Water body detection method, system, storage medium and electronic device for spaceborne SAR images based on DBO-CNN model

Abstract

The present invention discloses a method, system, storage medium and electronic device for water body detection using satellite-borne SAR images based on the DBO-CNN model, including: applying a GRD image of a dual-polarized SAR image to obtain backscatter features VV and VH; applying an SLC image of a dual-polarized SAR image to obtain five polarization features; applying a feature combination method to combine two backscatter features and five polarization features to obtain a total of eight candidate feature combinations; applying a DBO algorithm to find the optimal hyperparameters of the CNN model to obtain the optimal CNN model; applying two decision-level fusion methods to fuse the two optimal feature combinations to optimize water body detection results. The present invention can effectively contribute to the development of water body detection technology using remote sensing images, and provide technical support for the precise layout and effective promotion of regional water ecological protection, water resources management, and coastline reconstruction.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing images, and in particular to a method, system, storage medium and electronic equipment for detecting water bodies in spaceborne SAR images based on a DBO-CNN model. Background Art

[0002] The development of remote sensing technology has made it possible to conduct near-real-time, large-scale water monitoring on a global scale, with unparalleled spatial and temporal coverage. Remote sensing has greatly facilitated water monitoring research. However, despite its widespread application and significant effectiveness, remote sensing technology faces some unique and inherent challenges in the specific field of water monitoring.

[0003] First, the complexity of water body detection lies in the diversity and complexity of water itself. Due to the variability of water's natural properties, its reflection and scattering characteristics vary with factors such as water quality, depth, and surface conditions (such as waves and suspended matter). For example, lakes, rivers, and oceans may exhibit different reflection and scattering characteristics due to their unique physical and chemical properties. Furthermore, complex terrestrial environments, such as urban buildings, vegetation, and soil types, can interfere with water body detection. This makes water body detection using remote sensing technology particularly difficult. Second, traditional remote sensing water body detection methods primarily rely on backscatter analysis, which primarily analyzes and interprets backscatter information from SAR imagery. However, this approach generally overlooks the potential of polarization decomposition methods for water body detection. Finally, although deep learning and neural networks have demonstrated powerful capabilities in processing remote sensing images, how to effectively utilize these tools remains a problem that requires further research. Most deep learning methods focus on model optimization, but the setting of hyperparameters has always been based on empirical values. Improving the accuracy, robustness, and efficiency of water body detection remains an unresolved issue. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system, storage medium and electronic equipment for water body detection in spaceborne SAR images based on the DBO-CNN model, which can obtain accurate water body detection results and improve the accuracy, robustness and efficiency of water body detection.

[0005] The technical solution adopted in the present invention is:

[0006] A water body detection method for spaceborne SAR images based on a DBO-CNN model, comprising:

[0007] Step S101: pre-processing the GRD image data to obtain backscattering features VV and VH;

[0008] Step S102: pre-process the SLC image and proceed to the next step;

[0009] Step S103: Applying the H / α polarization decomposition method to perform polarization decomposition on the pre-processed SLC image to obtain two polarization features, H and α, where H represents polarization entropy and α represents average scattering angle;

[0010] Step S104: Apply the model-based polarization decomposition method to perform polarization decomposition on the pre-processed SLC image to obtain m s , m v and m dif Three polarization characteristics; m s represents the surface scattering component, m v represents the volume scattering component, m dif Represents the difference between the surface scattering component and the volume scattering component;

[0011] Step S105: Apply the DBO algorithm to find the optimal hyperparameters of the CNN model to obtain the optimal CNN model, i.e., the DBO-CNN model;

[0012] Step S106: Apply the DBO-CNN model to perform pixel-by-pixel dense water body prediction tasks to robustly and accurately detect water bodies, and select two optimal feature combinations based on the detection results, which are derived from the fusion of polarization features and backscattering features obtained by the two polarization decomposition methods respectively;

[0013] Step S107: Apply two decision-level fusion methods to fuse the two optimal features to optimize the water body detection result.

[0014] The pre-processing in step S101 specifically includes orbit correction, thermal noise removal, radiometric calibration, multi-look, speckle filtering, terrain correction and decibelization.

[0015] Step S103 is specifically as follows:

[0016] The scattering matrix in the VV-VH polarization mode contains only two non-zero vectors:

[0017]

[0018] The Pauli basis corresponding to the scattering matrix of the Mth pixel can be expressed as:

[0019]

[0020] Therefore, the multi-view coherence matrix based on dual-polarization SAR images can be expressed as:

[0021]

[0022] Where N represents the number of views, the superscript H represents the vector conjugate transpose, and j represents the imaginary unit;

[0023] Therefore, the 2×2 coherence matrix T can be eigen-decomposed as follows:

[0024]

[0025] Among them, λ q (q=1,2) represents the eigenvalue, u q represents the eigenvector;

[0026] Each eigenvector can be expressed using α q ,β q ,φ q and δ q The four corners represent:

[0027]

[0028] where α represents the scattering mechanism, β represents the azimuth angle, and φ and δ represent the phase angles;

[0029] The mean value of the α angle of the dual-polarization SAR image is defined as:

[0030]

[0031] in,

[0032] The entropy value can be obtained by: Calculated.

[0033] Step S104 is specifically as follows:

[0034] After preprocessing the SLC image, each pixel will obtain a C2 matrix:

[0035]

[0036] Where <> represents multi-look or speckle filtering;

[0037] C 2×2 Convert to Stokes matrix:

[0038]

[0039] Where Re(c 12 ) and Im(c 12 ) represent c 12 The real and imaginary parts of the wave, S1 is proportional to the total amplitude of the wave, S2 represents the amplitude difference between the horizontal and vertical components, S3 and S4 represent the phase difference between the horizontal and vertical components;

[0040] The Stokes vector S can be decomposed into:

[0041]

[0042] Among them, s v and s p represent partially polarized waves and fully polarized waves, respectively, and m v and m s represents the corresponding energy; and m v It can be calculated by a linear equation of two variables;

[0043]

[0044] Where a, b and c are calculated by the random dipole cloud model;

[0045] There is only one root that satisfies the law of conservation of energy, that is, when m v ≤s1, from which we can get the unique m v The solution is, obviously, m s and m dif Can be:

[0046] m s =s1-m v

[0047] β dif =m v -m s

[0048] Calculated, m s represents the surface scattering component, m dif Represents the difference between surface scattering and volume scattering components.

[0049] Step S105 is specifically as follows:

[0050] The DBO algorithm is applied to find the optimal hyperparameters of the CNN model, including the learning rate, the number of iterations and the number of small samples.

[0051] Step S106 specifically includes: applying the optimal hyperparameters to establish the optimal CNN model, namely DBO-CNN, taking the eight candidate feature combinations as input, and using DBO-CNN to perform pixel-by-pixel dense water body prediction tasks:

[0052] First, the optimal feature combination of backscattering features and polarization features obtained by H / α polarization decomposition method is selected for feature fusion, which is called combination H.

[0053] Then, the optimal feature combination for feature fusion of the backscattering feature and the polarization feature obtained by the model-based polarization decomposition method is selected, which is named combination M.

[0054] Step S107 is specifically as follows:

[0055] The two best features are combined, which are the fusion H and M of the polarization features and backscattering features from the two polarization decomposition methods respectively. Let the combination H be Let the combination M be The decision-level fusion formula used is:

[0056]

[0057]

[0058] Among them, HaM is the union of combination H and combination M, and HoM is the intersection of combination H and combination M. According to the results of HaM and HoM, the optimal decision-level fusion method is selected.

[0059] A system for water body detection in spaceborne SAR images based on a DBO-CNN model, comprising:

[0060] a scattering feature acquisition and combination unit, configured to obtain backscattering features and polarization features through the dual-polarization SAR image, and perform feature combination on the backscattering features and the polarization features;

[0061] The model building and prediction unit is configured to apply the DBO algorithm to find the optimal hyperparameters of CNN to build the optimal CNN model, namely DBO-CNN, and use DBO-CNN to perform pixel-by-pixel dense water body prediction tasks and select the optimal feature combination H and combination M;

[0062] The decision-level fusion unit is configured to perform two decision-level fusions on the combination H and the combination M.

[0063] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the device where the computer-readable storage medium is located executes the method for water body detection in spaceborne SAR images based on a DBO-CNN model.

[0064] An electronic device includes: a memory and a processor, wherein the memory stores a program that can be run on the processor, and when the processor executes the program, the spaceborne SAR image water body detection method based on the DBO-CNN model is implemented.

[0065] This paper uses dual-polarization SAR images to extract backscatter and polarization features, performs feature combination, and then uses the DBO algorithm to find the optimal hyperparameters of the CNN model to establish an optimal CNN model. The DBO-CNN model is used for intensive water classification tasks, and the optimal feature combination of backscatter features and polarization features obtained from two polarization decomposition methods is selected. Finally, two decision-level fusion methods are used to optimize water detection results. This method can obtain accurate water detection results, effectively increasing the robustness and automation of the water detection process, and provides new insights and methods for water detection tasks using dual-polarization SAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0067] Figure 1 is a flow chart of the present invention;

[0068] Figure 2 This is a structural block diagram of the invented system. DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0070] like Figure 1 and 2 As shown, the present invention includes the following steps:

[0071] Step S101: Apply the GRD image of the dual-polarized SAR image to obtain backscattering features VV and VH;

[0072] In the embodiment of the present application, the GRD data is subjected to orbit correction, thermal noise removal, radiometric calibration, multi-look, speckle filtering, terrain correction and decibelization preprocessing to obtain the backscattering characteristics of VV and VH polarizations.

[0073] Step S102: applying the SLC image of the dual-polarized SAR image to obtain backscattering features;

[0074] In an embodiment of the present application, the SLC image is preprocessed with orbit correction, radiation correction, deburst, polarization matrix, multi-view, speckle removal and geocoding, and then five polarization features are obtained by using two polarization decomposition methods.

[0075] Step S103: applying the H / α polarization decomposition method to perform polarization decomposition on the SLC image to obtain two polarization features, H and α;

[0076] In the embodiment of the present application, the scattering matrix in the VV-VH polarization mode contains only two non-zero vectors:

[0077]

[0078] The Pauli basis corresponding to the scattering matrix of the Mth pixel can be expressed as:

[0079]

[0080] Therefore, the multi-view coherence matrix based on dual-polarization SAR images can be expressed as:

[0081]

[0082] Where N represents the number of views and the superscript H represents the vector conjugate transpose.

[0083] Therefore, the 2×2 coherence matrix T can be eigen-decomposed as follows:

[0084]

[0085] Among them, λ q (q=1,2) represents the eigenvalue, u q represents the feature vector.

[0086] Each eigenvector can be expressed using α q ,β q ,φ q and δ q The four corners represent:

[0087]

[0088] where α represents the scattering mechanism, β represents the azimuth angle, and φ and δ represent the phase angles.

[0089] The mean value of the α angle of the dual-polarization SAR image is defined as:

[0090]

[0091] in,

[0092]

[0093] The entropy value can be obtained by:

[0094]

[0095] Calculated.

[0096] Step S104: Apply the model-based polarization decomposition method to perform polarization decomposition on the SLC image to obtain m s , m v and m dif Three polarization characteristics;

[0097] In the embodiment of the present application, after pre-processing the SLC image, each pixel will obtain a C2 matrix:

[0098]

[0099] Where <> represents multi-look or speckle filtering, S VH Indicates the cross-polarization component of vertical transmission and horizontal reception, S VV Indicates vertical transmission and vertical reception of co-polarization components.

[0100] C 2×2 Convert to Stokes matrix:

[0101]

[0102] Where Re(c 12 ) and Im(c 12 ) represent c 12 The real and imaginary parts of .

[0103] The Stokes vector S can be decomposed into:

[0104]

[0105] Among them, s v and s p represent partially polarized waves and fully polarized waves, respectively, and m v and m s Represents the corresponding energy. And m v It can be calculated through a linear equation of two variables.

[0106]

[0107] where a, b and c are all calculated using the random dipole cloud model.

[0108] There is only one root that satisfies the law of conservation of energy, that is, when m v ≤s1. From this we can get the unique mv The solution.

[0109] Obviously, m s and m dif Can be:

[0110] m s =s1-m v

[0111] m di f=m v -m s

[0112] Calculated.

[0113] Step S105: Apply the DBO algorithm to find the optimal hyperparameters of the CNN model to obtain the optimal CNN model, i.e., the DBO-CNN model;

[0114] In the embodiments of this application, the DBO algorithm is used to find the optimal hyperparameters of the CNN model, including the learning rate, number of iterations, and number of small samples. The learning rate is used to control the step size for adjusting the neural network weights. If the learning rate is set too high, training may oscillate around the optimal solution and fail to converge. If it is set too low, while the optimal solution can be better approached, the training speed will be very slow, and a large number of iterations may be required to reach the optimal solution. Therefore, choosing an appropriate learning rate is crucial to the speed and quality of neural network training. The number of iterations refers to the number of times data passes through the neural network during training. If the number of iterations is too low, the model may underfit, meaning that the model fails to learn all the patterns in the data. Conversely, if the number of iterations is too high, it may lead to overfitting, meaning that the model learns the training data too deeply and cannot generalize well to unknown data. Therefore, choosing an appropriate number of iterations to achieve a good model fit is very critical. The number of small samples refers to the number of samples in a single forward and backward propagation during neural network training. The choice of the number of small samples also has a significant impact on model training. If the number of small samples is too large, computational efficiency can be improved but the model's generalization ability may be reduced. If the number of small samples is too small, the model's generalization ability may be enhanced but the training time may increase significantly and the training process may become unstable. Therefore, choosing an appropriate number of small samples can improve the model's generalization ability while maintaining efficient training.

[0115] Step S106: Apply the DBO-CNN model to perform pixel-by-pixel dense water body prediction tasks to robustly and accurately detect water bodies, and select two optimal feature combinations based on the detection results, which are derived from the fusion of polarization features and backscattering features obtained by the two polarization decomposition methods respectively;

[0116] In an embodiment of the present application, an optimal CNN model, namely DBO-CNN, is established with optimal hyperparameters, and eight candidate feature combinations are used as input. DBO-CNN is used to perform pixel-by-pixel dense water body prediction tasks. First, the optimal feature combination when fusing backscattering features and polarization features obtained by the H / α polarization decomposition method is selected, and is named combination H. Then, the optimal feature combination when fusing backscattering features and polarization features obtained by the model-based polarization decomposition method is selected, and is named combination M.

[0117] Step S107: Apply two decision-level fusion methods to fuse the two optimal features to optimize the water body detection result;

[0118] In the embodiment of the present application, two optimal features are combined, which are the fusion H and M of the polarization features and backscattering features from two polarization decomposition methods. Let the combination H be Let the combination M be

[0119] In the embodiment of the present application, the decision-level fusion formula is:

[0120]

[0121]

[0122] Among them, HaM is the union of combination H and combination M, and HoM is the intersection of combination H and combination M. According to the results of HaM and HoM, the optimal decision-level fusion method is selected.

[0123] like Figure 2 As shown, the DBO-CNN-based water body detection system for spaceborne SAR images includes a scattering feature acquisition and combination unit, a model building and prediction unit, and a decision-level fusion unit. The scattering feature acquisition and combination unit is configured to obtain backscattering features and polarization features from dual-polarization SAR images and perform feature combination of the backscattering features and polarization features. The model building and prediction unit is configured to apply the DBO algorithm to find the optimal hyperparameters of the CNN to establish the optimal CNN model, namely DBO-CNN. The DBO-CNN is used to perform pixel-by-pixel dense water body prediction tasks and select the optimal feature combinations H and M. The decision-level fusion unit is configured to perform two decision-level fusions on the combinations H and M.

[0124] In some optional embodiments, the scattering feature acquisition and combination unit is further configured to use the GRD (Ground Range Detected, GRD) image to obtain the backscattering features VV and VH. Using the SLC (Single Look Complex, SLC) image of the corresponding position, after pre-processing the SLC image, each pixel will obtain a C z matrix:

[0125]

[0126] Here, <> represents multi-look or speckle filtering.

[0127] C 2×2 Convert to Stokes matrix:

[0128]

[0129] Where Re(c 12 ) and Im(c 12 ) represent c 12 The real and imaginary parts of .

[0130] The Stokes vector S can be decomposed into:

[0131]

[0132] Among them, s v and s p represent partially polarized waves and fully polarized waves, respectively, and m v and m s Indicates the corresponding energy. And m v It can be calculated through a linear equation of two variables.

[0133]

[0134] where a, b and c are all calculated using the random dipole cloud model.

[0135] There is only one root that satisfies the law of conservation of energy, that is, when m v ≤s1. From this we can get the unique m v The solution.

[0136] Obviously, m s and m dif Can be:

[0137] m s =s1-m v

[0138] m di f=mv -m s

[0139] Calculated.

[0140] For the H / α decomposition method, the scattering matrix in the VV-VH polarization mode contains only two non-zero vectors:

[0141]

[0142] The Pauli basis corresponding to the scattering matrix of the Mth pixel can be expressed as:

[0143]

[0144] Therefore, the multi-view coherence matrix based on dual-polarization SAR images can be expressed as:

[0145]

[0146] Where N represents the number of views and the superscript H represents the vector conjugate transpose.

[0147] Therefore, the 2×2 coherence matrix T can be eigen-decomposed as follows:

[0148]

[0149] Among them, λ q (q=1,2) represents the eigenvalue, u q represents the feature vector.

[0150] Each eigenvector can be expressed using α q ,β q ,φ q and δ q The four corners represent:

[0151]

[0152] where α represents the scattering mechanism, β represents the azimuth angle, and φ and δ represent the phase angles.

[0153] The mean value of the α angle of the dual-polarization SAR image is defined as:

[0154]

[0155] in,

[0156]

[0157] The entropy value can be obtained by:

[0158]

[0159] Calculated.

[0160] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the device where the computer-readable storage medium is located executes the method for water body detection in spaceborne SAR images based on a DBO-CNN model.

[0161] An electronic device includes: a memory and a processor, wherein the memory stores a program that can be run on the processor, and the processor executes the spaceborne SAR image water body detection method based on the DBO-CNN model.

[0162] The proposed method utilizes a DBO-CNN model to perform pixel-by-pixel intensive classification, using fused backscatter and polarimetric features as model input. This method leverages the potential of deep learning for water body detection based on SAR imagery and improves water body detection performance by optimizing the hyperparameters of the CNN model.

[0163] First, a significant advantage of this method is that it utilizes the DBO algorithm to optimize the hyperparameters of the CNN model. In most existing studies, researchers have focused primarily on optimizing the structure of the CNN model while neglecting the potential impact of optimizing the model's hyperparameters. However, the hyperparameters of the CNN model have a significant impact on model performance. Selecting the optimal set of hyperparameters can improve the performance of the CNN model in water body detection without changing the CNN structure. Therefore, it is crucial to select appropriate hyperparameters to fully realize the performance potential of the CNN model.

[0164] Second, the proposed method uses fused backscatter and polarization features as model input, which provides richer ground object information to the CNN model. Backscatter features provide information about the backscatter intensity of water bodies, while polarization features provide information about the polarization of water bodies. The fusion of these two features may improve the performance of water body detection in SAR imagery.

[0165] Finally, two decision-level fusion strategies are proposed to improve the accuracy of water body detection results. Decision-level fusion is to fuse the output results of multiple different fusion methods together through a certain strategy to improve the accuracy of detection results.

[0166] Overall, the proposed method not only improves model performance but also fully exploits the potential of deep learning for water detection in SAR images. We look forward to further research to further improve the performance of this method and apply it to a wider range of scenarios.

[0167] The system for water body detection in spaceborne SAR images based on the DBO-CNN model provided in the embodiments of the present application can implement the steps and processes of any of the above-mentioned method embodiments for water body detection in spaceborne SAR images based on the DBO-CNN model, and achieve the same technical effects, which will not be repeated here.

[0168] In the description of the present invention, it should be noted that, for directional words, such as the terms "center", "horizontal", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like, indicating directions and positional relationships, are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and cannot be understood as limiting the specific scope of protection of the present invention.

[0169] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or apparatuses.

[0170] Note that the above are only preferred embodiments of the present invention and the principles of the technology used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention is described in detail through the above embodiments, the present invention is not limited to the specific embodiments described herein. Without departing from the concept of the present invention, it may also include many other effective embodiments, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for water body detection in spaceborne SAR images based on the DBO-CNN model, characterized in that: include: Step S101: pre-processing the GRD image data to obtain backscattering features VV and VH; Step S102: pre-process the SLC image and proceed to the next step; Step S103: Applying the H / α polarization decomposition method to perform polarization decomposition on the pre-processed SLC image to obtain two polarization features, H and α, where H represents polarization entropy and α represents average scattering angle; Step S104: Apply the model-based polarization decomposition method to perform polarization decomposition on the pre-processed SLC image to obtain m s , m v and m dif Three polarization characteristics; m s represents the surface scattering component, m v represents the volume scattering component, m dif Represents the difference between the surface scattering component and the volume scattering component; Step S105: Apply the DBO algorithm to find the optimal hyperparameters of the CNN model to obtain the optimal CNN model, that is, the DBO-CNN model. Step S105 specifically includes: applying the DBO algorithm to find the optimal hyperparameters of the CNN model, the optimal hyperparameters including the learning rate, the number of iterations, and the number of small samples; Step S106: Apply the DBO-CNN model to perform pixel-by-pixel dense water body prediction tasks to robustly and accurately detect water bodies, and select two optimal feature combinations based on the detection results, which are derived from the fusion of polarization features and backscattering features obtained by the two polarization decomposition methods respectively; Step S107: Apply two decision-level fusion methods to fuse the two optimal features to optimize the water body detection result.

2. The method for water body detection in spaceborne SAR images based on the DBO-CNN model according to claim 1, characterized in that: The pre-processing in step S101 specifically includes orbit correction, thermal noise removal, radiometric calibration, multi-look, speckle filtering, terrain correction and decibelization.

3. The method for water body detection in spaceborne SAR images based on the DBO-CNN model according to claim 1, characterized in that: Step S103 is specifically as follows: The scattering matrix in the VV-VH polarization mode contains only two non-zero vectors: The Pauli basis corresponding to the scattering matrix of the Mth pixel can be expressed as: Therefore, the multi-view coherence matrix based on dual-polarization SAR images can be expressed as: Where N represents the number of views, the superscript H represents the vector conjugate transpose, and j represents the imaginary unit; Therefore, the 2×2 coherence matrix T can be eigen-decomposed as follows: Among them, λ q (q=1,2) represents the eigenvalue, u q represents the eigenvector; Each eigenvector can be expressed using α q ,β q ,φ q and δ q The four corners represent: where α represents the scattering mechanism, β represents the azimuth angle, and φ and δ represent the phase angles; The mean value of the α angle of the dual-polarization SAR image is defined as: in, The entropy value can be obtained by: Calculated.

4. The method for water body detection in spaceborne SAR images based on the DBO-CNN model according to claim 1, characterized in that: Step S104 is specifically as follows: After preprocessing the SLC image, each pixel will obtain a C2 matrix: Where <> represents multi-look or speckle filtering; C 2×2 Convert to Stokes matrix: Where Re(c 12 ) and Im(c 12 ) represent c 12 The real and imaginary parts of the wave, S1 is proportional to the total amplitude of the wave, S2 represents the amplitude difference between the horizontal and vertical components, S3 and S4 represent the phase difference between the horizontal and vertical components; The Stokes vector S can be decomposed into: Among them, s v and s p represent partially polarized waves and fully polarized waves, respectively, and m v and m s represents the corresponding energy; and m v It can be calculated by a linear equation of two variables; Where a, b and c are calculated by the random dipole cloud model; There is only one root that satisfies the law of conservation of energy, that is, when m v ≤s1, from which we can get the unique m v The solution is, obviously, m s and m dif Can be: m s =s1-m v m dif =m v -m s Calculated, m s represents the surface scattering component, m dif Represents the difference between surface scattering and volume scattering components.

5. The method for water body detection in spaceborne SAR images based on the DBO-CNN model according to claim 1, characterized in that: Step S106 specifically includes: applying the optimal hyperparameters to establish the optimal CNN model, namely DBO-CNN, taking the eight candidate feature combinations as input, and using DBO-CNN to perform pixel-by-pixel dense water body prediction tasks: First, the optimal feature combination of backscattering features and polarization features obtained by H / α polarization decomposition method is selected for feature fusion, which is called combination H. Then, the optimal feature combination for feature fusion of the backscattering feature and the polarization feature obtained by the model-based polarization decomposition method is selected, which is named combination M.

6. The method for water body detection in spaceborne SAR images based on the DBO-CNN model according to claim 1, characterized in that: Step S107 is specifically as follows: The two best features are combined, which are the fusion H and M of the polarization features and backscattering features from the two polarization decomposition methods respectively; let the combination H be Let the combination M be The decision-level fusion formula used is: Among them, HaM is the union of combination H and combination M, and HoM is the intersection of combination H and combination M. According to the results of HaM and HoM, the optimal decision-level fusion method is selected.

7. A system based on the DBO-CNN model-based water body detection method for spaceborne SAR images according to claim 1, characterized in that: include: a scattering feature acquisition and combination unit, configured to obtain backscattering features and polarization features through the dual-polarization SAR image, and perform feature combination on the backscattering features and the polarization features; The model building and prediction unit is configured to apply the DBO algorithm to find the optimal hyperparameters of CNN to build the optimal CNN model, namely DBO-CNN, and use DBO-CNN to perform pixel-by-pixel dense water body prediction tasks and select the optimal feature combination H and combination M; The decision-level fusion unit is configured to perform two decision-level fusions on the combination H and the combination M.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the device where the computer-readable storage medium is located executes the method for water body detection in spaceborne SAR images based on the DBO-CNN model according to any one of claims 1 to 7.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a program that can be run on the processor, and when the processor executes the program, the method for water body detection in spaceborne SAR images based on the DBO-CNN model according to any one of claims 1 to 7 is implemented.

Images