A vegetation and building classification method based on polarization and interference information

Abstract

The present invention relates to a vegetation and building classification method based on polarization and interference information, including: S1, data acquisition and preprocessing; S2, polarization SAR classification; S3, polarization SAR image segmentation, segmenting the polarization SAR image acquired by S1 to form connected homogeneous areas, and dividing vegetation areas and building areas; S4, calculating the coherence coefficient in combination with the complex image, and calculating the average coherence coefficient of each area in combination with all vegetation areas and building areas acquired by S3; S5, performing 45° inclined angle straight line detection on all vegetation areas and building areas acquired by S3; S6, reclassifying vegetation areas and building areas. On the basis of traditional polarization SAR ground object classification, the present invention further distinguishes vegetation areas and building areas by using coherence coefficient and straight line detection, so that misclassified vegetation and building areas can be correctly distinguished, and the accuracy of ground object classification is improved.

Description

Technical Field

[0001] The invention relates to the field of urban surveying and mapping, and in particular to a vegetation and building classification method based on polarization and interference information. Background Art

[0002] In polarimetric SAR object classification, due to the existence of polarization orientation angle (especially polarization orientation angle close to 45°), a large number of volume scattering components are often identified in buildings, which can easily misclassify buildings as vegetation. Since the scattering environment in some vegetation areas is complex and the scattering types are rich, such as lotus ponds in summer, this type of vegetation is often accompanied by strong secondary scattering, which can easily misclassify vegetation as buildings during classification.

[0003] The traditional land object classification method based on polarization information is prone to misclassification of vegetation and buildings, which is not conducive to the direct application of polarimetric SAR land object classification results in urban surveying and mapping. Summary of the invention

[0004] In order to solve the existing technical problems, the present invention provides a vegetation and building classification method based on polarization and interference information.

[0005] The specific content of the present invention is as follows: A vegetation and building classification method based on polarization and interference information comprises the following steps:

[0006] S1, data acquisition and preprocessing, acquire polarimetric SAR images and interferometric SAR image pairs for the test and preprocess them. The SAR images used for the test include: the complex image (HHm) of horizontal polarization transmission and horizontal polarization reception of the main antenna, the complex image (VHm) of horizontal polarization transmission and vertical polarization reception of the main antenna, the complex image (HVm) of vertical polarization transmission and horizontal polarization reception of the main antenna, the complex image (VVm) of vertical polarization transmission and vertical polarization reception of the main antenna, and the complex image (HHs) of horizontal polarization transmission and horizontal polarization reception of the auxiliary antenna.

[0007] S2, polarimetric SAR classification, uses the traditional polarimetric SAR object classification method to perform polarimetric classification on the polarimetric SAR images (HHm, HVm, VHm and VVm) preprocessed in S1 to obtain vegetation pixels and building pixels.

[0008] S3, polarimetric SAR image segmentation, segments the polarimetric SAR image obtained in S1 to form connected homogeneous regions. According to the proportion of vegetation pixels and building pixels classified in S2 in the homogeneous region, the vegetation region and building region are divided.

[0009] S4, combining the HHm and HHs complex images in S1 to calculate the coherence coefficient, combining all vegetation areas and building areas obtained in S3 to calculate the average coherence coefficient of each area.

[0010] S5, performing straight line detection at a 45° inclination angle on all vegetation areas and building areas obtained in S3.

[0011] S6, combining the coherence coefficient of each area obtained by S4 and the 45° tilt angle straight line detection result of S5 to reclassify the vegetation area and the building area.

[0012] Furthermore, in S1, the image is preprocessed including radiation calibration, polarization calibration, interference calibration and coherent speckle removal.

[0013] Furthermore, in S2, the polarimetric SAR method used includes a Wishart-H / alpha classification method or an unsupervised classification method based on a scattering model.

[0014] Furthermore, an unsupervised classification method based on a scattering model is used to classify objects, including:

[0015] S21, firstly, polarization decomposition is performed to obtain surface scattering, secondary scattering and volume scattering power; then, pixels are marked as one of the three scattering types according to the maximum of surface scattering, secondary scattering and volume scattering power; then, pixels in each scattering type are divided into 30 clusters; then, the Wishart classifier is used to merge the clusters, reducing the number of cluster categories to 15.

[0016] S22, the category with the lowest power among the 15 categories is divided into the first category, which includes shadows, water bodies, and areas with extremely low power; the categories with surface scattering as the main scattering type are merged into the second category, which includes ground objects with dominant surface scattering such as bare soil; the categories with secondary scattering as the main scattering type are merged into the third category, which includes building areas, and the pixels in this category are building pixels; the categories with volume scattering as the main scattering type are merged into the fourth category, which includes vegetation areas, and the pixels in this category are vegetation pixels.

[0017] Furthermore, in S3, the polarimetric SAR image is segmented using a watershed segmentation algorithm to obtain multiple connected regions, and the pixels in each connected region are homogeneous and belong to the same category;

[0018] Count the pixel category with the highest proportion and its proportion in each connected area. If its proportion exceeds 80%, set all pixel categories in the area as the category with the highest proportion.

[0019] If the proportion of the highest-occupying pixels in the region is less than 80%, the region is further segmented to obtain smaller connected regions;

[0020] Count the proportion of pixels with the highest proportion in the small connected area. If the proportion of pixels with the highest proportion exceeds 80%, set the category of all pixels in the area to the category with the highest proportion.

[0021] Repeat the above steps until the highest proportion of the category in each connected area exceeds 80%.

[0022] Furthermore, in S4, the coherence coefficient γ is calculated as

[0023]

[0024] Among them, <·> is the mathematical expectation operator, * represents conjugation, γ is less than or equal to 1, and the average coherence coefficient of each area is calculated by combining the vegetation area and building area extracted by S3. Set the threshold T1, if the average coherence coefficient If the average coherence coefficient exceeds the threshold T1, the corresponding area is marked as 1; If the threshold value T1 is not exceeded, the corresponding area is marked as 0.

[0025] Furthermore, in S5, the specific steps of performing 45° inclined angle straight line detection on all vegetation areas and building areas obtained in S3 include:

[0026] S51, using the Canny operator to detect image edges of three images, namely, the complex image transmitted and received by the horizontal polarization of the main antenna, the complex image transmitted and received by the vertical polarization of the main antenna, and the complex image transmitted and received by the vertical polarization of the main antenna, and superimposing the edge detection results of the three images, and recording the superimposed edge detection result as G;

[0027] S52, using Hough transform to perform straight line detection on the edge detection result G extracted in S51, retaining straight lines with inclination angles between -25° and -65° and between 25° and 65°, and removing straight lines with too short lengths;

[0028] S53, counting the number of straight lines detected in S52 in all vegetation areas and building areas, setting a threshold T2, if the number of straight lines detected in the area exceeds the threshold T2, the corresponding area is marked as 1; if the number of straight lines detected in the area does not exceed the threshold T2, the corresponding area is marked as 0.

[0029] Furthermore, the reclassification of vegetation areas and building areas in S6 includes: for all vegetation areas, if the average coherence coefficient in S4 is marked as 0 and the number of straight lines with an inclination angle of 45° in S53 is marked as 1, then the area is divided into a building area; for all building areas, if the average coherence coefficient in S4 is marked as 1 and the number of straight lines with an inclination angle of 45° in S53 is marked as 0, then the area is divided into a vegetation area.

[0030] In view of the disadvantage that traditional polarimetric SAR object classification methods easily cause misclassification of vegetation and buildings, the present invention aims to provide a vegetation and building classification method based on polarization and interference information. On the basis of traditional polarimetric SAR object classification, the coherence coefficient and straight line detection are used to further distinguish vegetation areas and building areas, so that the misclassified vegetation and building areas can be correctly distinguished, thereby improving the accuracy of object classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The specific implementation of the present invention will be further explained below in conjunction with the accompanying drawings.

[0032] Figure 1 It is the main flow chart of the vegetation and building classification method based on polarization and interference information of the present invention;

[0033] Figure 2 is a flow chart of the steps of polarization SAR classification of the present invention;

[0034] Figure 3 This is the main flow chart of the step 45° tilt angle straight line detection of the present invention;

[0035] Figure 4 is the Paul i basis fusion graph of polarization SAR data used in the embodiment of the present invention;

[0036] Figure 5 This is a polarization SAR classification result diagram using a traditional method in an embodiment of the present invention;

[0037] Figure 6 is a data coherence coefficient diagram of an embodiment of the present invention;

[0038] Figure 7 It is the result of reclassification of vegetation area and building area in the embodiment of the present invention. DETAILED DESCRIPTION

[0039] Combination Figure 1-Figure 7 , the vegetation and building classification method based on polarization and interference information of the present invention is used to process the polarization interferometric SAR data acquired by the N-SAR system produced by the 14th Institute of China Electronics. The data was acquired in Weinan City, Shaanxi Province in April 2017. The polarization fusion diagram of the data is shown in the figure below: Figure 4 The scene includes buildings with 0° and 45° orientation angles, the polarization fusion image of the building with 0° orientation angle is red, and the polarization fusion image of the building with 45° orientation angle is green. This data is suitable for verifying the present invention.

[0040] like Figure 1 As shown, the present invention provides a vegetation and building classification method based on polarization and interference information, comprising the following steps:

[0041] S1, data acquisition and preprocessing, acquire polarization SAR images and interference image pairs for the test and perform preprocessing. The SAR images used for the test include the complex image (HHm) of horizontal polarization transmission and horizontal polarization reception of the main antenna, the complex image (VHm) of vertical polarization transmission and vertical polarization reception of the main antenna, the complex image (HVm) of vertical polarization transmission and horizontal polarization reception of the main antenna, the complex image (VVm) of vertical polarization transmission and vertical polarization reception of the main antenna, and the complex image (HHs) of horizontal polarization transmission and horizontal polarization reception of the auxiliary antenna. And perform preprocessing operations such as radiation calibration, polarization calibration, interference calibration and coherent speckle removal on the above complex images. Figure 4 Shown is a polarimetric fusion diagram of the data used in the example.

[0042] S2, polarimetric SAR classification, uses the traditional polarimetric SAR object classification method to polarimetrically classify the polarimetric SAR images (HHm, HVm, VHm and VVm) preprocessed in S1, and merges the categories to obtain vegetation pixels and building pixels. The polarimetric SAR classification method used can be the Wishart-H / alpha classification method, the unsupervised classification method based on the scattering model, or other similar classification methods. The specific steps are as follows:

[0043] S21, such as Figure 2 As shown, without loss of generality, this embodiment uses an unsupervised classification method based on a scattering model to classify ground objects. First, polarization decomposition is performed to obtain surface scattering, secondary scattering, and volume scattering power; then, based on the maximum of surface scattering, secondary scattering, and volume scattering power, the pixel is marked as one of the three scattering types; then, the pixels in each scattering type are divided into 30 clusters; then, the Wishart classifier is used to merge the clusters, so that the cluster categories are reduced to 15 categories.

[0044] S22, the category with the lowest power among the 15 categories is divided into category 1, which usually contains shadows, water bodies, and areas with extremely low power; the categories with surface scattering as the main scattering type are merged into category 2, which usually contains ground objects with dominant surface scattering such as bare soil; the categories with secondary scattering as the main scattering type are merged into category 3, which usually contains building areas, and the pixels in this category are building pixels; the categories with volume scattering as the main scattering type are merged into category 4, which usually contains vegetation areas, and the pixels in this category are vegetation pixels.

[0045] like Figure 5 As shown, the data of the embodiment is classified using an unsupervised classification method based on a scattering model and the result of class merging is shown, wherein red pixels represent buildings, green pixels represent vegetation pixels, yellow pixels represent bare soil pixels, and black pixels represent weak scattering pixels.

[0046] S3, segment the polarimetric SAR image obtained in S1 to form connected homogeneous regions. According to the proportion of vegetation pixels and building pixels classified in S2 in the homogeneous region, the vegetation region and the building region are divided. The specific steps are as follows:

[0047] Use the watershed segmentation algorithm to segment the polarimetric SAR image to obtain multiple connected regions. The pixels in each connected region are homogeneous and can be considered to belong to the same category. Count the pixel category with the highest proportion and its proportion in each connected region. If its proportion exceeds 80%, set all pixel categories in the region to the category with the highest proportion. If the proportion of the pixels with the highest proportion in the region is less than 80%, further segment the region to obtain smaller connected regions. Count the proportion of the pixels with the highest proportion in the small connected region. If the proportion of the pixels with the highest proportion exceeds 80%, set all pixel categories in the region to the category with the highest proportion. Repeat the above steps until the category with the highest proportion in each connected region exceeds 80%.

[0048] S4, calculate the coherence coefficient by combining the HHm and HHs complex images in S1, and calculate the average coherence coefficient of each area by combining all vegetation areas and building areas obtained in S3. The coherence coefficient γ is calculated as

[0049]

[0050] Among them, <·> is the mathematical expectation operator, * represents conjugation, and γ is less than or equal to 1. Combine the vegetation area and building area extracted by S3 to calculate the average coherence coefficient of each area Set the threshold T1, if the average coherence coefficient If the average coherence coefficient exceeds the threshold T1, the corresponding area is marked as 1; If the threshold value T1 is not exceeded, the corresponding area is marked as 0. In this embodiment, T1 is preferably 0.8. Figure 6 Shown is a coherence coefficient diagram of the experimental data of the embodiment.

[0051] S5, detects straight lines with a 45° inclination angle for all vegetation areas and building areas obtained in S3. The volume scattering in the building area is mainly caused by the scattering of the building at a 45° angle to the aircraft route. Therefore, detecting the straight lines with a 45° orientation in the ground objects can determine whether there is a 45° dihedral angle in the ground objects, and further distinguish the vegetation areas from the building areas. Figure 3 As shown, the specific steps for 45° tilt angle straight line detection are:

[0052] S51, using the Canny operator to detect image edges for the three images HHm, HVm and VVm respectively, and superimposing the edge detection results of the three images, and recording the superimposed edge detection result as G.

[0053] S52, using Hough transform to perform straight line detection on the edge detection result G extracted in S51. Straight lines with inclination angles between -25° and -65° and between 25° and 65° are retained, and straight lines with too short lengths are removed.

[0054] S53, counting the number of straight lines detected in S52 in all vegetation areas and building areas. Set a threshold T2, if the number of straight lines detected in the area exceeds the threshold T2, the corresponding area is marked as 1; if the number of straight lines detected in the area does not exceed the threshold T2, the corresponding area is marked as 0. In this embodiment, T2 is preferably 15.

[0055] S6, combining the coherence coefficient of each area obtained by S4 and the 45° tilt angle straight line detection result of S5 to reclassify the vegetation area and the building area.

[0056] For all vegetation areas, if the average coherence coefficient in S4 is marked as 0 and if the number of straight lines with a 45° inclination angle in S53 is marked as 1, the area is classified as a building area.

[0057] For all built-up areas, if the average coherence coefficient in S4 is marked as 1 and if the number of straight lines with a 45° inclination angle in S53 is marked as 0, the area is classified as a vegetation area.

[0058] like Figure 7 The figure shows the result of reclassification of vegetation area and building area. It can be seen from the figure that the vegetation and buildings that were originally misclassified have been correctly distinguished.

[0059] The present invention starts from the misclassification mechanism. Buildings are misclassified as vegetation because the radar and the building form a 45° (or close to 45°) orientation angle. Therefore, the category of the area in the scene with an inclination angle close to 45° needs to be confirmed again; vegetation is misclassified as buildings because the vegetation area has rich scattering types and may contain strong secondary scattering (for example, the trunk and the ground form a dihedral angle). For the classified vegetation and building areas, they are further classified through 45° straight line detection and coherence coefficient extraction. After the vegetation and building areas are reclassified by the method of the present invention, the probability of misclassification of vegetation and buildings can be greatly reduced, and the accuracy of polarization SAR ground object classification can be improved, which has great application prospects in agriculture, forestry, surveying and mapping, etc.

[0060] Many specific details are described in the above description to facilitate a full understanding of the present invention. However, the above description is only a preferred embodiment of the present invention. The present invention can be implemented in many other ways different from those described herein, so the present invention is not limited to the specific implementation disclosed above. At the same time, any person familiar with the art can make many possible changes and modifications to the technical solution of the present invention using the methods and technical contents disclosed above without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A vegetation and building classification method based on polarization and interference information, characterized in that: The steps include: S1, data acquisition and preprocessing, acquiring polarimetric SAR images and interferometric SAR image pairs for the test and preprocessing them; The SAR images used for the test include: the complex image of horizontal polarization transmission and horizontal polarization reception of the main antenna, the complex image of vertical polarization transmission and vertical polarization reception of the main antenna, the complex image of vertical polarization transmission and horizontal polarization reception of the main antenna, the complex image of vertical polarization transmission and vertical polarization reception of the main antenna, and the complex image of horizontal polarization transmission and horizontal polarization reception of the auxiliary antenna; S2, polarimetric SAR classification, uses the traditional polarimetric SAR object classification method to polarimetrically classify the polarimetric SAR image preprocessed in S1, and merges the categories to obtain vegetation pixels and building pixels; S3, polarimetric SAR image segmentation, segmenting the polarimetric SAR image obtained in S1 to form connected homogeneous regions, and dividing the vegetation region and the building region according to the proportion of vegetation pixels and building pixels classified in S2 in the homogeneous region; S4, combining the complex image of horizontal polarization transmission and horizontal polarization reception of the main antenna in S1 and the complex image of horizontal polarization transmission and horizontal polarization reception of the auxiliary antenna in S1 to calculate the coherence coefficient, and combining all vegetation areas and building areas obtained in S3 to calculate the average coherence coefficient of each area; S5, performing straight line detection at a 45° inclination angle on all vegetation areas and building areas obtained in S3; S6, combining the coherence coefficient of each area obtained by S4 and the 45° tilt angle straight line detection result of S5 to reclassify the vegetation area and the building area.

2. The vegetation and building classification method based on polarization and interference information according to claim 1, characterized in that: In S1, the image is preprocessed including radiation calibration, polarization calibration, interference calibration and coherent speckle removal.

3. The vegetation and building classification method based on polarization and interference information according to claim 1, characterized in that: In S2, the polarimetric SAR methods used include the Wishart-H / alpha classification method or the unsupervised classification method based on the scattering model.

4. The vegetation and building classification method based on polarization and interference information according to claim 3 is characterized in that: Use unsupervised classification methods based on scattering models to classify objects, including: S21, perform polarization decomposition to obtain surface scattering, secondary scattering and volume scattering power; mark the pixel as one of the three scattering types according to the maximum of surface scattering, secondary scattering and volume scattering power; divide the pixels in each scattering type into 30 clusters; then use Wishart classifier to merge clusters, so that the cluster categories are reduced to 15 categories; S22, the category with the lowest power among the 15 categories is divided into the first category, which includes shadows, water bodies, and areas with extremely low power; the categories with surface scattering as the main scattering type are merged into the second category, which includes ground objects with dominant surface scattering such as bare soil; the categories with secondary scattering as the main scattering type are merged into the third category, which includes building areas, and the pixels in this category are building pixels; the categories with volume scattering as the main scattering type are merged into the fourth category, which includes vegetation areas, and the pixels in this category are vegetation pixels.

5. The vegetation and building classification method based on polarization and interference information according to claim 1, characterized in that: In S3, the polarimetric SAR image is segmented using the watershed segmentation algorithm to obtain multiple connected regions, and the pixels in each connected region are homogeneous and belong to the same category; Count the pixel category with the highest proportion and its proportion in each connected area. If its proportion exceeds 80%, set all pixel categories in the area as the category with the highest proportion. If the proportion of the highest-occupying pixels in the region is less than 80%, the region is further segmented to obtain smaller connected regions; Count the proportion of pixels with the highest proportion in the small connected area. If the proportion of pixels with the highest proportion exceeds 80%, set the category of all pixels in the area to the category with the highest proportion. Repeat the above steps until the highest proportion of the category in each connected area exceeds 80%.

6. The vegetation and building classification method based on polarization and interference information according to claim 1, characterized in that: In S4, the coherence coefficient γ is calculated as follows: Among them, <·> is the mathematical expectation operator, * represents conjugation, and γ is less than or equal to 1; Combine the vegetation area and building area extracted by S3 to calculate the average coherence coefficient of each area Set the threshold T1, if the average coherence coefficient If the average coherence coefficient exceeds the threshold T1, the corresponding area is marked as 1; If the threshold value T1 is not exceeded, the corresponding area is marked as 0.

7. The vegetation and building classification method based on polarization and interference information according to claim 1, characterized in that: In S5, the specific steps of performing 45° inclined angle straight line detection on all vegetation areas and building areas obtained in S3 include: S51, using the Canny operator to detect image edges of three images, namely, the complex image transmitted and received by the horizontal polarization of the main antenna, the complex image transmitted and received by the vertical polarization of the main antenna, and the complex image transmitted and received by the vertical polarization of the main antenna, and superimposing the edge detection results of the three images, and recording the superimposed edge detection result as G; S52, using Hough transform to perform straight line detection on the edge detection result G extracted in S51, retaining straight lines with inclination angles between -25° and -65° and between 25° and 65°, and removing straight lines with too short lengths; S53, counting the number of straight lines detected in S52 in all vegetation areas and building areas, setting a threshold T2, if the number of straight lines detected in the area exceeds the threshold T2, the corresponding area is marked as 1; if the number of straight lines detected in the area does not exceed the threshold T2, the corresponding area is marked as 0.

8. The vegetation and building classification method based on polarization and interference information according to claim 7, characterized in that: The reclassification of vegetation areas and building areas in S6 includes: for all vegetation areas, if the average coherence coefficient in S4 is marked as 0 and the number of straight lines with an inclination angle of 45° in S53 is marked as 1, then the area is classified as a building area; for all building areas, if the average coherence coefficient in S4 is marked as 1 and the number of straight lines with an inclination angle of 45° in S53 is marked as 0, then the area is classified as a vegetation area.

Images