Snow identification method and device based on full polarization SAR, equipment and storage medium

Abstract

Embodiments of the present application provide a snow identification method and device based on full polarization SAR, equipment and storage medium, relating to the technical field of snow identification. The snow identification method comprises: S1, acquiring full polarization Radarsat-2 image of a target area. S2, extracting four backscatter characteristics through polarization. S3, acquiring a first difference value and a first ratio value according to the VV backscatter characteristic and the VH backscatter characteristic. S6, acquiring three polarization characteristics through Pauli decomposition. S7, acquiring three polarization characteristics through Freeman decomposition. S8, acquiring four polarization characteristics through decomposition. S9, acquiring four polarization characteristics through Yamaguchi decomposition. S10, combining the above-mentioned 20 characteristics to obtain a multi-band characteristic image composed of 20 bands. S11, identifying using a random forest classifier according to the multi-band characteristic image composed of 20 bands to obtain a snow identification result. By combining 20 characteristics, especially the first difference value and the first ratio value, the fragmentation of the snow identification result can be greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of snow cover recognition technology, and more specifically, to a snow cover recognition method, apparatus, device, and storage medium based on fully polarimetric SAR. Background Technology

[0002] Snow cover is one of the most extensive and active elements in the cryosphere, playing a significant role in global climate change, the ecological environment, and socio-economic development. In recent years, with global warming, the duration and extent of global snow cover have decreased significantly. Accurate snow cover identification and monitoring are crucial for climate change assessment, disaster assessment, dynamic monitoring of snowmelt, and ecological security.

[0003] Snow cover is mostly distributed in high-altitude, high-latitude mountainous areas, making data acquisition difficult using traditional snow cover research methods. The emergence of remote sensing technology has provided new possibilities for large-scale, high-precision snow cover identification and monitoring. Compared with optical remote sensing, SAR has the ability to penetrate clouds and fog, monitor snow cover around the clock and in all weather conditions, and has partial penetration capabilities, making it widely used in the field of snow cover remote sensing.

[0004] However, in the prior art, the snow accumulation recognition results are relatively fragmented and the accuracy is low. In view of this, the applicant hereby submits this application after studying the existing technology. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for snow cover identification based on fully polarimetric SAR, in order to improve the above-mentioned technical problems.

[0006] First aspect

[0007] This invention provides a snow cover identification method based on fully polarimetric SAR, which includes steps S1 to S2.

[0008] S1. Acquire the fully polarized Radarsat-2 image of the target region.

[0009] S2. Based on the fully polarized Radarsat-2 image, extract the backscattering coefficients and obtain the HH backscattering characteristics, HV backscattering characteristics, VH backscattering characteristics, and VV backscattering characteristics.

[0010] S3. Based on the VV backscattering characteristics and the VH backscattering characteristics, obtain the first difference and the first ratio. The first difference is the difference between the VV backscattering characteristics and the VH backscattering characteristics. The first ratio is the ratio between the VV backscattering characteristics and the VH backscattering characteristics.

[0011] S6. Based on the fully polarized Radarsat-2 image, obtain the Pauli_Dbl polarization features, Pauli_Odd polarization features, and Pauli_Vol polarization features through Pauli decomposition.

[0012] S7. Based on the fully polarized Radarsat-2 image, obtain the Freeman_Dbl polarization features, Freeman_Odd polarization features, and Freeman_Vol polarization features through Freeman decomposition.

[0013] S8. Based on the fully polarized Radarsat-2 imagery, via HA- Decompose the data to obtain Entropy polarization features, Anisotropy polarization features, Alpha polarization features, and Lambda polarization features.

[0014] S9. Based on the fully polarimetric Radarsat-2 image, obtain the polarization features of Yamaguchi4_Dbl, Yamaguchi4_Odd, Yamaguchi4_Vol, and Yamaguchi4_Hlx through Yamaguchi decomposition.

[0015] S10. Based on the backscattering characteristics of HH, HV, VH, and VV, the first difference, the first ratio, the polarization characteristics of Pauli_Dbl, Pauli_Odd, Pauli_Vol, Freeman_Dbl, Freeman_Odd, Freeman_Vol, Entropy, Anisotropy, Alpha, Lambda, Yamaguchi4_Dbl, Yamaguchi4_Odd, Yamaguchi4_Vol, and Yamaguchi4_Hlx, band combinations are performed to obtain a multi-band feature image composed of 20 bands.

[0016] S11. Based on the multi-band feature image composed of 20 bands, a random forest classifier is used for identification to obtain snow cover identification results.

[0017] The second aspect

[0018] This invention provides a snow cover identification device based on fully polarimetric SAR, comprising:

[0019] The initial image acquisition module is used to acquire fully polarized Radarsat-2 images of the target region.

[0020] The first feature acquisition module is used to extract backscattering coefficients from fully polarized Radarsat-2 images and obtain HH backscattering features, HV backscattering features, VH backscattering features and VV backscattering features.

[0021] The second feature acquisition module is used to obtain a first difference and a first ratio based on the VV backscattering feature and the VH backscattering feature. The first difference is the difference between the VV backscattering feature and the VH backscattering feature. The first ratio is the ratio between the VV backscattering feature and the VH backscattering feature.

[0022] The third feature acquisition module is used to obtain Pauli_Dbl polarization features, Pauli_Odd polarization features, and Pauli_Vol polarization features from the fully polarized Radarsat-2 image through Pauli decomposition.

[0023] The fourth feature acquisition module is used to obtain Freeman_Dbl polarization features, Freeman_Odd polarization features, and Freeman_Vol polarization features from the fully polarized Radarsat-2 image through Freeman decomposition.

[0024] The fifth feature acquisition module is used to acquire features from fully polarimetric Radarsat-2 images via HA- Decompose the data to obtain Entropy polarization features, Anisotropy polarization features, Alpha polarization features, and Lambda polarization features.

[0025] The sixth feature acquisition module is used to acquire the Yamaguchi4_Dbl polarization features, Yamaguchi4_Odd polarization features, Yamaguchi4_Vol polarization features, and Yamaguchi4_Hlx polarization features based on the fully polarized Radarsat-2 image through Yamaguchi decomposition.

[0026] The feature combination module is used to combine bands based on HH backscattering features, HV backscattering features, VH backscattering features, VV backscattering features, first difference, first ratio, Pauli_Dbl polarization features, Pauli_Odd polarization features, Pauli_Vol polarization features, Freeman_Dbl polarization features, Freeman_Odd polarization features, Freeman_Vol polarization features, Entropy polarization features, Anisotropy polarization features, Alpha polarization features, Lambda polarization features, Yamaguchi4_Dbl polarization features, Yamaguchi4_Odd polarization features, Yamaguchi4_Vol polarization features, and Yamaguchi4_Hlx polarization features to obtain a multi-band feature image composed of 20 bands.

[0027] The snow cover recognition module is used to identify snow cover based on a multi-band feature image composed of 20 bands using a random forest classifier.

[0028] Third aspect

[0029] This invention provides a snow cover identification device based on fully polarimetric SAR, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement the snow cover identification method based on fully polarimetric SAR as described in any paragraph of the first aspect.

[0030] Fourth aspect

[0031] This invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the snow cover identification method based on fully polarimetric SAR as described in any paragraph of the first aspect.

[0032] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0033] By combining 20 features, especially the first difference and the first ratio, the fragmentation of snow cover recognition results can be greatly reduced, resulting in better recognition performance, which has great practical significance.

[0034] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic flowchart of the snow cover recognition method provided in the first embodiment of the present invention.

[0037] Figure 2 This is Radarsat-2 imagery of the Manas River basin.

[0038] Figure 3 yes Figure 2 Radarsat backscattering coefficient diagram.

[0039] Figure 4 These are the results of snow cover recognition based on different combinations of features.

[0040] Figure 5 This is a comparison chart of classification accuracy.

[0041] Figure 6 This is a schematic diagram of a snow accumulation recognition device provided in the second embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0044] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0045] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0046] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0047] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0048] The use of "first" and "second" in the embodiments is merely to distinguish similar objects and does not represent a specific ordering of objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0049] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0050] Example 1:

[0051] Please see Figures 1 to 5 The first embodiment of the present invention provides a snow cover identification method based on fully polarimetric SAR, which can be executed by a snow cover identification device. In particular, it is executed by one or more processors in the snow cover identification device to implement steps S1 to S11.

[0052] S1. Acquire the fully polarized Radarsat-2 image of the target region.

[0053] It is understood that Radarsat-2 imagery is an image captured by a synthetic aperture radar (SAR) system. Specifically, this invention uses fully polarized Radarsat-2 imagery as the data source, with the imaging band being the C-band, the imaging mode being fine full polarization (including four polarization modes: HH, HV, VH, and VV), and the product mode being single-view complex data (SLC).

[0054] It should be noted that the snow recognition device can be an electronic device with computing power, such as a portable laptop computer, desktop computer, server, smartphone, or tablet computer.

[0055] S2. Based on the fully polarized Radarsat-2 image, extract the backscattering coefficients and obtain the HH backscattering characteristics, HV backscattering characteristics, VH backscattering characteristics, and VV backscattering characteristics.

[0056] In an optional embodiment, step S2 specifically includes:

[0057] Based on the fully polarimetric Radarsat-2 image, preprocessing was performed including multi-view processing, filtering, geocoding, and radiometric calibration. Noise reduction was performed using Refind Lee filtering with a window size of 5*5. Then, backscattering coefficients were extracted to obtain HH, HV, VH, and VV backscattering features.

[0058] Specifically, backscattering feature extraction is crucial, as the backscattering coefficients of different polarization modes play an important role in image classification. Therefore, four backscattering features—HH, HV, VH, and VV—were extracted as input features for snow cover classification.

[0059] S3. Obtain the first difference based on the backscattering characteristics of VV and VH. The ratio to the first The first difference is the difference between the VV backscattering characteristics and the VH backscattering characteristics. The first ratio is the ratio between the VV backscattering characteristics and the VH backscattering characteristics.

[0060] Through extensive creative research, the inventors discovered that in prior art, snow identification is achieved by using only a single feature of SAR images. However, due to the influence of speckle noise in SAR images, the measurement of the snow scattering coefficient is greatly deviated, and some snow information is also lost during the filtering process, resulting in fragmented snow identification results.

[0061] Therefore, through extensive and creative research, the inventors proposed two new parameters by processing the cross-polarization and co-polarization backscattering coefficients to improve the fragmented snow cover recognition results in prior art. Specifically, the two parameters are the first difference... The ratio to the first Among them, the first difference The ratio to the first The calculation formula is:

[0062] (1)

[0063] (2)

[0064] In the formula, Let V be the backscattering coefficient. VH is the backscattering coefficient.

[0065] S4. Obtain the polarization scattering matrix based on the fully polarized Radarsat-2 image.

[0066] S5. Convert the polarization scattering matrix into a polarization coherence matrix, and then use Refind Lee filtering to denoise the coherence matrix.

[0067] S6. Based on the fully polarized Radarsat-2 image, obtain the Pauli_Dbl polarization features, Pauli_Odd polarization features, and Pauli_Vol polarization features through Pauli decomposition.

[0068] Specifically, the Pauli decomposition decomposes the scattering matrix S into a weighted sum of complex forms of the Pauli basis matrices. Each Pauli basis matrix corresponds to a basic scattering mechanism, as shown in the following formula:

[0069] (3)

[0070] S7. Based on the fully polarized Radarsat-2 image, obtain the Freeman_Dbl polarization features, Freeman_Odd polarization features, and Freeman_Vol polarization features through Freeman decomposition.

[0071] Specifically, Freeman decomposition, based on Van Zyl, decomposes the polarization covariance matrix into three scattering mechanisms: surface scattering, volume scattering, and dihedral scattering. This method, grounded in physical reality, models these three fundamental scattering mechanisms separately, providing a good description of natural scatterers. The polarization covariance matrix can be expressed as:

[0072] (4)

[0073] S8. Based on the fully polarized Radarsat-2 imagery, via HA- Decompose the data to obtain Entropy polarization features, Anisotropy polarization features, Alpha polarization features, and Lambda polarization features.

[0074] Specifically, HA- The decomposition, also known as the Cloude-Pottier decomposition, decomposes the target into a weighted sum of three components based on the eigenvalue and eigenvector decomposition methods of the polarization coherence matrix. It then further calculates the dominant scattering component in the target's scattered echo. This method first transforms the scattering matrix [S] into a coherence matrix [T], and then calculates the eigenvalues. And the eigenvector U, the formula is as follows:

[0075] （5）

[0076] The polarization scattering entropy H, scattering anisotropy A, and average scattering angle are defined using the obtained eigenvalues ​​and eigenvectors. The polarization scattering entropy (H) is typically used to describe the degree of randomness between isotropic and completely random scattering of different scattering types of targets; the scattering anisotropy (A) is a complement to the scattering entropy, characterizing the relationship between the second and third eigenvalues; the average scattering angle has four components. This corresponds to the change from surface scattering to volume scattering, and then from volume scattering to dihedral scattering; the mean eigenvalue is the average of these three eigenvalues.

[0077] S9. Based on the fully polarimetric Radarsat-2 image, obtain the polarization features of Yamaguchi4_Dbl, Yamaguchi4_Odd, Yamaguchi4_Vol, and Yamaguchi4_Hlx through Yamaguchi decomposition.

[0078] Specifically, in order to be more widely applicable to cases where the scatterer has a complex geometric scattering structure, the Yamaguchi decomposition introduces a fourth scattering component—spiral scattering—on top of the Freeman decomposition, thus overcoming the limitation of the Freeman decomposition in being unsuitable for asymmetric problems.

[0079] S10. Based on the backscattering characteristics of HH, HV, VH, and VV, the first difference, the first ratio, the polarization characteristics of Pauli_Dbl, Pauli_Odd, Pauli_Vol, Freeman_Dbl, Freeman_Odd, Freeman_Vol, Entropy, Anisotropy, Alpha, Lambda, Yamaguchi4_Dbl, Yamaguchi4_Odd, Yamaguchi4_Vol, and Yamaguchi4_Hlx, band combinations are performed to obtain a multi-band feature image composed of 20 bands.

[0080] S11. Based on the multi-band feature image composed of 20 bands, a random forest classifier is used for identification to obtain snow cover identification results.

[0081] Specifically, the Random Forest (RF) algorithm is an ensemble learning method with decision trees as the basic classifiers. This method uses the Bagging method to extract an independent and identically distributed sample set for each decision tree, trains the sample set to generate decision trees, and determines the final result through decision tree voting. Compared with other machine learning algorithms, the RF algorithm can train a small number of samples to establish classification rules when dealing with high-dimensional data, suppress the occurrence of overfitting, and has good robustness and stability to outliers and noisy data [23, 24]. At the same time, the RF algorithm has the ability of feature optimization. It uses the Out-Of-Bag (OOB) data that is not involved in training as test samples, calculates the OOB misclassification rate of a single decision tree, and estimates the importance of each feature.

[0082] In this study, the number of decision trees was set to 100, and the formula for calculating the number of features of each decision tree is:

[0083] m = sqr(M) (6)

[0084] In the formula, M is the total number of features used for classification.

[0085] In this embodiment, the random forest classifier includes, from the original training sample set with a capacity of N, each time N examples are drawn with replacement from the original sample set by the Bootstraping self-sampling method, and a total of k draws are made to generate k training sets (D1, D2,..., Dk); from the feature space with a feature size of M, m features (m < M) are randomly selected to form a new feature set, where m = sqr(M). A total of k decision trees are generated in the k training sets, and each decision tree is independent of each other. The generated multiple decision trees are combined into a random forest; the final classification result is determined according to the voting of the k decision trees.

[0086] In an optional embodiment, step S10 specifically includes:

[0087] S101. Obtain the original training sample set according to the multi-band feature image composed of 20 bands. Among them, the original training sample set contains 55 snow samples and 66 non-snow samples.

[0088] S102. According to the original training sample set, randomly select a sample with replacement by the Bootstraping method and put it into the training set, and repeat the sampling 100 times to obtain 100 training sets. Among them, each training set contains 20 features.

[0089] Specifically, the original training sample set contains 121 samples (55 snow samples and 66 non-snow samples). From the original training sample data, a sample is randomly selected each time using the Bootstraping method and placed into the training set Di. Then, the sample is put back into the original training sample set so that the sample will still have a chance to be selected in the next sampling. After repeating the sampling 100 times, 100 training sets of the same size are obtained.

[0090] S103. Randomly select from the 100 training sets respectively. Each feature is used to form a new feature set. By calculating the information content of the features, the optimal feature is selected to split the node and generate a decision tree, thereby obtaining 100 decision tree models. These 100 decision trees are then combined into a random forest.

[0091] Specifically, the input feature set contains 20 features. √20 features are randomly selected from these 20 to form a new feature set. By calculating the information content of each feature, the optimal feature is selected to split the nodes, generating a decision tree. 100 decision tree models are constructed using 100 extracted training sets, and these decision trees are then combined to form a random forest.

[0092] S104. Classify and identify the multi-band feature image composed of 20 bands using random forest to obtain snow cover identification results. The identification results include snow-covered areas and non-snow-covered areas.

[0093] Specifically, each decision tree can predict a classification result. Based on the voting results of 100 decision trees, the final snow and non-snow areas of the image to be classified are determined.

[0094] In this embodiment, by combining 20 features, especially the first difference and the first ratio, the fragmentation of the snow cover recognition results can be greatly reduced, resulting in better recognition performance, which has great practical significance.

[0095] To facilitate understanding of the present invention, a specific example is given below to illustrate the application of this embodiment.

[0096] This case study focuses on the Manas River Basin in Xinjiang. Figure 2Located on the northern slopes of the central Tianshan Mountains in Xinjiang, in the southern part of the Junggar Basin (43°20′~45°55′N, 85°00′~87°00′E), the Manas River is the inland river with the largest glacier scale in the Junggar inland area. The terrain within its basin slopes from southeast to northwest, with the highest elevation at 544m and the lowest at 256m. Most areas above 3600m are covered by glaciers and permanent snow cover, with an average snowline of 3970m. The Manas River has a typical mountain-basin structure in an arid region, characterized by low moisture sources, scarce precipitation, and an arid climate, belonging to a typical temperate continental arid and semi-arid climate. Under the influence of continental climate conditions, the snow cover in the basin is characterized by low density, low water content, and thin thickness. The average snow cover reaches its peak in January and February, begins to melt in March, ends in mid-April, and reaches its lowest point in August. Meltwater from glaciers and snow is the main source of runoff for the Manas River, accounting for 47% of its annual runoff, making it an important water source for the northern slope of the Tianshan Mountains.

[0097] The fully polarized Radarsat-2 imagery in this case was acquired on December 13, 2013, with an incident angle of 43.45°, a range pixel of 4.733m, and an azimuth pixel of 4.799m.

[0098] First, four backscattering features were extracted from the fully polarized Radarsat-2 data ( Figure 3 ).

[0099] Then, the first difference and the first ratio are calculated based on the backscattering characteristics.

[0100] Next, the target polarization decomposition features are extracted using four different decomposition methods, resulting in 14 polarization decomposition features. The table below shows the 20 feature parameters extracted in the previous three steps.

[0101] Table 1 Feature parameters based on Radarsat-2 images

[0102]

[0103] Table 2 Classification Feature Set

[0104]

[0105] To verify the effectiveness of the two feature parameters proposed by the inventors after their creative work, this case divides the aforementioned 20 features into three combinations: T1, T2, and T3. The following steps are performed for each combination, and the results of snow accumulation recognition are compared. Feature set T1 includes Pauli decomposition, Freeman decomposition, and HA- Fourteen polarimetric feature parameters derived from the Yamaguchi four-component target polarimetric decomposition were used in feature set T1. These polarimetric features contain rich snow cover information. The purpose of using this feature set was to explore the effectiveness of using only features derived from multiple polarimetric target decompositions for snow cover identification. To explore the contribution of the fully polarimetric backscattering coefficient to snow cover identification, four fully polarimetric SAR backscattering features (HH, HV, VV, and VH) were added to feature set T2, based on feature set T1. Feature set T3 added the difference (Diff) and ratio (Ratio) between VV and VH to feature set T2.

[0106] Finally, the three feature sets T1, T2, and T3 were used as input features for the RF classifier, and the results of snow classification with different feature sets were compared. Figure 4 ).

[0107] For accuracy evaluation, the study referenced GF-1 optical imagery to select and determine snow-covered and non-snow-covered samples, ensuring the sample points were distributed as evenly as possible within the study area. Training samples were selected from the GF-1 imagery, including 55 snow-covered regions and 66 non-snow-covered regions. 200 validation samples were randomly generated from the GF-1 satellite imagery, including 94 snow-covered samples and 106 non-snow-covered samples.

[0108] The accuracy of the classification results in Section 3.2 is evaluated using the confusion matrix to calculate accuracy metrics and the F-measure. Accuracy metrics include Overall Accuracy (OA), Kappa Coefficient, Producer's Accuracy (PA), and User's Accuracy (UA); F-measure represents the harmonic mean of PA and UA, calculated as follows:

[0109] (7)

[0110] From the classification results image ( Figure 4 As can be seen, the T1, T2, and T3 feature sets can basically distinguish between snow-covered and non-snow-covered areas, but differences exist in the identification results in local regions. Classification using only the target polarization feature (T1) is insufficient. Figure 4 In (a), the snow cover area is relatively complete, but small patches appear in the internal areas of the snow cover. Four backscattering features (T2) were added. Figure 4 (b) and the difference (Diff) and ratio (Ratio) between VV and VH (in the text) Figure 4 After step c), the number of patches in each classification result image decreased, and the obtained snow-covered areas were more complete. From the overall recognition results,

[0111] The identification results based on polarization features show that the snow-covered area has a large number of broken patches. Compared with the polarization feature set, the identification results after adding the fully polarized backscattering feature show that the snow-covered area is more complete, but there are still a few small fragments. Adding the difference (Diff) and ratio (Ratio) of VV and VH to the above features results in the best snow recognition effect and reduces the fragmentation of the classification results.

[0112] The accuracy of the classification results was evaluated, and the confusion matrix is ​​shown in Table 3. Accuracy verification was performed as follows: Figure 5 The comparison revealed that, among the classification accuracies of the three feature combinations, using only polarization features (T1) resulted in low classification accuracy for both snow cover and non-snow cover, with some degree of confusion between the two. The accuracy for snow cover and non-snow cover producers was 77.66% and 77.36%, respectively, with F1 scores of 0.76 and 0.78, respectively. The overall classification accuracy was 77.50%, and the Kappa coefficient was only 0.55. Adding four fully polarimetric backscattering features (T2) improved the overall classification accuracy. The accuracy for snow cover and non-snow cover producers increased by 3.19% and 2.83%, respectively, and the F1 score improved by 0.03 for both. The overall classification accuracy increased to 80.50%, and the Kappa coefficient increased to 0.61.

[0113] Table 3. Confusion matrix of classification results

[0114]

[0115] Combining polarization features, backscattering features, and the difference (Diff) and ratio (Ratio) features between VV and VH (T3) for classification, the overall classification accuracy and Kappa coefficient were the highest, at 83.00% and 0.66, respectively. The accuracy for both snow-covered and non-snow-covered producers and users was the highest, with F1 scores exceeding 0.80, an improvement of 0.06 compared to using only polarization features (T1). In summary, comparing the classification results of the three feature sets, feature set T3 showed the highest overall classification accuracy and best performance, followed by feature set T2, while feature set T1 showed the lowest classification accuracy.

[0116] Example 2

[0117] This invention provides a snow cover identification device based on fully polarimetric SAR, comprising:

[0118] Initial image acquisition module 1 is used to acquire fully polarized Radarsat-2 images of the target area.

[0119] The first feature acquisition module 2 is used to extract backscattering coefficients from the fully polarized Radarsat-2 image and obtain HH backscattering features, HV backscattering features, VH backscattering features and VV backscattering features.

[0120] The second feature acquisition module 3 is used to acquire a first difference and a first ratio based on the VV backscattering feature and the VH backscattering feature. The first difference is the difference between the VV backscattering feature and the VH backscattering feature. The first ratio is the ratio between the VV backscattering feature and the VH backscattering feature.

[0121] The third feature acquisition module 4 is used to obtain Pauli_Dbl polarization features, Pauli_Odd polarization features, and Pauli_Vol polarization features from the fully polarized Radarsat-2 image through Pauli decomposition.

[0122] The fourth feature acquisition module 5 is used to obtain Freeman_Dbl polarization features, Freeman_Odd polarization features, and Freeman_Vol polarization features from the fully polarized Radarsat-2 image through Freeman decomposition.

[0123] The fifth feature acquisition module 6 is used to acquire features based on the fully polarimetric Radarsat-2 image via HA- Decompose the data to obtain Entropy polarization features, Anisotropy polarization features, Alpha polarization features, and Lambda polarization features.

[0124] The sixth feature acquisition module 7 is used to acquire the Yamaguchi4_Dbl polarization features, Yamaguchi4_Odd polarization features, Yamaguchi4_Vol polarization features, and Yamaguchi4_Hlx polarization features based on the fully polarized Radarsat-2 image through Yamaguchi decomposition.

[0125] Feature combination module 8 is used to combine bands based on HH backscattering features, HV backscattering features, VH backscattering features, VV backscattering features, first difference, first ratio, Pauli_Dbl polarization features, Pauli_Odd polarization features, Pauli_Vol polarization features, Freeman_Dbl polarization features, Freeman_Odd polarization features, Freeman_Vol polarization features, Entropy polarization features, Anisotropy polarization features, Alpha polarization features, Lambda polarization features, Yamaguchi4_Dbl polarization features, Yamaguchi4_Odd polarization features, Yamaguchi4_Vol polarization features, and Yamaguchi4_Hlx polarization features to obtain a multi-band feature image composed of 20 bands.

[0126] Snow cover recognition module 9 is used to identify snow cover based on a multi-band feature image composed of 20 bands using a random forest classifier to obtain snow cover recognition results.

[0127] In this embodiment, by combining 20 features, especially the first difference and the first ratio, the fragmentation of the snow cover recognition results can be greatly reduced, resulting in better recognition performance, which has great practical significance.

[0128] In an optional embodiment, the first feature acquisition module 2 is specifically used for:

[0129] Based on the fully polarimetric Radarsat-2 image, multi-view processing, filtering, geocoding, and radiometric calibration preprocessing were performed. Refind Lee filtering was used for noise reduction with a window size of 5*5. Then, backscattering coefficients were extracted to obtain HH, HV, VH, and VV backscattering features.

[0130] In an optional embodiment, the snow recognition device further includes a polarization scattering matrix acquisition module and a polarization coherence matrix conversion module.

[0131] The polarization scattering matrix acquisition module is used to acquire the polarization scattering matrix based on the fully polarized Radarsat-2 image.

[0132] The polarization coherence matrix conversion module is used to convert the polarization scattering matrix into a polarization coherence matrix, and then the coherence matrix is ​​denoised using Refind Lee filtering.

[0133] In an optional embodiment, the snow accumulation recognition module 9 includes:

[0134] The training sample acquisition unit is used to acquire the original training sample set based on the multi-band feature image composed of 20 bands. The original training sample set contains 55 snow samples and 66 non-snow samples.

[0135] The training set acquisition unit is used to randomly select a sample with replacement from the original training sample set using the Bootstraping method and add it to the training set. This sampling is repeated 100 times to obtain a training set of 100 samples. Each training set contains 20 features.

[0136] Training units are used to randomly select from 100 training sets. Each feature is used to form a new feature set. By calculating the information content of the features, the optimal feature is selected to split the node and generate a decision tree, thereby obtaining 100 decision tree models. These 100 decision trees are then combined into a random forest.

[0137] The identification unit is used to classify and identify multi-band feature images composed of 20 bands based on random forest to obtain snow cover identification results. These results include snow-covered areas and non-snow-covered areas.

[0138] Example 3

[0139] This invention provides a snow cover identification device based on fully polarimetric SAR, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement the snow cover identification method based on fully polarimetric SAR as described in any paragraph of Embodiment 1.

[0140] Example 4

[0141] This invention provides a computer-readable storage medium, which includes a stored computer program. When the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the snow cover identification method based on fully polarimetric SAR as described in any paragraph of Embodiment 1.

[0142] In the several embodiments provided in this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0143] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0144] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0145] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for snow identification based on full polarimetric SAR, characterized in that, Comprising: acquiring a full polarization Radarsat-2 image of a target area; extracting backscattering coefficients from the full polarization Radarsat-2 image to obtain HH backscattering features, HV backscattering features, VH backscattering features and VV backscattering features; obtaining a first difference value and a first ratio value according to the VV backscattering features and the VH backscattering features; wherein the first difference value is the difference between the VV backscattering features and the VH backscattering features; and the first ratio value is the ratio of the VV backscattering features to the VH backscattering features; obtaining Pauli_Dbl polarization features, Pauli_Odd polarization features and Pauli_Vol polarization features by Pauli decomposition according to the full polarization Radarsat-2 image; obtaining Freeman_Dbl polarization features, Freeman_Odd polarization features and Freeman_Vol polarization features by Freeman decomposition according to the full polarization Radarsat-2 image; According to the full polarimetric Radarsat-2 image, by Decomposition, obtain Entropy polarization features, Anisotropy polarization features, Alpha polarization features and Lambda polarization features; obtaining Yamaguchi4_Dbl polarization features, Yamaguchi4_Odd polarization features, Yamaguchi4_Vol polarization features and Yamaguchi4_Hlx polarization features by Yamaguchi decomposition according to the full polarization Radarsat-2 image; performing band combination according to the HH backscattering features, the HV backscattering features, the VH backscattering features, the VV backscattering features, the first difference value, the first ratio value, the Pauli_Dbl polarization features, the Pauli_Odd polarization features, the Pauli_Vol polarization features, the Freeman_Dbl polarization features, the Freeman_Odd polarization features, the Freeman_Vol polarization features, the Entropy polarization features, the Anisotropy polarization features, the Alpha polarization features, the Lambda polarization features, the Yamaguchi4_Dbl polarization features, the Yamaguchi4_Odd polarization features, the Yamaguchi4_Vol polarization features and the Yamaguchi4_Hlx polarization features to obtain a multi-band feature image composed of 20 bands; performing recognition by using a random forest classifier according to the multi-band feature image composed of 20 bands to obtain a snow identification result.

2. The snow recognition method according to claim 1, characterized by, extracting backscattering coefficients from the full polarization Radarsat-2 image to obtain HH backscattering features, HV backscattering features, VH backscattering features and VV backscattering features according to the full polarization Radarsat-2 image, specifically comprising: According to the full polarization Radarsat-2 image, multi-view processing, filtering processing, geographic coding and radiation scaling preprocessing are performed, Refind Lee filtering is used for noise reduction processing, the window size is set to 5*5, then the backscattering coefficient is extracted, and HH backscattering characteristics, HV backscattering characteristics, VH backscattering characteristics and VV backscattering characteristics are obtained.

3. The snow recognition method according to claim 1, characterized by, Before the Pauli decomposition, the Freeman decomposition, decomposition, and the Yamaguchi decomposition, further comprising: According to the full polarization Radarsat-2 image, a polarized scattering matrix is obtained; The polarized scattering matrix is converted into a polarized coherence matrix, and then Refind Lee filtering is used for noise reduction processing on the coherence matrix.

4. The snow recognition method according to claim 1, characterized by, According to the 20-band multi-band feature image, a random forest classifier is used for identification to obtain a snow identification result, specifically including: According to the 20-band multi-band feature image, an original training sample set is obtained; wherein the original training sample set contains 55 snow samples and 66 non-snow samples; According to the original training sample set, a Bootstraping method is used to randomly select a sample into the training set with replacement, and the sampling is repeated 100 times to obtain 100 training sets; wherein each training set contains 20 features; According to the 100 training sets respectively, randomly select A new feature set is composed of 100 features, the optimal feature is selected for node splitting by calculating the information amount of the feature, a decision tree is generated, thereby 100 decision tree models are acquired, and the 100 decision trees are combined to form a random forest. According to the random forest, the 20-band multi-band feature image is classified and identified to obtain a snow identification result; wherein the identification result includes snow area and non-snow area.

5. The device for recognizing snow based on full polarization SAR, characterized in that, Comprise: An initial image acquisition module for acquiring a full polarization Radarsat-2 image of a target area; A first feature acquisition module for extracting backscattering coefficients from the full polarization Radarsat-2 image to obtain HH backscattering characteristics, HV backscattering characteristics, VH backscattering characteristics and VV backscattering characteristics; A second feature acquisition module for obtaining a first difference and a first ratio according to the VV backscattering characteristics and the VH backscattering characteristics; wherein the first difference is the difference between the VV backscattering characteristics and the VH backscattering characteristics; and the first ratio is the ratio of the VV backscattering characteristics to the VH backscattering characteristics; A third feature acquisition module for obtaining Pauli_Dbl polarization characteristics, Pauli_Odd polarization characteristics and Pauli_Vol polarization characteristics by Pauli decomposition according to the full polarization Radarsat-2 image; A fourth feature acquisition module for obtaining Freeman_Dbl polarization characteristics, Freeman_Odd polarization characteristics and Freeman_Vol polarization characteristics by Freeman decomposition according to the full polarization Radarsat-2 image; A fifth feature acquisition module is configured to acquire, according to the full-polarization Radarsat-2 image, an Entropy polarization feature, an Anisotropy polarization feature, an Alpha polarization feature and a Lambda polarization feature by decomposition. A sixth feature acquisition module for obtaining Yamaguchi4_Dbl polarization characteristics, Yamaguchi4_Odd polarization characteristics, Yamaguchi4_Vol polarization characteristics and Yamaguchi4_Hlx polarization characteristics by Yamaguchi decomposition according to the full polarization Radarsat-2 image; a feature combination module, configured to perform band combination according to the HH backscattering feature, the HV backscattering feature, the VH backscattering feature, the VV backscattering feature, the first difference value, the first ratio value, the Pauli_Dbl polarization feature, the Pauli_Odd polarization feature, the Pauli_Vol polarization feature, the Freeman_Dbl polarization feature, the Freeman_Odd polarization feature, the Freeman_Vol polarization feature, the Entropy polarization feature, the Anisotropy polarization feature, the Alpha polarization feature, the Lambda polarization feature, the Yamaguchi4_Dbl polarization feature, the Yamaguchi4_Odd polarization feature, the Yamaguchi4_Vol polarization feature, and the Yamaguchi4_Hlx polarization feature, to obtain a multi-band feature image composed of 20 bands; a snow identification module, configured to perform identification on the multi-band feature image composed of 20 bands by using a random forest classifier, to obtain a snow identification result.

6. A device for snow cover identification based on full polarimetric SAR, characterized by A computer readable storage medium includes a computer program stored therein, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the method for identifying snow based on full-polarimetric SAR according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a computer program stored therein, wherein the computer program, when executed, controls a device in which the computer readable storage medium is located to perform the method for identifying snow based on full-polarimetric SAR according to any one of claims 1 to 4.

Images