A method of ground object classification based on polarimetric interferometric SAR

Abstract

The present invention provides a method for classifying land objects based on polarimetric interferometry SAR, comprising: polarimetric interferometry SAR data acquisition; data preprocessing; polarimetric SAR image segmentation; DEM elevation extraction; division of strong and weak scattering regions; weak scattering region classification; strong scattering region classification; vegetation classification; classification result fusion. The present invention provides a method for classifying land objects based on polarimetric interferometry SAR, which comprehensively utilizes the polarization information and interference information of land objects, combines polarimetric SAR image segmentation results, power information, elevation information, graphic features and statistical information, so that the distinguishable land object categories are increased to 8 categories, and the land object classification accuracy is significantly improved.

Description

Technical Field

[0001] The invention belongs to the technical field of radars and relates to a ground object classification method based on polarization interferometry SAR. Background Art

[0002] Polarimetric interferometric SAR can simultaneously obtain polarimetric information and interferometric information of the observed object. Polarimetric information is sensitive to information such as the orientation, shape, and roughness of the object, and can be processed in a specific way to invert the scattering type (odd-order scattering, even-order scattering, and volume scattering) of the observed object. Interferometric information is sensitive to the elevation of the object, and the elevation of the object can be obtained by unwrapping the interference phase. Polarimetric interferometric SAR can obtain these two types of information at the same time, which increases the dimension of object classification. However, in practical applications, there are few studies on how to effectively combine these two types of information for object classification. In practical applications, how to fuse these two types of information, increase the categories of object classification, and improve the accuracy of object classification is still very challenging.

[0003] Traditional polarimetric SAR object classification methods are mainly improved on the basis of the classification method proposed by Cloude and Pottier in 1997 (Cloude SR, Pottier E. An entropy based classification scheme for land applications of polarimetric SAR [J]. IEEE Transactions on Geoscience & Remote Sensing, 1997, 35 (1): 68-78.). However, this type of method does not use the interference information of the object, the types of objects that can be distinguished are limited, and the classification accuracy is not high. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a method for ground object classification based on polarization interferometry SAR, comprising the following steps:

[0005] 1. Polarimetric interferometric SAR data acquisition: Acquire polarimetric interferometric SAR data of the observation scene and perform phase-preserving imaging processing on the data to obtain multiple complex images containing polarization information and interferometric information;

[0006] 2. Data preprocessing: performing data preprocessing on the complex image obtained in step 1, wherein the preprocessing operation includes radiation calibration, polarization interferometry calibration and coherent speckle filtering;

[0007] 3. Polarimetric SAR image segmentation: For the preprocessed complex image in step 2, fuse the information of different polarization channels, extract the full polarization multi-azimuth gradient map, and segment the image into different homogeneous regions, i.e. different grids;

[0008] 4. DEM elevation extraction: Extraction step 1 Obtain HH in the complex image m Complex image and HH s DEM elevation information of the objects in the complex image; the HH m The complex image represents the complex image obtained by the horizontal polarization transmission and horizontal polarization reception of the main antenna. s The complex image represents the complex image obtained by the auxiliary antenna transmitting horizontal polarization and receiving horizontal polarization;

[0009] 5. Divide strong and weak scattering areas: Based on the polarimetric SAR image segmentation result in step 3 and the power of the complex image, divide the area in the complex image into strong scattering areas and weak scattering areas;

[0010] 6. Classification of weak scattering areas: subdivide the weak scattering areas extracted in step 5 through the DEM elevation information and the corner points of the complex image;

[0011] 7. Classification of strong scattering areas: perform polarization decomposition on the strong scattering areas divided in step 5 to obtain surface scattering components, even-order scattering components, and volume scattering components corresponding to different pixels; and classify the strong scattering areas into bare soil, artificial buildings, and vegetation areas according to the relative relationship of these scattering components, and perform adaptive classification;

[0012] 8. Vegetation classification: Polarization decomposition of the vegetation area extracted in step 7 is performed, and the vegetation is subdivided into subdivisions in combination with the DEM elevation information extracted in step 4;

[0013] 9. Classification result fusion: Fusion the classification results of step 6, step 7, and step 8, color the classification results, and form a land feature classification map.

[0014] Furthermore, the classification types of the weak scattering area classification in step 6 include shadows and water bodies; and the classification types of the vegetation classification in step 8 include short broad-leaved vegetation, short coniferous vegetation, tall broad-leaved vegetation and tall coniferous vegetation.

[0015] Furthermore, the complex image specifically includes: HH m Complex image, VH m Complex image, HV m Complex image, VV m Complex image, HH s Complex image, VH s Complex image, HV s Complex image and VV s Complex image; the HH m The complex image represents the complex image obtained by the horizontal polarization transmission and horizontal polarization reception of the main antenna. m The complex image represents the complex image obtained by the main antenna transmitting horizontal polarization and receiving vertical polarization.m The complex image represents the complex image obtained by the vertical polarization transmission and horizontal polarization reception of the main antenna, wherein VV m The complex image represents the complex image obtained by the vertical polarization transmission and vertical polarization reception of the main antenna. s The complex image represents the complex image obtained by the auxiliary antenna horizontally polarized transmission and horizontally polarized reception. s The complex image represents the complex image obtained by the auxiliary antenna transmitting horizontal polarization and receiving vertical polarization. s The complex image represents the complex image obtained by the auxiliary antenna transmitting vertical polarization and receiving horizontal polarization. s The complex image represents the complex image obtained by the auxiliary antenna vertically polarized transmission and vertically polarized reception.

[0016] Furthermore, the step 3 specifically includes the following steps:

[0017] Based on the grid obtained by step 3, calculate the average power of the pixels in each grid; assuming that the number of pixels in grid G ​​is n, the average power SPAN of the pixels in grid G ​​is G It can be expressed as:

[0018]

[0019] in They represent the power corresponding to a certain pixel point in the four complex images acquired by the main antenna; (i, j) represents the coordinates of the pixel point, which is limited to the grid G;

[0020] Select a grid determined as a shadow area, and use M times its average power as the threshold T; the area where the average power of the pixels in the grid obtained by step 3 is less than or equal to the threshold T is classified as a weak scattering area; the area where the average power of the pixels in the grid is greater than the threshold T is classified as a strong scattering area.

[0021] Furthermore, the M is 1.2.

[0022] Furthermore, the operation of subdividing the weak scattering area extracted in step 5 in step 6 specifically includes the following steps:

[0023] According to the polarimetric SAR image segmentation result in step 3 and the DEM elevation information extracted in step 4, the average elevation in each grid is calculated; the average elevations of adjacent grids are compared in order from the near end to the far end of the radar. If the average elevation of the grid at the near end is higher than the average elevation of the grid at the far end, the grid at the near end is marked as an elevation-prominent area;

[0024] Traverse each weak scattering area. If there is an elevation protrusion area at the adjacent close-range end of the area, the area is listed as a suspicious shadow area. If there is no elevation protrusion area at the adjacent close-range end of the area, the area is a water area. The shadow area is confirmed by comparing the number of corner points of the suspicious shadow area and the elevation protrusion area. The method for extracting the corner points of the suspicious shadow area and the elevation protrusion area is as follows: for a certain area, according to the polarization SAR image segmentation result of step 3, the pixel grayscale in the area to be extracted is set to 1, and the pixel grayscale outside the area is set to 0; construct an N×N template, where N is an odd number; place the center of the template at the area to be extracted. Take a pixel on the edge of the area where the corner point is taken, calculate the sum of the grayscale values ​​in the template, if the sum of the grayscale values ​​is less than N×N / 3, the edge point where the center of the template is located is recorded as the corner point; slide the template over each pixel on the edge of the area where the corner point is to be extracted, calculate the sum of the grayscale values ​​in the template, and extract all the corner points; the number of corner points extracted from the suspicious shadow area is n1, and the number of corner points in the adjacent elevation protrusion area is n2. If |n1-n2| / max(n1, n2)<Q, the suspicious shadow area is a shadow area; if |n1-n2| / max(n1, n2)>=Q, the suspicious shadow area is classified as a water body.

[0025] Furthermore, the Q is 0.2.

[0026] Furthermore, the specific steps of classifying the strong scattering areas are:

[0027] Construct the coherence matrix T3,

[0028] T3= <k·k *T > (2)

[0029] in, The superscript * indicates conjugation, the superscript T indicates transposition, and <·> indicates multi-view processing;

[0030] If T3(1,1)>T3(2,2)+T3(3,3), then surface scattering is dominant, and dihedral scattering is small. The three components are solved by equation (3):

[0031]

[0032] If T3(1,1)≤T3(2,2)+T3(3,3), then dihedral scattering is dominant, and surface scattering is small. The three components are solved by equation (4):

[0033]

[0034] Among them, the three components P s , P d and P v represent odd-order scattering component, even-order scattering component and volume scattering respectively;

[0035] Compare the odd scattered components P s , even-order scattered component P d and volume scattering P v The size of the initial classification result is obtained. If the largest component is P v , the corresponding pixel is identified as vegetation; if the largest component is P d , the corresponding pixel is identified as an artificial building; if the largest component is P s , then consider the second largest component, if the second largest component is P v , the corresponding pixel is identified as bare soil; if the largest component is P s , the second largest component is P d , the corresponding pixels are identified as artificial buildings;

[0036] Adaptive classification is performed by combining the polarimetric SAR image segmentation result of step 3, the above preliminary classification results and statistical information.

[0037] Furthermore, the specific steps for obtaining the statistical information are as follows: the polarimetric SAR image segmentation result of step 3 and the above-mentioned preliminary classification result are used to count the types of objects in each grid, and to calculate the proportion of each type of object in the grid; if the proportion of the object with the highest proportion exceeds 70% of the number of pixels in the grid, all objects in the grid are set as the objects with the highest proportion; if the proportion of the object with the highest proportion is less than 70%, the grid is further segmented using the watershed segmentation algorithm to obtain grids of smaller scales; the proportion of the types of objects in the smaller scale grids is calculated in turn, and similarly, if the proportion of the object with the highest proportion exceeds 70% of the number of pixels in the grid, all objects in the grid are set as the objects with the highest proportion; if the proportion of the object with the highest proportion is less than 70%, the grid is further segmented; the above operations are repeated until the proportion of the objects with the highest proportion in all grids exceeds 70%.

[0038] Furthermore, the step of subdividing vegetation specifically includes:

[0039] If H≤α1, δ≤α2, the vegetation pixel is identified as short broad-leaved vegetation;

[0040] If H≤α1, δ>α2, the vegetation pixel is identified as dwarf coniferous vegetation;

[0041] If H>α1, δ≤α2, the vegetation pixel is identified as tall broad-leaved vegetation;

[0042] If H>α1, δ>α2, the vegetation pixel is identified as tall coniferous vegetation;

[0043] Where H represents the elevation of vegetation above the ground, which is obtained by subtracting the average elevation of the bare soil area closest to each vegetation pixel from the elevation of each vegetation pixel; α1 and α2 represent the threshold of vegetation above the ground elevation α1 and the threshold of parameter δ α2, respectively; δ represents the parameter that characterizes the change of vegetation leaf shape from circular to needle-shaped, which is obtained by decomposing the main radar polarization data of the vegetation area. The specific calculation formula is (5)

[0044]

[0045] Here <·> indicates multi-view processing.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] (1) The strong scattering area and the weak scattering area are distinguished by power information, and then classified separately. This avoids the misclassification caused by the extremely low power in the weak scattering area, the noise pollution of the polarization information and interference information, and the inaccurate extraction, thus improving the robustness of the classification algorithm;

[0048] (2) Classify weak scattering areas by fusing interference information, image corner information, etc., point out the cause of shadows from a physical mechanism, and distinguish between shadows and water areas;

[0049] (3) Using polarization information, the strong scattering area is divided into building, vegetation and bare soil pixels. The polarization segmentation results are creatively combined to determine the segmentation scale adaptively by counting whether the proportion of the same type of categories in each grid exceeds the threshold. This operation can improve classification accuracy;

[0050] (4) The introduction of elevation information and the further extraction of vegetation elevation above the ground and parameters characterizing vegetation leaf type increase the differentiation dimension of vegetation and further classify vegetation into tall broad-leaved vegetation, tall coniferous vegetation, short broad-leaved vegetation, and short coniferous vegetation, thereby increasing the number of classifications. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a main flow chart of a method for classifying land objects based on polarization interferometric SAR of the present invention.

[0052] Figure 2 It is a flow chart of weak scattering area classification in the steps of the present invention.

[0053] Figure 3 It is a schematic diagram of the elevation highlight area in the weak scattering area classification step of the present invention.

[0054] Figure 4 It is a schematic diagram of corner point extraction in weak scattering area classification in the step of the present invention.

[0055] Figure 5 This is a flow chart of the strong scattering area classification process of the present invention.

[0056] Figure 6 It is a Pauli basis polarization fusion diagram of the experimental data of an embodiment of the present invention.

[0057] Figure 7 This is a diagram of polarization segmentation results of experimental data of an embodiment of the present invention.

[0058] Figure 8 It is a DEM elevation result map extracted from the test data of the embodiment of the present invention.

[0059] Fig. 9 This is a result diagram of dividing the experimental data of an embodiment of the present invention into strong and weak scattering areas.

[0060] Fig.10 This is a diagram of classification results of weak scattering regions of test data of an embodiment of the present invention.

[0061] Fig.11 This is a diagram showing the classification results of strong scattering regions of the test data of an embodiment of the present invention.

[0062] Fig.12 This is a diagram of the vegetation area classification results of the test data of an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to solve the above technical problems, the present invention provides a method for classifying ground objects based on polarimetric interferometry SAR. The present invention is further described in detail below in conjunction with embodiments and with reference to the accompanying drawings.

[0064] Embodiment 1:

[0065] This embodiment uses a ground feature classification method based on polarimetric interferometric SAR of the present invention to process polarimetric interferometric SAR data acquired by an N-SAR system produced by a certain company. The data was acquired in the Weihe Bridge area of ​​Weinan City, Shaanxi Province in August 2017. The polarimetric fusion diagram of the data is shown in the figure below: Figure 6 This scene includes various landforms such as urban areas, roads, rivers, farmlands, shrubs, and trees, and is suitable for verifying the present invention.

[0066] like Figure 1 As shown, the present invention provides a method for classifying objects based on polarization interferometry SAR, comprising the following steps:

[0067] 1. Polarimetric interferometry SAR data acquisition: The polarimetric interferometry SAR data of the observation scene is acquired through a satellite-borne or airborne polarimetric interferometry SAR system, and the data is processed by phase-preserving imaging to obtain 8 complex images containing polarization information and interferometry information (main antenna: HH m 、HV m , VH m and VV m ; Auxiliary antenna: HHs 、HV s , VH s and VV s ).

[0068] 1.1. Obtain polarimetric interferometric SAR data of the observation scene through a SAR system with polarimetric interferometry function, that is, obtain SAR echo data of eight scenes including main antenna horizontal polarization transmission and horizontal polarization reception, main antenna horizontal polarization transmission and vertical polarization reception, main antenna vertical polarization transmission and horizontal polarization reception, main antenna vertical polarization transmission and vertical polarization reception, auxiliary antenna horizontal polarization transmission and horizontal polarization reception, auxiliary antenna horizontal polarization transmission and vertical polarization reception, auxiliary antenna vertical polarization transmission and horizontal polarization reception, and auxiliary antenna vertical polarization transmission and vertical polarization reception.

[0069] 1.2. Use the imaging algorithm with phase-preserving performance to perform high-quality imaging on the 8-scene SAR echo data obtained in step 1.1, and obtain 8 well-focused complex images, including HH m Complex image (complex image obtained by horizontal polarization transmission and horizontal polarization reception of the main antenna), VH m Complex image (complex image obtained by horizontal polarization transmission of the main antenna and vertical polarization reception), HV m Complex image (complex image obtained by vertical polarization transmission of the main antenna and horizontal polarization reception), VV m Complex image (complex image obtained by vertical polarization transmission and vertical polarization reception of the main antenna), HH s Complex image (complex image obtained by horizontal polarization transmission and horizontal polarization reception of auxiliary antenna), VH s Complex image (complex image obtained by transmitting horizontal polarization and receiving vertical polarization with auxiliary antenna), HV s Complex image (complex image obtained by the auxiliary antenna vertically polarized transmission and horizontally polarized reception) and VV s Complex image (complex image obtained by vertically polarized transmission and vertically polarized reception of the auxiliary antenna).

[0070] 2. Data preprocessing: Perform radiometric calibration, polarization interferometry calibration, and speckle filtering on the eight complex images obtained in step 1. Radiometric calibration is used to correct the unevenness of the antenna pattern in the complex image distance direction; polarization interferometry calibration is used to correct the amplitude, phase imbalance, and polarization leakage between different complex images; and speckle filtering is performed by multi-viewing, Refined-Lee filter, etc. to reduce the speckle noise in the SAR image.

[0071] 3. Polarimetric SAR image segmentation: For the preprocessed complex image in step 2, fuse the information of different polarization channels, extract the full polarization multi-directional gradient map, and use the labeled watershed segmentation algorithm to segment the image into different homogeneous areas, i.e. different grids. Figure 7As shown, it is the result of segmenting the test data. The pixels in the same grid are considered to have the same scattering characteristics.

[0072] 4. DEM elevation extraction: Select the HH obtained in step 1 m Complex image and HH s The DEM elevation information of the ground objects is extracted from the complex image through image registration, deflating phase, phase unwrapping, phase filtering and other steps. Figure 8 The figure shows the result of DEM elevation extraction for the test data.

[0073] 5. Divide strong and weak scattering areas: Based on the polarimetric SAR image segmentation result in step 3 and the power of the complex image, divide the area in the complex image into strong scattering areas and weak scattering areas.

[0074] 5.1. Calculate the average power of each pixel in the grid obtained by segmentation in step 3. Assuming that the number of pixels in the grid G ​​is n, the average power SPAN of the pixels in the grid G ​​is G It can be expressed as:

[0075]

[0076] in They respectively represent the power corresponding to a pixel point in the four complex images acquired by the main antenna; (i, j) represents the coordinates of the pixel point, which is limited to the grid G.

[0077] 5.2. Manually select a grid that is determined to be a shadow area, and use 120% of its average power as the threshold T. The area where the average power of the pixels in the grid obtained by step 3 is less than or equal to the threshold T is classified as a weak scattering area; the area where the average power of the pixels in the grid is greater than the threshold T is classified as a strong scattering area. Fig. 9 The figure shows the result of dividing the test data into strong and weak scattering areas, where the dark area is the weak scattering area and the light area is the strong scattering area.

[0078] 6. Classification of weak scattering areas: Figure 2 As shown in the figure, the weak scattering area extracted in step 5 is subdivided by DEM elevation information and image corner points. The weak scattering area is mainly shadow and water area. The radar observation geometry is a typical oblique imaging geometry. If there is occlusion of high objects (such as buildings and vegetation) in the scene, a shadow related to the shape of the high objects will be formed in the radar line of sight.

[0079] 6.1. Extract the area with prominent elevation. Based on the polarimetric SAR image segmentation result in step 3 and the DEM elevation information extracted in step 4, calculate the average elevation in each grid. Compare the average elevations of adjacent grids in the order from the near-range end to the far-range end of the radar. If the average elevation of the grid at the near-range end is higher than the average elevation of the grid at the far-range end, mark the grid at the near-range end as an area with prominent elevation. Figure 3 As shown, the left side of the schematic area is the near end, and the right side is the far end. The numbers in the grids represent the average elevation. The grids marked in black are the elevation-prominent areas, and their average elevations are higher than the average elevations of the adjacent grids on their left.

[0080] 6.2. Extract suspicious shadow areas. For each weak scattering area extracted in step 5, check whether there is an elevation protrusion area at the adjacent close-range end of the area. If there is an elevation protrusion area at the adjacent close-range end of the area, the area is listed as a suspicious shadow area. If there is no elevation protrusion area at the adjacent close-range end of the area, the area is a water area.

[0081] 6.3. Confirmation of shadow area: For the suspected shadow area in step 6.2, the shadow area is confirmed by comparing the boundary of the suspected shadow area and the elevation protrusion area at the near end. If the suspected shadow area is caused by the occlusion of the elevation protrusion area at the near end, then their boundaries will have a high degree of similarity. The shadow area is confirmed by comparing the number of corner points of the suspected shadow area and the elevation protrusion area.

[0082] For a certain area, its corner points are extracted by the following method: Figure 4 As shown in the figure, according to the polarization SAR image segmentation result of step 3, the grayscale of pixels in the area to be extracted is set to 1, and the grayscale of pixels outside the area is set to 0. Construct an N×N template, where N is an odd number. Place the center of the template on a pixel at the edge of the area to be extracted, calculate the sum of the grayscale values ​​in the template, and if the sum of the grayscale values ​​is less than N×N / 3, record the edge point where the center of the template is located as the corner point. Slide the template over each pixel at the edge of the area to be extracted, calculate the sum of the grayscale values ​​in the template, and extract all the corner points.

[0083] The above method is used to determine the number of corner points of the suspicious shadow area and the elevation protrusion area at the near end. Assuming that the number of corner points extracted from the suspicious shadow area is n1, and the number of corner points of the elevation protrusion area at the adjacent distance end is n2, if |n1-n2| / max(n1, n2) < 0.2, it means that the corner point data of the two areas are not much different, and it is believed that the suspicious shadow area is caused by the occlusion of the elevation protrusion area, and the suspicious shadow area is a shadow area; if |n1-n2| / max(n1, n2) >= 0.2, it means that the corner point data of the two areas are very different, and it is believed that the suspicious shadow area is not caused by occlusion, and it is classified as a water body. Fig.10 The figure shows the result of classifying the weak scattering regions of the test data.

[0084] 7. Classification of strong scattering areas: e.g. Figure 5 As shown, the strong scattering area divided in step 5 is polarized and decomposed to obtain the surface scattering component, even scattering component and volume scattering component corresponding to different pixels. According to the relative relationship of these scattering components, the strong scattering area is classified into bare soil, artificial buildings and vegetation areas and adaptively classified.

[0085] 7.1. Using HH after data preprocessing in step 2 m 、HV m , VH m and VV m The complex image is combined with the strong scattering area obtained in step 5 to extract polarization information for preliminary classification. As shown in formula (2), the coherence matrix T3 is constructed:

[0086] T3= <k·k *T > (2)

[0087] in, The superscript * indicates conjugation, the superscript T indicates transposition, and <·> indicates multi-view processing. The odd scattered component P of the strong scattered area is extracted by the following method: s , even-order scattered component P d and volume scattering P v .

[0088] If T3(1,1)>T3(2,2)+T3(3,3), then surface scattering is dominant, and dihedral scattering is small. The three components are solved by equation (3):

[0089]

[0090] If T3(1,1)≤T3(2,2)+T3(3,3), then dihedral scattering is dominant, and surface scattering is small. The three components are solved by equation (4):

[0091]

[0092] Compare the odd scattered components P s , even-order scattered component P d and volume scattering P v If the largest component is P v , the corresponding pixel is identified as vegetation; if the largest component is P d , the corresponding pixel is identified as an artificial building; if the largest component is P s , then consider the second largest component, if the second largest component is P v, the corresponding pixel is identified as bare soil; if the largest component is P s , the second largest component is P d , the corresponding pixels are identified as artificial buildings.

[0093] 7.2. Perform adaptive classification based on the polarimetric SAR image segmentation result of step 3, the preliminary classification result of step 7.1 and the statistical information.

[0094] According to the segmentation results, count the types of features in each grid and calculate the proportion of each feature type in the grid. If the proportion of the feature with the highest proportion exceeds 70% of the number of pixels in the grid, all features in the grid are set as the features with the highest proportion; if the proportion of the feature with the highest proportion is less than 70%, use the watershed segmentation algorithm to continue segmenting the grid to obtain smaller-scale grids; calculate the proportion of feature categories in the smaller-scale grids in turn. Similarly, if the proportion of the feature with the highest proportion exceeds 70% of the number of pixels in the grid, all features in the grid are set as the features with the highest proportion; if the proportion of the feature with the highest proportion is less than 70%, continue segmenting the grid. Repeat the above steps until the proportion of the features with the highest proportion in all grids exceeds 70%. Fig.11 Shown is the result of classifying the strong scattering areas of the test data.

[0095] 8. Vegetation classification: Polarization decomposition is performed on the vegetation area extracted in step 7, and combined with the DEM elevation information extracted in step 4, the vegetation is subdivided into tall broad-leaved vegetation, tall coniferous vegetation, short broad-leaved vegetation, and short coniferous vegetation.

[0096] 8.1. Subtract the average elevation of the bare soil area closest to each vegetation pixel from its elevation to obtain the vegetation elevation above ground. This elevation represents the height of the vegetation relative to the ground, not the absolute altitude.

[0097] 8.2. Decompose the main radar polarization data of the vegetation area and extract the parameter δ that represents the change of the vegetation leaf shape from circular to needle-shaped. δ can be calculated by formula (5):

[0098]

[0099] Here <·> indicates multi-view processing.

[0100] 8.3. Combine the vegetation elevation above ground extracted in step 8.1 and the parameter δ extracted in step 8.2 to classify vegetation. Let the vegetation elevation above ground be H, set the threshold of vegetation elevation above ground α1 and the threshold of parameter δ α2,

[0101] If H≤α1, δ≤α2, the vegetation pixel is identified as short broad-leaved vegetation;

[0102] If H≤α1, δ>α2, the vegetation pixel is identified as dwarf coniferous vegetation;

[0103] If H>α1, δ≤α2, the vegetation pixel is identified as tall broad-leaved vegetation;

[0104] If H>α1, δ>α2, the vegetation pixel is identified as tall coniferous vegetation.

[0105] like Fig.12 Shown is the result of vegetation area classification for the test data.

[0106] 9. Fusion of classification results: Fusion of the classification results of steps 6, 7, and 8, color matching of the classification results, and formation of a land feature classification map.

[0107] The invention discloses a method for classifying ground objects based on polarization interferometry SAR, which creatively integrates polarization information, interference information, power information, image information and statistical information of polarization interferometry SAR to classify ground objects.

[0108] The strong scattering area and the weak scattering area are distinguished by power information, and then classified separately. This avoids the misclassification caused by the extremely low power in the weak scattering area, the noise pollution of the polarization information and interference information, and the inaccurate extraction, and improves the robustness of the classification algorithm.

[0109] By fusing interference information, image corner points and other information, the weak scattering areas are classified, the causes of shadows are pointed out from the physical mechanism, and shadows and water areas are distinguished.

[0110] Using polarization information, we can distinguish strong scattering areas into building, vegetation, and bare soil pixels. We can creatively combine the polarization segmentation results and adaptively determine the segmentation scale by counting the proportion of the same type of categories in each grid to see whether it exceeds the threshold. This operation can improve classification accuracy.

[0111] The introduction of elevation information and the further extraction of vegetation elevation above the ground and parameters characterizing vegetation leaf type increase the vegetation differentiation dimension and further classify vegetation into tall broad-leaved vegetation, tall coniferous vegetation, short broad-leaved vegetation and short coniferous vegetation, thereby increasing the number of classifications.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for ground object classification based on polarimetric interferometry SAR, characterized in that: The following steps are involved:

1. Polarimetric interferometric SAR data acquisition: Acquire polarimetric interferometric SAR data of the observation scene and perform phase-preserving imaging processing on the data to obtain multiple complex images containing polarization information and interferometric information; 2. Data preprocessing: performing data preprocessing on the complex image obtained in step 1, wherein the preprocessing operation includes radiation calibration, polarization interferometry calibration and coherent speckle filtering; 3. Polarimetric SAR image segmentation: For the preprocessed complex image in step 2, fuse the information of different polarization channels, extract the full polarization multi-azimuth gradient map, and segment the image into different homogeneous regions, i.e. different grids; 4. DEM elevation extraction: Extraction step 1 Get the Complex images and DEM elevation information of the objects in the complex image; The complex image represents the complex image obtained by the horizontal polarization transmission and horizontal polarization reception of the main antenna. The complex image represents the complex image obtained by the auxiliary antenna transmitting horizontal polarization and receiving horizontal polarization; 5. Divide strong and weak scattering areas: Based on the polarimetric SAR image segmentation result in step 3 and the power of the complex image, divide the area in the complex image into strong scattering areas and weak scattering areas; 6. Classification of weak scattering areas: subdivide the weak scattering areas extracted in step 5 through the DEM elevation information and the corner points of the complex image; 7. Classification of strong scattering areas: perform polarization decomposition on the strong scattering areas divided in step 5 to obtain surface scattering components, even-order scattering components, and volume scattering components corresponding to different pixels; According to the relative relationship of these scattering components, the strong scattering areas are classified into bare soil, artificial buildings and vegetation areas and adaptively classified; 8. Vegetation classification: Polarization decomposition of the vegetation area extracted in step 7 is performed, and the vegetation is subdivided into subdivisions in combination with the DEM elevation information extracted in step 4; 9. Classification result fusion: Fusion the classification results of step 6, step 7, and step 8, color the classification results, and form a land feature classification map.

2. The method for ground object classification based on polarimetric interferometry SAR according to claim 1, characterized in that: The classification types of the weak scattering area classification in step 6 include shadows and water bodies; the classification types of the vegetation classification in step 8 include short broad-leaved vegetation, short coniferous vegetation, tall broad-leaved vegetation and tall coniferous vegetation.

3. The method for ground object classification based on polarimetric interferometry SAR according to claim 1, characterized in that: The complex image specifically includes: Re-image, Re-image, Re-image, Re-image, Re-image, Re-image, Complex image and complex image; The complex image represents the complex image obtained by the horizontal polarization transmission and horizontal polarization reception of the main antenna. The complex image represents the complex image obtained by the main antenna transmitting horizontal polarization and receiving vertical polarization. The complex image represents the complex image obtained by the main antenna transmitting vertical polarization and receiving horizontal polarization. The complex image represents the complex image obtained by the vertical polarization transmission and vertical polarization reception of the main antenna. The complex image represents the complex image obtained by the auxiliary antenna horizontally polarized transmission and horizontally polarized reception. The complex image represents the complex image obtained by the auxiliary antenna transmitting horizontal polarization and receiving vertical polarization. The complex image represents the complex image obtained by the auxiliary antenna transmitting vertical polarization and receiving horizontal polarization. The complex image represents the complex image obtained by the auxiliary antenna vertically polarized transmission and vertically polarized reception.

4. The method for ground object classification based on polarimetric interferometry SAR according to claim 3, characterized in that: The step 3 specifically comprises the following steps: Based on the grid obtained by step 3, calculate the average power of the pixels in each grid; assuming that the number of pixels in grid G ​​is n, the average power of the pixels in grid G ​​is It can be expressed as: in , , , They respectively represent the power corresponding to a certain pixel point in the four complex images acquired by the main antenna; represents the coordinates of the pixel point, which is limited to the grid G; select a grid determined as the shadow area, and use M times its average power as the threshold T; the area where the average power of the pixel points in the grid obtained by step 3 is less than or equal to the threshold T is classified as a weak scattering area; the area where the average power of the pixel points in the grid is greater than the threshold T is classified as a strong scattering area.

5. The method for ground object classification based on polarimetric interferometry SAR according to claim 4, characterized in that: The M is taken as 1.

2.

6. The method for ground object classification based on polarimetric interferometry SAR according to claim 4, characterized in that: The operation of subdividing the weak scattering area extracted in step 5 in step 6 specifically includes the following steps: According to the polarimetric SAR image segmentation result in step 3 and the DEM elevation information extracted in step 4, the average elevation in each grid is calculated; the average elevations of adjacent grids are compared in order from the near end to the far end of the radar. If the average elevation of the grid at the near end is higher than the average elevation of the grid at the far end, the grid at the near end is marked as an elevation-prominent area; Traverse each weak scattering area. If there is an elevation protrusion area at the adjacent close-range end of the area, the area is listed as a suspicious shadow area. If there is no elevation protrusion area at the adjacent close-range end of the area, the area is a water area. The shadow area is confirmed by comparing the number of corner points in the suspicious shadow area and the elevation prominent area. The method for extracting the corner points of the suspicious shadow area and the elevation prominent area is as follows: For a certain area, according to the polarization SAR image segmentation result in step 3, the pixel gray value in the area where the corner points are to be extracted is set to 1, and the pixel gray value outside this area is set to 0; construct an N×N template, where N is an odd number; place the center of the template on a pixel at the edge of the area where the corner points are to be extracted, calculate the sum of the gray values within the template, if the sum of the gray values is less than N×N / 3, then mark the edge point where the template center is located as a corner point; slide the template over each pixel at the edge of the area where the corner points are to be extracted, calculate the sum of the gray values within the template, and extract all corner points; the number of corner points extracted from the suspicious shadow area is n1, and the number of corner points in the adjacent near-end elevation prominent area is n2. If |n1 - n2| / max(n1, n2) < Q, this suspicious shadow area is a shadow area; if |n1 - n2| / max(n1, n2) >= Q, this suspicious shadow area is classified as water body.

7. The method for ground object classification based on polarization interferometry SAR according to claim 6, characterized in that: The Q is taken as 0.

2.

8. The method for ground object classification based on polarization interferometry SAR according to claim 6, characterized in that: The specific steps for classifying the strong scattering area are as follows: Constructing the coherence matrix , (2) in, , the superscript * indicates conjugation, and the superscript T indicates transposition. represents multi-view processing; like , then surface scattering is dominant, and dihedral scattering is small. The three components are solved by equation (3): (3) like , then dihedral scattering is dominant, and surface scattering is small. The three components are solved by equation (4): (4) Among them, the three components , and represent odd-order scattering component, even-order scattering component and volume scattering respectively; Comparison of odd-order scattered components , even-order scattered components and volume scattering The size of the initial classification result is obtained. If the largest component is , the corresponding pixel is identified as vegetation; if the largest component is , the corresponding pixel is identified as an artificial building; if the largest component is , then consider the second largest component. If the second largest component is , the corresponding pixel is identified as bare soil; if the largest component is , the second largest component is , the corresponding pixels are identified as artificial buildings; Perform adaptive classification by combining the polarization SAR image segmentation result in step 3, the above preliminary classification result, and the statistical information.

9. The method for ground object classification based on polarimetric interferometry SAR according to claim 8, characterized in that: The specific steps for obtaining the statistical information are as follows: For the polarization SAR image segmentation result in step 3 and the above preliminary classification result, count the land cover types in each grid, and calculate the proportion of each land cover type in this grid; if the proportion of the land cover type with the highest proportion exceeds 70% of the number of pixels in this grid, then set all the land cover in this grid to the land cover type with the highest proportion; if the proportion of the land cover type with the highest proportion is less than 70%, then use the watershed segmentation algorithm to continue to divide this grid to obtain smaller-scale grids; Calculate the proportion of land cover types in the smaller-scale grids in turn. Similarly, if the proportion of the land cover type with the highest proportion exceeds 70% of the number of pixels in this grid, then set all the land cover in this grid to the land cover type with the highest proportion; if the proportion of the land cover type with the highest proportion is less than 70%, then continue to divide this grid; repeat the above operations until the proportion of the land cover type with the highest proportion in all grids exceeds 70%.

10. The method for ground object classification based on polarization interferometry SAR according to claim 9, characterized in that: The steps for subdividing vegetation specifically include: like , , then the vegetation pixel is identified as short broad-leaved vegetation; like , , then the vegetation pixel is identified as dwarf coniferous vegetation; like , , then the vegetation pixel is identified as tall broad-leaved vegetation; like , , then the vegetation pixel is identified as tall coniferous vegetation; in, It represents the elevation of vegetation above the ground, which is obtained by subtracting the average elevation of the nearest bare soil area from the elevation of each vegetation pixel; and Represents the vegetation elevation threshold and parameters Threshold ; It represents the parameter that characterizes the change of vegetation leaf shape from circular to needle-shaped. It is obtained by decomposing the main radar polarization data of the vegetation area. Its specific calculation formula is (5) (5) in Indicates multi-view processing.

Images