GHz. This represents the brightness temperature value for vertical polarization at 23.8 GHz. This represents the brightness temperature value for vertical polarization at 36.5 GHz. This represents the brightness temperature value for vertical polarization at 89.0 GHz. This represents the brightness temperature value for horizontal polarization at 89.0 GHz.

[0015] In conjunction with the first aspect, in one implementation, the step of training a Pix2PixGAN network based on the obtained image pairs, inputting some sea ice concentration inversion parameters into the Pix2PixGAN network, and obtaining ice-water classification results specifically includes: This involves removing data from the resampled SAR data that contains only seawater and no sea ice, thus enabling a masking operation on the SAR data. Obtain the sea ice concentration product with the preset distance resolution of the official AMSR2+MODIS, use the same land mask as the obtained sea ice concentration product, remove the land portion in the resampled AMSR2 data, and realize the masking operation on the AMSR2 data. The maximum and minimum values ​​of SAR data and AMSR2 data after the masking operation are statistically analyzed, and the statistically obtained values ​​are used as parameters to input into the Pix2PixGAN network. The number of rows and columns of the obtained image pair is expanded, and the SAR data and AMSR2 data in the image pair are stitched together left and right to obtain an image pair with the same number of rows but twice the number of columns as the original image, thus realizing the construction of the image pair required for input to the Pix2PixGAN network. The Pix2PixGAN network was trained based on the statistically obtained values ​​and the constructed image pairs. Will and Input the Pix2PixGAN network and output the original image and the fused image; The difference between the original image and the fused image is calculated to obtain a difference map; Based on the obtained difference map, the ice-water classification results are obtained.

[0016] In conjunction with the first aspect, in one implementation method, Training the Pix2PixGAN network also includes normalizing the training data. The normalization process specifically involves:

[0017] in, This represents the normalized value. Represents the original value of a point. This represents the minimum value. This represents the maximum value.

[0018] In conjunction with the first aspect, in one implementation, the step of removing the seawater region from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results, inputting the remaining sea ice concentration inversion parameters after the removal operation into the Pix2PixGAN network, and using the ASI algorithm to calculate the output results to obtain the sea ice concentration specifically includes: Based on the obtained ice-water classification results, remove The seawater area in the middle will be the area after removing the seawater area. Input the Pix2PixGAN network and obtain the output results; The ASI algorithm is used to calculate the sea ice concentration from the output results.

[0019] In conjunction with the first aspect, in one implementation, the ASI algorithm is used to calculate the sea ice concentration, and the specific calculation method is as follows:

[0020] in, Indicates sea ice concentration. express polarization difference, , , , This represents the coefficient, and the coefficient is determined as follows:

[0021] in, , This represents the value of the coordinate point.

[0022] Secondly, embodiments of this application provide a sea ice concentration inversion device for passive microwave and SAR data fusion, the sea ice concentration inversion device for passive microwave and SAR data fusion includes: The acquisition module is used to acquire Sentinel-1 GRD data to obtain SAR data and preprocess the SAR data, and perform geographic projection and resampling on the preprocessed SAR data. The cropping module is used to calculate the sea ice concentration inversion parameters, resample AMSR2 data, align the resampled AMSR2 data with the resampled SAR data, and crop out the overlapping image areas to obtain image pairs. The adjustment module is used to train the Pix2PixGAN network based on the obtained image pairs, inputting some sea ice concentration inversion parameters into the Pix2PixGAN network to obtain ice and water classification results. The calculation module is used to remove the seawater region from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results. The remaining sea ice concentration inversion parameters after the removal operation are input into the Pix2PixGAN network, and the ASI algorithm is used to calculate the output results to obtain the sea ice concentration.

[0023] The beneficial effects of the technical solutions provided in this application include: By introducing the Pix2PixGAN network, high-precision sea ice classification is achieved, providing auxiliary information for high-precision sea ice concentration inversion. Based on this, a sea ice concentration inversion model using fused data with the ASI algorithm is established to achieve high-precision sea ice concentration inversion. The sea ice concentration inversion research based on the fusion of passive microwave data and SAR data in this application can obtain high-precision sea ice category and concentration data, thereby improving the spatial resolution and ice zone edge texture details of existing sea ice concentration products and providing reference data for regional Arctic sea ice changes. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the sea ice concentration inversion method based on the fusion of passive microwave and SAR data in this application. Figure 2 This is a schematic diagram of the Pix2PixGAN network structure; Figure 3 This is a schematic diagram of the generator's structure; Figure 4 This is a schematic diagram of the discriminator. Figure 5 for The difference plot and ice-water classification results; Figure 6 A schematic diagram showing the results and comparison of sea ice concentration inversion; Figure 7 This is a detailed schematic diagram of the sea ice concentration inversion method based on the fusion of passive microwave and SAR data in this application; Figure 8 This is a schematic diagram of the functional modules of the sea ice concentration inversion device for passive microwave and SAR data fusion in this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0027] In a first aspect, embodiments of this application provide a method for inverting sea ice concentration by fusing passive microwave and SAR data.

[0028] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the sea ice concentration inversion method based on the fusion of passive microwave and SAR data presented in this application. Figure 1 As shown, the sea ice concentration inversion method based on the fusion of passive microwave and SAR data includes: S1: Acquire Sentinel-1 GRD data to obtain SAR data and preprocess the SAR data, then perform geographic projection and resampling on the preprocessed SAR data; Sentinel-1 GRD data is synthetic aperture radar (SAR) data, mainly used for Earth observation and disaster monitoring. S2: Calculate the sea ice concentration inversion parameters, resample the AMSR2 data, align the resampled AMSR2 data with the resampled SAR data, and crop the overlapping image area to obtain the image pair; AMSR2, or Advanced Microwave Scanning Radiometer 2, is a microwave scanning radiometer carried on the GCOM-W1 satellite, mainly used to observe Earth's surface and atmospheric parameters, and its data is widely used in climate research, hydrological monitoring and other fields; Specifically, the HV polarization data of Sentinel-1 GRD is acquired, and then the acquired SAR data is preprocessed through the SNAP platform developed by the European Space Agency (ESA) to improve the reliability of the SAR data. Then, the preprocessed SAR data is geographically projected and resampled to the same resolution as the AMSR2 data. Finally, the overlapping parts of the images are cropped to obtain the image pairs required for training the Pix2PixGAN network. S3: Train the Pix2PixGAN network based on the obtained image pairs, input some sea ice concentration inversion parameters into the Pix2PixGAN network, and obtain the ice and water classification results; the Pix2PixGAN network is an image translation model based on conditional generative adversarial network (CGAN); Specifically, the Pix2PixGAN network is trained based on the images obtained above, and the Pix2PixGAN network is adjusted to adapt to the input data. Some sea ice concentration inversion parameters are fed into the Pix2PixGAN network, and the ice and water classification results can be obtained based on the fused results. S4: Based on the obtained ice-water classification results, remove the seawater region from the remaining sea ice concentration inversion parameters. Input the remaining sea ice concentration inversion parameters after the removal operation into the Pix2PixGAN network, and use the ASI algorithm to calculate the output results to obtain the sea ice concentration. ASI, or Advanced Sea Ice Algorithm, is an algorithm for sea ice concentration estimation developed by the University of Bremen.

[0029] Specifically, the ice and water classification results obtained above are applied to the remaining sea ice concentration inversion parameters, and then the remaining sea ice concentration inversion parameters are input into the Pix2PixGAN network. The ASI algorithm is used to calculate the output results to obtain the sea ice concentration. Furthermore, it can be compared and evaluated with other concentration products.

[0030] Furthermore, in one embodiment, the preprocessing includes orbit correction, thermal noise removal, radiometric calibration, and speckle filtering.

[0031] For orbit correction, the orbital information accompanying Sentinel-1 data acquisition is a rapid orbit, which has limited accuracy and may contain positioning errors of tens to hundreds of meters. Therefore, it is necessary to update the satellite position and attitude information using accurate orbit files and correct these errors. Orbit files can be downloaded from the European Space Agency's server and applied through the "Apply Orbit File" module in SNAP (Sentinel Application Platform) to improve the spatial positioning accuracy of the data.

[0032] For thermal noise removal, SAR data is affected by the system's own thermal noise during imaging, especially in low backscattering regions (such as calm sea surfaces, dark areas, and polar sea ice), where streaks or uneven background brightness may appear. This phenomenon is more pronounced in the HV band compared to the HH band. This noise is not the true scattering characteristics of the ground objects, but rather spurious signals introduced by the imaging system. In SNAP, the "Thermal Noise Removal" module can be used to denoise the data, making dark areas closer to the true backscattering levels, thereby ensuring the accuracy of subsequent calibration and analysis.

[0033] For radiometric calibration, the raw Sentinel-1 GRD data stores digital quantization values ​​(DN), which only represent relative intensity and have no physical meaning. In SNAP, the "Calibration" module is used for radiometric calibration, which converts these amplitude values ​​into physically meaningful backscattering coefficients. (Sigma Nought), in linear units. Then, it is converted to decibels using the "Linear to / from dB" module. The units are converted to decibels (dB). Radiometrically calibrated data can be quantitatively compared across time series, regional comparisons, and different polarization modes.

[0034] For speckle filtering, SAR imaging utilizes the principle of microwave coherent imaging, inevitably resulting in a large amount of speckle noise, which manifests as granular texture. This noise is caused by the coherent interference of different scatterers and can obscure the true texture and statistical characteristics of ground features. By using SNAP's "Single Product Speckle Filter" function and applying the Refined Lee algorithm, speckle noise can be reduced, improving the visualization and statistical stability of the image.

[0035] Furthermore, in one embodiment, the preprocessed SAR data undergoes geographic projection and resampling, specifically including: S101: Projecting GRD data onto the NSIDC Sea Ice Polar Stereographic North projection with EPSG (an encoding standard) code of 3411 via the SNAP platform; GRD data is ground range detected data acquired by the Sentinel-1 satellite; NSIDC Sea Ice Polar Stereographic North is a coordinate system used for polar geographic data projection; S102: Resample SAR data from the original resolution to a preset distance resolution.

[0036] Specifically, in SNAP, the "Average Height Range-Doppler" module is used to project GRD data onto the NSIDC Sea Ice Polar Stereographic North projection with EPSG code 3411, and to resample SAR data from the original resolution of 40m to a resolution of 1km.

[0037] Furthermore, in one embodiment, AMSR2 data is resampled and the resampled AMSR2 data and resampled SAR data are aligned. The overlapping image regions are then cropped to obtain image pairs, specifically including: S201: Using bicubic convolution, the acquired AMSR2 data is resampled from the original resolution to a preset distance resolution; S202: Align the resampled AMSR2 data and the resampled SAR data, crop out the overlapping areas of the AMSR2 and SAR data of the same date, and reassemble them into an image pair.

[0038] Specifically, the acquired AMSR2 data is resampled from the original 12.5km to 1km using bicubic convolution. Then, the resampled SAR data and AMSR2 data are aligned, and the overlapping areas of the AMSR2 and SAR data on the same date are cropped and reassembled into an image pair.

[0039] Furthermore, in one embodiment, the calculation of sea ice concentration inversion parameters specifically includes:

[0040]

[0041]

[0042] in, , , Indicates the parameters for sea ice concentration inversion. This represents the brightness temperature value for vertical polarization at 18.7 GHz. This represents the brightness temperature value for vertical polarization at 23.8 GHz. This represents the brightness temperature value for vertical polarization at 36.5 GHz. This represents the brightness temperature value for vertical polarization at 89.0 GHz. This represents the brightness temperature value for horizontal polarization at 89.0 GHz.

[0043] Furthermore, in one embodiment, a Pix2PixGAN network is trained based on the obtained image pairs, and some sea ice concentration inversion parameters are input into the Pix2PixGAN network to obtain ice-water classification results, specifically including: S301: Removes data from the resampled SAR data that contains only seawater and no sea ice, thus performing a masking operation on the SAR data; S302: Acquire sea ice concentration products with preset distance resolution from the official AMSR2+MODIS, use the same land mask as the acquired sea ice concentration products to remove the land portion from the resampled AMSR2 data, and perform masking operations on the AMSR2 data; MODIS stands for Medium Resolution Imaging Spectroradiometer data. S303: Calculate the maximum and minimum values ​​of SAR data after the masking operation, and the maximum and minimum values ​​of AMSR2 data after the masking operation. Input the statistically obtained values ​​as parameters into the Pix2PixGAN network. S304: Expand the number of rows and columns of the obtained image pair, and stitch the SAR data and AMSR2 data of the image pair left and right to obtain an image pair with the same number of rows but twice the number of columns as the original image, thus realizing the construction of the image pair required for input to the Pix2PixGAN network. Specifically, the sampled SAR data was examined, and data containing only seawater and no sea ice were removed. Furthermore, since existing thermal noise removal methods still leave some residual thermal noise in the first sub-strip, and the thermal noise removal module of SNAP software cannot completely remove thermal noise, data significantly affected by thermal noise was also removed. Therefore, a 1km resolution sea ice concentration product of AMSR2+MODIS was downloaded from the University of Bremen website, and the same land mask as this sea ice concentration product was used to remove the land portion from the sampled AMSR2 data. Then, the maximum and minimum values ​​of the SAR data and AMSR2 data after the masking operation were statistically analyzed and used as parameters input into the Pix2PixGAN network. Since the Pix2PixGAN network requires at least 256 rows and columns of input images, images with fewer than 256 rows and columns after cropping were padded to 256 rows and columns using invalid values. Then, the SAR data and AMSR2 data were stitched together left and right to obtain an image pair with the same number of rows but twice the number of columns as the original image. This completes the construction of the image pairs required for input into the Pix2PixGAN network. S305: Train the Pix2PixGAN network based on the statistically obtained values ​​and the constructed image pairs; The structure of the Pix2PixGAN network in this application is as follows: Figure 2 As shown, the structure of the generator is as follows: Figure 3 As shown, the structure of the discriminator is as follows: Figure 4 As shown; since the Pix2PixGAN network only considered the use of optical data when it was proposed, when using floating-point data that is different from optical data, it is necessary to additionally input the maximum and minimum values ​​of the left and right sides of the image data to complete the data normalization before training; that is, when training the Pix2PixGAN network, it also includes normalization processing of the training data; the normalization processing specifically includes:

[0044] in, This represents the normalized value. Represents the original value of a point. This represents the minimum value. Indicates the maximum value; The Pix2PixGAN network requires a pair of images as input and outputs fused data. Its generator and discriminator are optimized alternately under the adversarial learning framework. The generator aims to generate a fused image that is as realistic as possible, while the discriminator aims to identify the authenticity of the image as much as possible. When the discriminator cannot distinguish between the original image and the fused image, it outputs the fused image, which is the fused data. The fused data has both the high-definition texture details of SAR data and the brightness temperature range of AMSR2 data. S306: Will and Input the Pix2PixGAN network and output the original image and the fused image; S307: Subtract the original image and the fused image to obtain a difference map; S308: Based on the obtained difference map, the ice-water classification results are obtained.

[0045] Soon and Inputting the Pix2PixGAN network will output original and fused images. Subtracting the original and fused images yields a difference map, as shown below. Figure 5 As shown, Figure 5 In the image, the left image shows the difference map, and the right image shows the ice-water classification results. Due to the different physical properties of seawater and sea ice, the correspondence between their brightness temperature values ​​and backscattering coefficients is not consistent. When sea ice melts, its brightness temperature value gradually increases, while its backscattering coefficient gradually decreases. Utilizing this opposite trend, seawater and sea ice can be distinguished on the difference map. In the sea ice region, the pixel values ​​of the original image are lower than those of the fused image, while in the seawater region, the pixel values ​​of the original image are higher than those of the fused image. By setting a certain threshold, the two can be distinguished, thus obtaining the ice-water classification results.

[0046] Furthermore, in one embodiment, the seawater region is removed from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results. The remaining sea ice concentration inversion parameters after the removal operation are then input into the Pix2PixGAN network, and the sea ice concentration is calculated using the ASI algorithm. Specifically, this includes: S401: Based on the obtained ice-water classification results, remove... The seawater area in the middle will be the area after removing the seawater area. Input the Pix2PixGAN network and obtain the output results; S402: The ASI algorithm is used to calculate the sea ice concentration.

[0047] Specifically, based on the obtained ice-water classification results, remove The seawater region is then input into the Pix2PixGAN network, and the ASI algorithm is used to calculate the sea ice concentration.

[0048] In this application, the ASI algorithm is used to calculate the sea ice concentration from the output results. The specific calculation method is as follows:

[0049] in, Indicates sea ice concentration. express polarization difference, , , , This represents the coefficient, and the coefficient is determined as follows:

[0050] in, , This represents the value of the coordinate point.

[0051] The ASI algorithm uses information from the high-frequency channel (89.0 GHz) and calculates... The polarization difference is used to distinguish between seawater and sea ice. This is because, in the high-frequency band, the polarization difference of various types of ice is much smaller than that of open water. The concentration calculation equation is a third-order polynomial obtained by interpolation between two boundary expressions.

[0052] If the value of the coefficient point and If the value is known, then the sea ice concentration at any point can be calculated. In one possible implementation, The value is 45k. The value is 10k. Finally, the retrieved sea ice density inversion product is compared with other products, such as... Figure 6 As shown, Figure 6 In the middle, the first row from left to right shows the 5km density result based on ice and water classification, the 12.5km density result of the original image ASI algorithm, and the 1km density product from the University of Bremen. The second row from left to right shows the 3km density result based on ice and water classification, the 1km density result of the fused image ASI algorithm, and the 3.125km density product from the University of Bremen.

[0053] The following combinations are as follows Figure 7 The method for inverting sea ice concentration by fusing passive microwave and SAR data in this application is described in detail.

[0054] Step 1: Data Preprocessing. Based on the acquired Sentinel-1 GRD HV polarization data, SAR data preprocessing was performed using the SNAP platform developed by the European Space Agency (ESA). This included orbit correction, thermal noise removal, radiometric calibration, and filtering to improve the reliability of the SAR data. The SAR data was then decibel-diminished, projected onto a geographic projection consistent with the AMSR2 data, and resampled to the same spatial resolution. A land mask provided by the University of Bremen was used to remove land portions from the data, and images not containing sea ice were also removed. Using the AMSR2 multi-frequency polarization data, the gradient ratio and polarization difference parameters required for inversion were calculated. After aligning the multi-source data, overlapping areas from the same date were cropped and stitched together in pairs to provide high-quality data for subsequent training of the adversarial neural network.

[0055] Step 2: Sea ice classification based on original images and fused data. Using the image pairs of AMSR2 gradient ratio data and decibel-ratio SAR data obtained in Step 1, a Pix2PixGAN network (model) is trained. The difference map is obtained by subtracting the original images from the output fused data, and an appropriate difference threshold is found to distinguish between sea ice and seawater.

[0056] Step 3: Sea ice concentration inversion based on fused data. Using the ice-water classification information obtained in Step 2 as auxiliary information, the seawater portion of the AMSR2 polarization difference data is removed, and then input into the Pix2PixGAN network. The ASI algorithm is then used to calculate sea ice concentration from the fused data, and the results are compared and evaluated with existing products.

[0057] This application addresses the problems of sea ice classification and sea ice concentration inversion, proposing a method based on adversarial neural networks (ANNs) using Sentinel-1 and AMSR2 data. To address the issue of low resolution in sea ice concentration products, a method for obtaining high-resolution sea ice concentration products is provided by introducing higher-resolution SAR data and employing an adversarial generative neural network.

[0058] First, preprocessing techniques such as orbit correction, thermal noise removal, and radiometric calibration are used to obtain registered SAR HV polarization data and AMSR2 gradient ratio and polarization difference data with the same resolution, providing high-quality data for subsequent sea ice information extraction. Based on this, a Pix2PixGAN network is used to obtain high-resolution fused AMSR2 gradient ratio data, which is then used to obtain high-precision sea ice classification results. Then, based on the sea ice classification results, the seawater portion of the AMSR2 polarization difference data is removed before being input into the network. Finally, the ASI algorithm is used on the output fused polarization difference data to obtain high-resolution sea ice concentration results.

[0059] Secondly, embodiments of this application also provide a sea ice concentration inversion device that fuses passive microwave and SAR data.

[0060] In one embodiment, reference is made to Figure 8 , Figure 8 This is a schematic diagram of the functional modules of the sea ice concentration inversion device for passive microwave and SAR data fusion according to this application. Figure 8 As shown, the sea ice concentration inversion device for passive microwave and SAR data fusion includes: an acquisition module, a cropping module, an adjustment module, and a calculation module.

[0061] The acquisition module acquires Sentinel-1 GRD data to obtain SAR data and preprocesses the SAR data, performing geographic projection and resampling on the preprocessed SAR data. The cropping module calculates the sea ice concentration inversion parameters, resamples AMSR2 data, aligns the resampled AMSR2 data with the resampled SAR data, and cropps out overlapping image areas to obtain image pairs. The adjustment module trains a Pix2PixGAN network based on the obtained image pairs, inputs some sea ice concentration inversion parameters into the Pix2PixGAN network, and obtains ice-water classification results. The calculation module removes seawater regions from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results, inputs the remaining sea ice concentration inversion parameters after removal into the Pix2PixGAN network, and uses the ASI algorithm to calculate the output results to obtain sea ice concentration.

[0062] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0063] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0064] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0065] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0066] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0067] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for inverting sea ice concentration by fusing passive microwave and SAR data, characterized in that, The sea ice concentration inversion method based on the fusion of passive microwave and SAR data includes: Acquire Sentinel-1 GRD data to obtain SAR data and preprocess the SAR data, then perform geographic projection and resampling on the preprocessed SAR data; The sea ice concentration inversion parameters were calculated, the AMSR2 data were resampled, and the resampled AMSR2 data and the resampled SAR data were aligned. The overlapping areas of the images were cropped to obtain image pairs. The Pix2PixGAN network is trained based on the obtained image pairs. Some sea ice concentration inversion parameters are input into the Pix2PixGAN network to obtain ice and water classification results. Based on the obtained ice-water classification results, the seawater region is removed from the remaining sea ice concentration inversion parameters. The remaining sea ice concentration inversion parameters after the removal operation are then input into the Pix2PixGAN network, and the ASI algorithm is used to calculate the output results to obtain the sea ice concentration.

2. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 1, characterized in that: The preprocessing includes track correction, thermal noise removal, radiometric calibration, and speckle filtering.

3. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 1, characterized in that, The process of geographic projection and resampling of the preprocessed SAR data specifically includes: The GRD data was projected onto the NSIDC Sea Ice PolarStereographic North projection with EPSG code 3411 using the SNAP platform. Resample SAR data from its original resolution to a preset distance resolution.

4. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 1, characterized in that, The process of resampling AMSR2 data, aligning the resampled AMSR2 data with the resampled SAR data, and cropping the overlapping image region to obtain an image pair specifically includes: The acquired AMSR2 data is resampled from the original resolution to a preset distance resolution using a bicubic convolution method. The resampled AMSR2 data and resampled SAR data are aligned, and the overlapping areas of the AMSR2 and SAR data on the same date are cropped and reassembled into an image pair.

5. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 1, characterized in that, The calculated sea ice concentration inversion parameters specifically include: in, , , Indicates the parameters for sea ice concentration inversion. This represents the brightness temperature value for vertical polarization at 18.7 GHz. This represents the brightness temperature value for vertical polarization at 23.8 GHz. This represents the brightness temperature value for vertical polarization at 36.5 GHz. This represents the brightness temperature value for vertical polarization at 89.0 GHz. This represents the brightness temperature value for horizontal polarization at 89.0 GHz.

6. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 5, characterized in that, The Pix2PixGAN network is trained based on the obtained image pairs. Some sea ice concentration inversion parameters are input into the Pix2PixGAN network to obtain ice-water classification results, specifically including: This involves removing data from the resampled SAR data that contains only seawater and no sea ice, thus enabling a masking operation on the SAR data. Obtain the sea ice concentration product with the preset distance resolution of the official AMSR2+MODIS, use the same land mask as the obtained sea ice concentration product, remove the land portion in the resampled AMSR2 data, and realize the masking operation on the AMSR2 data. The maximum and minimum values ​​of SAR data and AMSR2 data after the masking operation are statistically analyzed, and the statistically obtained values ​​are used as parameters to input into the Pix2PixGAN network. The number of rows and columns of the obtained image pair is expanded, and the SAR data and AMSR2 data in the image pair are stitched together left and right to obtain an image pair with the same number of rows but twice the number of columns as the original image, thus realizing the construction of the image pair required for input to the Pix2PixGAN network. The Pix2PixGAN network was trained based on the statistically obtained values ​​and the constructed image pairs. Will and Input the Pix2PixGAN network and output the original image and the fused image; The difference between the original image and the fused image is calculated to obtain a difference map; Based on the obtained difference map, the ice-water classification results are obtained.

7. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 6, characterized in that: Training the Pix2PixGAN network also includes normalizing the training data. The normalization process specifically involves: in, This represents the normalized value. Represents the original value of a point. This represents the minimum value. This represents the maximum value.

8. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 6, characterized in that, The process involves removing the seawater region from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results, inputting the removed remaining sea ice concentration inversion parameters into the Pix2PixGAN network, and using the ASI algorithm to calculate the output results to obtain the sea ice concentration. Specifically, this includes: Based on the obtained ice-water classification results, remove The seawater area in the middle will be the area after removing the seawater area. Input the Pix2PixGAN network and obtain the output results; The ASI algorithm is used to calculate the sea ice concentration from the output results.

9. The sea ice concentration inversion method based on passive microwave and SAR data fusion as described in claim 8, characterized in that, The ASI algorithm is used to calculate the sea ice concentration from the output results. The specific calculation method is as follows: in, Indicates sea ice concentration. express polarization difference, , , , This represents the coefficient, and the coefficient is determined as follows: in, , This represents the value of the coordinate point.

10. A sea ice concentration inversion device for passive microwave and SAR data fusion, characterized in that, The sea ice concentration inversion device that fuses passive microwave and SAR data includes: The acquisition module is used to acquire Sentinel-1 GRD data to obtain SAR data and preprocess the SAR data, and perform geographic projection and resampling on the preprocessed SAR data. The cropping module is used to calculate the sea ice concentration inversion parameters, resample AMSR2 data, align the resampled AMSR2 data with the resampled SAR data, and crop out the overlapping image areas to obtain image pairs. The adjustment module is used to train the Pix2PixGAN network based on the obtained image pairs, inputting some sea ice concentration inversion parameters into the Pix2PixGAN network to obtain ice and water classification results. The calculation module is used to remove the seawater region from the remaining sea ice concentration inversion parameters based on the obtained ice-water classification results. The remaining sea ice concentration inversion parameters after the removal operation are input into the Pix2PixGAN network, and the ASI algorithm is used to calculate the output results to obtain the sea ice concentration.

Images