Salt marsh vegetation extraction method based on multi-polarization SAR target decomposition and random forest classification

Abstract

The invention relates to a salt marsh vegetation extraction method based on multi-polarization SAR target decomposition and random forest classification, and the method comprises the steps: obtaining multi-polarization SAR data of a target region at different times, and carrying out the preprocessing; converting the pixel value of the preprocessed image data into the back scattering intensity of an actual ground object and extracting a scattering matrix; obtaining a covariance matrix according to the scattering matrix; entropy / balance degree / main polarization angle decomposition is carried out on the covariance matrix; dividing the sample points into a test set and a training set; and classifying the salt marsh vegetation by adopting a random forest classification method. The target decomposition method has the beneficial effects that the high spatial-temporal variation characteristic of salt marsh vegetation polarization scattering and the high spatial heterogeneity and fine classification requirements of salt marsh vegetation distribution are considered, the target decomposition method of the multi-polarization SAR data is provided, and the multi-polarization SAR target decomposition parameters are obtained; and further realizing SAR fine identification of coastal wetland salt marsh vegetation in combination with random forest classification.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing images, and more precisely, it relates to a method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification. Background Art

[0002] The salt marsh wetland accounts for about 1.12% of the total area of coastal wetlands, but provides 10%-15% of the global carbon burial. Vegetation is the "engineer" of the coastal wetland ecosystem and also the "barometer" of the ecosystem health. As the producer of the coastal wetland ecosystem, salt marsh vegetation can significantly improve the carbon burial capacity of the coastal wetland system; and by increasing the surface roughness, hindering the water movement, enhancing the water retention and storage capacity of the wetland, it affects the development and evolution of the wetland. At the same time, the presence of vegetation can effectively weaken the wave energy and flow velocity, promote sediment deposition, and reduce the risk of coastal wetland degradation. Therefore, how to strengthen the change monitoring and protection of salt marsh vegetation in coastal wetlands, and timely master the types and spatial distribution of different salt marsh vegetation has become an important problem that urgently needs to be solved for the goals of coastal wetland restoration and reconstruction, wetland resource protection and rational utilization in China.

[0003] Synthetic Aperture Radar (SAR) has been widely used in the fields of surface coverage monitoring and vegetation classification due to its all-weather and all-day imaging ability. The SAR system in the full polarization observation mode can obtain more comprehensive polarization information of ground objects, and it has been proved that the polarization scattering characteristics obtained by polarization target decomposition are helpful to improve the classification accuracy of salt marsh vegetation. However, the limitations of the full polarization SAR data, such as small observation swath width and high price, still cannot be ignored. Especially for complex objects such as salt marsh vegetation in coastal wetlands with high dynamic changes, the small image swath width and single-scene / few full polarization SAR images are not suitable for large-scale salt marsh identification, and the long time series advantage of SAR data cannot be utilized. Summary of the Invention

[0004] The purpose of the present invention is to propose a method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification in view of the deficiencies of the prior art.

[0005] In the first aspect, a method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification is provided, including:

[0006] Step 1, obtain multi-polarization SAR data of the target area at different times and perform preprocessing;

[0007] Step 2, convert the pixel values of the preprocessed image data into the backscattering intensity of the actual ground objects and extract the scattering matrix S;

[0008] Step 3: Obtain the covariance matrix C2 according to the scattering matrix S;

[0009] Step 4: Decompose the covariance matrix C2 by entropy / equilibrium degree / principal polarization angle to obtain the polarization scattering parameters of the multi-polarization SAR target decomposition;

[0010] Step 6: Select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area, and divide the sample points into a test set and a training set;

[0011] Step 9: Classify the salt marsh vegetation using the random forest classification method to obtain the refined classification result of the salt marsh vegetation.

[0012] Preferably, Step 1 includes:

[0013] Step 1.1: Obtain multi-polarization SAR data of the target area at different times;

[0014] Step 1.2: Perform orbit correction on the multi-polarization SAR data, and replace the orbit information in the metadata to improve the geometric positioning accuracy;

[0015] Step 1.3: Perform radiometric calibration on the data after orbit correction to obtain the backscattering intensity data and phase information of the multi-polarization SAR;

[0016] Step 1.4: Perform terrain correction on the data after radiometric calibration based on terrain data;

[0017] Step 1.5: Perform denoising processing on the data after terrain correction using the adaptive filtering method.

[0018] Preferably, in Step 1.3, the radiometric calibration is achieved by complex calibration, and the backscattering intensity is expressed as:

[0019]

[0020] where γ i is the backscattering intensity of pixel i; P i is the complex amplitude of pixel i; B i is the scattering correction factor of pixel i;

[0021] The phase information is the delay of the wave propagating from the scatterer to the antenna, and the phase information is expressed as:

[0022]

[0023] where is the complex field strength of the scattered wave; is the complex field strength of the incident wave; d is the distance between the scatterer and the receiving antenna; k is the wave number of the incident wave; Represents the attenuation of the signal with distance; e is the base of the natural logarithm; j is the imaginary unit.

[0024] Preferably, in step 1.4, the distance-Doppler correction is performed on the radiometrically calibrated data, and the backscattering intensity after correction is γ cor , and the calculation formula is:

[0025]

[0026] where P i is the complex amplitude of pixel i, and D i is the terrain correction factor obtained from the SRTM data.

[0027] Preferably, in step 1.5, the adaptive filtering method is the Refined Lee filtering method, and the pixel value after filtering is expressed as:

[0028] γ filtered = γ mean + K * (γ cor - γ mean )

[0029] where γ filtered represents the pixel value after filtering; γ mean represents the local mean within the sliding window; K represents the dynamically adjusted weight for controlling the filtering intensity.

[0030] Preferably, step 2 includes:

[0031] Step 2.1, perform decibel conversion on the preprocessed SAR remote sensing image data to convert the unitless backscattering intensity into decibels, expressed as:

[0032]

[0033] where is the backscattering intensity of the ground object after decibel conversion;

[0034] Step 2.2, extract the polarization scattering matrix S based on the preprocessed SAR remote sensing image data, expressed as:

[0035]

[0036] S vv represents the combination of vertical polarization transmission and vertical polarization reception, representing the vertical polarization scattering intensity; S vh represents the scattering coefficient when the vertical polarization wave V is incident and the horizontal polarization wave H is received; represents the complex conjugate of S vh ; 0 indicates that the scattering coefficient of the horizontal polarization wave H incident and other polarization waves received is not included in this model and is filled with 0.

[0037] Preferably, in step 3, the covariance matrix C2 is used to describe the statistical relationship of the scattering characteristics in the dual-polarization data, and the calculation formula of the covariance matrix C2 is:

[0038]

[0039] where <|S vv | 2 +|S vh | 2 > represents the total scattering intensity of the VV and VH channels, that is, the overall energy of the signal; and are the covariance between the scattering intensities of the VV and VH channels, reflecting the coherence between different polarization channels; |S vh | 2 represents the scattering intensity of the VH channel.

[0040] In a second aspect, a salt marsh vegetation extraction system for multi-polarization SAR target decomposition and random forest classification is provided, which is used to execute the method described in any one of the first aspects, including:

[0041] A first acquisition module, configured to acquire multi-polarization SAR data of a target area at different times and perform preprocessing;

[0042] A conversion module, configured to convert the pixel values of the preprocessed image data into the backscattering intensity of the actual ground object and extract the scattering matrix S;

[0043] A second acquisition module, configured to acquire the covariance matrix C2 according to the scattering matrix S;

[0044] A decomposition module, configured to perform entropy / equilibrium degree / principal polarization angle decomposition on the covariance matrix C2 to obtain polarization scattering parameters for multi-polarization SAR target decomposition;

[0045] A selection module, configured to select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area and divide the sample points into a test set and a training set;

[0046] A classification module, configured to classify the salt marsh vegetation by using a random forest classification method to obtain a refined classification result of the salt marsh vegetation.

[0047] In a third aspect, a computer storage medium is provided, in which a computer program is stored; when the computer program runs on a computer, the computer is made to execute the method described in any one of the first aspects.

[0048] In a fourth aspect, an electronic device is provided, including:

[0049] A memory for storing computer programs;

[0050] A processor for executing the computer programs to implement the method according to any one of the first aspects.

[0051] The beneficial effects of the present invention are:

[0052] 1) Considering the high spatio-temporal variability characteristics of the polarimetric scattering of salt marsh vegetation, the high spatial heterogeneity of the distribution of salt marsh vegetation, and the need for fine classification, the present invention proposes a target decomposition method for multi-polarimetric SAR data, obtains multi-polarimetric SAR target decomposition parameters, and further combines random forest classification to achieve fine identification of salt marsh vegetation in coastal wetlands by SAR.

[0053] 2) The present invention first preprocesses multi-polarimetric SAR remote sensing data to obtain the backscattering intensity data of multi-polarimetric SAR and the scattering matrix S, and calculates the covariance matrix C2 for polarimetric decomposition through the scattering matrix. Based on the covariance matrix C2, H / A / Alpha decomposition is performed to obtain entropy (Entropy, H), anisotropy (Anistropy, A), and the main polarization angle (Alpha, α). Further, based on the differences in the backscattering intensity and polarimetric decomposition parameters of each land cover type, the random forest classification method is used to finely identify the salt marsh vegetation in coastal wetlands, so as to obtain the spatio-temporal distribution characteristics of the salt marsh vegetation and provide a scientific basis for the monitoring and protection of coastal wetland ecosystems. Description of the Drawings

[0054] Figure 1 is a flowchart of a method for target decomposition and salt marsh vegetation classification based on multi-polarimetric SAR according to the present invention;

[0055] Figure 2 is a distribution map of samples in an embodiment of the present invention;

[0056] Figure 3 is a polarimetric decomposition parameter image of a multi-polarimetric SAR remote sensing image on May 8, 2022 in an embodiment of the present invention;

[0057] Figure 4 is a polarimetric decomposition parameter image of a multi-polarimetric SAR remote sensing image on September 5, 2022 in an embodiment of the present invention;

[0058] Figure 5a is an upper threshold distribution map of Sentinel-1 polarization parameters on May 8, 2022 in an embodiment of the present invention;

[0059] Figure 5b is an upper threshold distribution map of Sentinel-1 polarization parameters on September 5, 2022 in an embodiment of the present invention;

[0060] Figure 6It is the result map of salt marsh vegetation classification implemented by using random forest. Specific implementation mode

[0061] The present invention will be further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several modifications can still be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0062] Embodiment 1:

[0063] The multi-polarization SAR target decomposition method has greater data advantages and technical potential to achieve fine identification and classification of salt marsh vegetation in coastal wetlands. However, different from the conventional wetland vegetation classification, the fine classification of salt marsh vegetation by multi-polarization SAR requires that the multi-polarization SAR target decomposition technology not only realizes the detailed distinction of different vegetation types, but also can accurately determine the spatial range of the distribution of different vegetation types. Therefore, considering the high spatio-temporal variability characteristics of salt marsh vegetation polarization scattering, the high spatial heterogeneity of salt marsh vegetation distribution and the requirements of fine classification, the present invention proposes a target polarization decomposition method for multi-polarization SAR data, obtains multi-polarization SAR target decomposition parameters, and further combines random forest classification to achieve fine SAR identification of salt marsh vegetation in coastal wetlands.

[0064] Specifically, Embodiment 1 of the present application provides a method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification, as Figure 1 shown, including:

[0065] Step 1, obtain multi-polarization SAR data of the target area at different times and perform preprocessing.

[0066] Step 1 includes:

[0067] Step 1.1, obtain multi-polarization SAR data of the target area at different times.

[0068] Exemplarily, the target area of the present application is the south bank of a certain bay, and multi-polarization SAR data at different times of the same year are collected.

[0069] Step 1.2, perform orbit correction on the multi-polarization SAR data, and replace the orbit information in the metadata to improve the geometric positioning accuracy.

[0070] Specifically, perform orbit correction processing on the multi-polarization SAR remote sensing image data in the SNAP software, and use accurate orbit data to replace the preliminary orbit information in the satellite image metadata, so as to improve the geometric positioning accuracy of the image. The principle is to recalculate the geometric relationship between the sensor and the earth's surface by replacing the satellite position and velocity data, and eliminate the influence of orbit errors on the geometric structure of the image.

[0071] Step 1.3: Perform radiometric calibration on the orbit-corrected data to obtain the backscattering intensity data and phase information of the multi-polarization SAR.

[0072] In Step 1.3, the radiometric calibration is achieved through complex calibration, that is, calibrating the real and imaginary parts of the complex number to obtain the backscattering intensity data and phase information of the multi-polarization SAR. The backscattering intensity is expressed as:

[0073]

[0074] where γ i is the backscattering intensity of pixel i; P i is the complex amplitude of pixel i; B i is the scattering correction factor of pixel i;

[0075] The phase information is the delay of the wave propagating from the scatterer to the antenna, and the phase information is expressed as:

[0076]

[0077] where is the complex field strength of the scattered wave; is the complex field strength of the incident wave; d is the distance between the scatterer and the receiving antenna; k is the wave number of the incident wave; represents the attenuation of the signal with distance; e is the natural base; j is the imaginary unit.

[0078] Step 1.4: Perform terrain correction on the radiometrically calibrated data based on the terrain data.

[0079] Specifically, use the SRTM data as the terrain data to perform Range-Doppler Terrain Correction on the radiometrically calibrated SAR remote sensing image data. The corrected pixel value is γ cor , and the calculation formula is:

[0080]

[0081] where γ cor is the backscattering intensity after terrain correction, P i is the complex amplitude of pixel i, and D i is the terrain correction factor obtained from the SRTM data.

[0082] Step 1.5: Use the adaptive filtering method to denoise the terrain-corrected data.

[0083] Specifically, the Refined Lee filtering method is adopted to filter the terrain-corrected SAR data. The filtering intensity is reduced in areas with strong target structures and increased in areas with strong speckle, thereby reducing the influence of speckle. Specifically, the filter works with a window size of 7×7. The local mean and variance are calculated within each window, and the filtering weights are determined based on the local mean and variance to smooth the noise while retaining the target edges. The filtered pixel value is expressed as:

[0084] γ filtered =γ mean +K*(γ cor -γ mean )

[0085] where γ filtered represents the filtered pixel value; γ mean represents the local mean within the sliding window; K represents the dynamically adjusted weight used to control the filtering intensity.

[0086] Step 2: Convert the pixel values of the preprocessed image data into the backscattering intensity of the actual ground objects and extract the scattering matrix S.

[0087] Step 3: Obtain the covariance matrix C2 according to the scattering matrix S.

[0088] Step 4: Perform entropy / equilibrium degree / principal polarization angle decomposition on the covariance matrix C2 to obtain the polarization scattering parameters of the multi-polarization SAR target decomposition.

[0089] Step 5: Select sample points of salt marsh vegetation and non-salt marsh vegetation within the target area, and divide the sample points into a test set and a training set.

[0090] Step 6: Use the random forest classification method to classify the salt marsh vegetation and obtain the refined classification result of the salt marsh vegetation.

[0091] Example 2:

[0092] Based on Example 1, Example 2 of the present application provides a more specific method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification, including:

[0093] Step 1: Obtain multi-polarization SAR data of the target area at different times and perform preprocessing.

[0094] Step 2: Convert the pixel values of the preprocessed image data into the backscattering intensity of the actual ground objects and extract the scattering matrix S.

[0095] Step 2 includes:

[0096] Step 2.1: Decibelize the preprocessed SAR remote sensing image data, convert the unitless backscattering intensity to decibels, and express it as:

[0097]

[0098] where is the backscattering intensity of the ground object after decibelization;

[0099] Step 2.2: Extract the polarization scattering matrix S based on the preprocessed SAR remote sensing image data. The scattering matrix reflects the physical structure, material, shape, azimuth angle and other characteristics of the ground object target.

[0100] The formula for the polarization scattering matrix S in Step 2.2 is:

[0101]

[0102] S vv represents the combination of vertical polarization transmission and vertical polarization reception, representing the vertical polarization scattering intensity; S vh represents the scattering coefficient when the vertical polarization wave V is incident and the horizontal polarization wave H is received; represents the vh complex conjugate of S; 0 indicates that the scattering coefficient of the horizontal polarization wave H incident and other polarization wave receptions is not included in this model and is filled with 0.

[0103] Step 3: Obtain the covariance matrix C2 according to the scattering matrix S.

[0104] In Step 3, the covariance matrix C2 is used to describe the statistical relationship of the scattering characteristics in the dual-polarization data. The calculation formula of the covariance matrix C2 is:

[0105]

[0106] where <|S vv | 2 +|S vh | 2 > represents the total scattering intensity of the VV and VH channels, that is, the total energy of the signal; and are the covariance between the scattering intensities of the VV and VH channels, reflecting the coherence between different polarization channels; |S vh | 2 represents the scattering intensity of the VH channel.

[0107] Step 4: Decompose the covariance matrix C2 by entropy / equilibrium degree / principal polarization angle to obtain the polarization scattering parameters of the multi-polarization SAR target decomposition.

[0108] In step 4, the embodiments of the present application decompose the target scattering into three main parameters: entropy (Entropy, H), anisotropy (Anistropy, A), and main polarization angle (Alpha, α), which respectively describe the randomness of the scattering mechanism, the relative contribution of the scattering mechanism, and the type of the main scattering mechanism. The polarization decomposition parameter images on different dates are as Figure 3 , 4 shown.

[0109] Among them, H is a parameter that measures the randomness or uncertainty of the scattering process. The lower the value of H, the more deterministic the scattering process is, usually a single scattering mechanism, such as specular reflection or even-order scattering; the higher the value of H, the more random the scattering process is, including multiple scattering mechanisms such as volume scattering and complex multiple scattering. A measures the contrast between the intensity of the secondary scattering mechanism and the intensity of the main scattering mechanism. The lower the value of A, the more dominant the main scattering mechanism is and the weaker the secondary scattering mechanism is; the higher the value of A, the stronger the secondary scattering mechanism is and the contributions of the main and secondary scattering mechanisms are close. α characterizes the main scattering mechanism of the target. 0°≤α≤45° indicates that the scattering mechanism is surface scattering, 45°≤α≤60° indicates that the scattering mechanism is volume scattering, and 60°≤α≤90° indicates that the scattering mechanism is even-order scattering or multiple scattering.

[0110] Step 5: Select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area, and divide the sample points into a test set and a training set.

[0111] Specifically, the sample points of salt marsh vegetation are selected from typical salt marsh vegetation types in the study area, and the non-salt marsh vegetation is selected from two types of ground objects, roads and water bodies. The number of sample points for each type of ground object is not less than 60, and the data of the sample points are divided into a training set and a test set by using the cross-validation method. The specific distribution of the sample points is as Figure 2 shown.

[0112] Step 6: Use the random forest classification method to classify the salt marsh vegetation and obtain the refined classification result of the salt marsh vegetation.

[0113] Specifically, the backscattering intensity of the multi-polarization SAR remote sensing image data at different times and the polarization parameter characteristics, namely Entropy, Anistropy, and Alpha, are used as input variables. Through random sampling with replacement, these input variables are combined into multiple subsets for training multiple decision trees. In each tree, a part of the features are randomly selected for node splitting, so as to improve the diversity and anti-interference ability of the model. After each tree independently predicts, the final classification result is determined by majority voting, and finally the classification label of the target ground object (such as Spartina alterniflora, Scirpus triqueter, etc. of the salt marsh vegetation) is output. Here, the number of trees in the random forest is set to 100, and the random seed is set to 42. The classification result is as Figure 6 shown.

[0114] It should be noted that the parts that are the same as or similar to those in Embodiment 1 in this embodiment can be referred to each other, and will not be elaborated in this application.

[0115] Embodiment 3:

[0116] Based on Embodiments 1 and 2, Embodiment 3 of this application provides a salt marsh vegetation extraction system for multi-polarization SAR target decomposition and random forest classification, including:

[0117] A first acquisition module, configured to acquire multi-polarization SAR data of a target area at different times and perform preprocessing;

[0118] A conversion module, configured to convert the pixel values of the preprocessed image data into the backscattering intensity of actual ground objects and extract the scattering matrix S;

[0119] A second acquisition module, configured to acquire a covariance matrix C2 according to the scattering matrix S;

[0120] A decomposition module, configured to perform entropy / equilibrium degree / principal polarization angle decomposition on the covariance matrix C2 to obtain polarization scattering parameters for multi-polarization SAR target decomposition;

[0121] A selection module, configured to select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area and divide the sample points into a test set and a training set;

[0122] A classification module, configured to classify salt marsh vegetation by using a random forest classification method to obtain a refined classification result of salt marsh vegetation.

[0123] It should be noted that the system provided in this embodiment is the system corresponding to the methods provided in Embodiments 1 and 2. Therefore, the parts that are the same as or similar to those in Embodiments 1 and 2 in this embodiment can be referred to each other, and will not be elaborated in this application.

Claims

1. A salt marsh vegetation extraction method based on multi-polarization SAR target decomposition and random forest classification, characterized in that: include: Step 1: Obtain multi-polarization SAR data of the target area at different times and perform preprocessing; Step 2: Convert the pixel value of the preprocessed image data into the backscattering intensity of the actual ground object And extract the scattering matrix S; Step 3: Obtain the covariance matrix C2 according to the scattering matrix S; Step 4, perform entropy / balance / main polarization angle decomposition on the covariance matrix C2 to obtain polarization scattering parameters of multi-polarization SAR target decomposition; Step 5: Select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area, and divide the sample points into a test set and a training set; Step 6: Use the random forest classification method to classify the salt marsh vegetation and obtain refined classification results of the salt marsh vegetation.

2. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 1, characterized in that: Step 1 includes: Step 1.1, obtaining multi-polarization SAR data of the target area at different times; Step 1.2: Perform orbit correction on multi-polarization SAR data and replace the orbit information in metadata to improve geometric positioning accuracy; Step 1.3, perform radiation calibration on the orbit-corrected data to obtain backscatter intensity data and phase information of the multi-polarization SAR; Step 1.4, performing terrain correction on the radiometrically calibrated data based on terrain data; Step 1.5: Use adaptive filtering method to denoise the terrain corrected data.

3. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 2, characterized in that: In step 1.3, radiation calibration is achieved through complex calibration, and the backscatter intensity is expressed as: Among them, γ i is the backscattering intensity of pixel i; P i is the complex amplitude of pixel i; B i is the scatter correction factor of pixel i; The phase information is the delay of the wave propagating from the scatterer to the antenna. The phase information is expressed as: in, is the complex field intensity of the scattered wave; is the complex field intensity of the incident wave; d is the distance between the scatterer and the receiving antenna; k is the wave number of the incident wave; Represents the attenuation of the signal with distance; e is the natural base; j is the imaginary unit.

4. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 3, characterized in that: In step 1.4, the data after radiation calibration is subjected to range Doppler correction, and the corrected backscatter intensity is γ cor , the calculation formula is: Among them, P i is the complex amplitude of pixel i, D i is the terrain correction factor obtained from the SRTM data.

5. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 4, characterized in that: In step 1.5, the adaptive filtering method is the Refined Lee filtering method, and the pixel value after filtering is expressed as: c filered =c mean +K*(γ cor -c mean ) Among them, γ filtered Represents the pixel value after filtering; γ mean represents the local mean within the sliding window; K represents the dynamically adjusted weight used to control the filtering strength.

6. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 5, characterized in that: Step 2 includes: Step 2.1: Convert the preprocessed SAR remote sensing image data into decibels, and convert the unitless backscatter intensity into decibels, expressed as: in, is the ground object backscattering intensity after decibelization; Step 2.2: Extract the polarization scattering matrix S based on the preprocessed SAR remote sensing image data, expressed as: S vv Represents the combination of vertical polarization transmission and vertical polarization reception, representing the vertical polarization scattering intensity; S vh It represents the scattering coefficient when vertical polarization wave V is incident and horizontal polarization wave H is received; Indicates S vh Complex conjugate; 0 means that the scattering coefficients of horizontally polarized wave H incident and other polarized waves received are not included in this model and are filled with 0. .

7. The method for extracting salt marsh vegetation by multi-polarization SAR target decomposition and random forest classification according to claim 6, characterized in that: In step 3, the covariance matrix C2 is used to describe the statistical relationship of the scattering characteristics in the dual-polarization data. The calculation formula of the covariance matrix C2 is: Among them, <|S vv | 2 +|S vh | 2 > represents the total scattering intensity of the VV and VH channels, that is, the overall energy of the signal; and is the covariance between the scattering intensities of the VV and VH channels, reflecting the coherence between channels with different polarizations; |S vh | 2 Represents the scattering intensity of the VH channel.

8. Salt marsh vegetation extraction system based on multi-polarization SAR target decomposition and random forest classification, characterized by: Used to perform the method according to any one of claims 1 to 7, comprising: The first acquisition module is used to acquire multi-polarization SAR data of the target area at different times and perform preprocessing; The conversion module is used to convert the pixel value of the preprocessed image data into the backscattering intensity of the actual ground object. And extract the scattering matrix S; A second acquisition module, used for acquiring a covariance matrix C2 according to the scattering matrix S; A decomposition module is used to perform entropy / balance / main polarization angle decomposition on the covariance matrix C2 to obtain polarization scattering parameters of multi-polarization SAR target decomposition; A selection module is used to select sample points of salt marsh vegetation and non-salt marsh vegetation in the target area, and divide the sample points into a test set and a training set; The classification module is used to classify salt marsh vegetation using the random forest classification method to obtain refined classification results of salt marsh vegetation.

9. A computer storage medium, characterized in that: The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 7.

10. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 7.

Images