Landslide identification method and device under complex background integrating polarization information

Abstract

This invention belongs to the field of fully polarimetric synthetic aperture radar (SAR) image processing technology, and specifically relates to a method and apparatus for identifying landslides in complex backgrounds by fusing polarization information. The method includes registering images with the same polarization combination as the pre-landslide single-polarimetric SAR image and the post-landslide PolSAR image to construct a dual-temporal SAR image pair; generating a coherence map through coherence change detection to extract the change region; extracting multiple polarization feature parameters from the preprocessed post-landslide PolSAR image; fully fusing these polarization feature parameters using weighted TOPSIS for landslide detection to obtain suspected landslide areas; and fusing the change region extracted from the coherence map and the suspected landslide area through a logical AND operation to obtain the final landslide area. This invention utilizes the superior-inferior solution distance method (TOPSIS) to fuse polarization information for landslide detection, combined with coherence change detection, which can effectively suppress false alarm rates and significantly improve the accuracy of landslide detection.

Description

Technical Field

[0001] The invention belongs to the technical field of polarization synthetic aperture radar (PolSAR) image processing, and in particular relates to a method and a device for identifying a landslide under a complex background by fusing polarization information. Background Art

[0002] Landslides are one of the most frequent and serious natural disasters. They can directly cause damage to buildings, interrupt transportation, and bring huge losses to residents' lives and property. Using synthetic aperture radar (SAR) remote sensing satellite images to detect the location and boundaries of post-disaster landslides and investigate the spatial distribution characteristics of landslides can provide important information for disaster prevention and reconstruction. There are two main ideas for landslide detection based on polarization decomposition. One is to use the multi-threshold method for landslide detection, that is, to set thresholds for multiple polarization features respectively, and the area that meets all threshold conditions is determined as the landslide area. The other is to use supervised classification or unsupervised classification methods to classify post-disaster PolSAR images to distinguish the areas belonging to landslides. Compared with the multi-threshold method, the latter can provide higher landslide detection accuracy due to the complex structure of the classifier. Therefore, the classification method is more commonly used for landslide identification than the multi-threshold method. However, such methods cannot effectively distinguish areas with similar scattering characteristics to landslides. In recent years, some scholars have combined Change Detection (CD) with Analytic Hierarchy Process (AHP) to realize landslide detection in complex terrain background. In this method, CD only uses the scattering information of PolSAR images, but does not use the phase information. Therefore, in areas with similar scattering mechanisms, this method may produce false detections; in the process of AHP fusing polarization information, the use of expert scoring method makes this method subjective. Therefore, a method to accurately identify landslide areas in a complex background with multiple types of landforms and better suppress false alarms caused by other landforms needs to be proposed. Summary of the invention

[0003] In view of the problems existing in the prior art, the present invention proposes a landslide identification method and device in a complex background by fusing polarization information. The landslide detection is performed by fusing polarization information using the superiority-inferior solution distance method (TOPSIS), and then combined with coherent change detection, which can effectively suppress the false alarm rate and greatly improve the accuracy of landslide detection.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] The present invention provides a method for identifying landslides in complex backgrounds by integrating polarization information, comprising the following steps:

[0006] The single-polarization SAR image before the landslide and the PolSAR image after the landslide with the same polarization combination as the single-polarization SAR image before the landslide are registered to construct a dual-phase SAR image pair, and a coherence map is generated by coherence change detection to extract the change area.

[0007] The post-landslide PolSAR images were preprocessed by radiation calibration, multi-viewing and terrain correction in sequence. Multiple polarization characteristic parameters were extracted from the preprocessed post-landslide PolSAR images, and the weighted TOPSIS was used to fully integrate the above polarization characteristic parameters for landslide detection. The area with the same scattering characteristics as the landslide surface was the suspected landslide area.

[0008] The change area extracted from the coherence map and the suspected landslide area are fused through the logical "AND" operation. The false alarm caused by the objects in the suspected landslide area with similar scattering characteristics to the real landslide area is suppressed by the change area extracted from the coherence map to obtain the final landslide area.

[0009] Furthermore, the coherence map is obtained using the dual-phase SAR image pair. The grayscale value in the coherence map indicates the degree of coherence. The mutation of the landslide will cause significant changes in the topography, that is, the coherence is low. The area that has not changed will maintain a high coherence. The coherence |γ| is calculated as follows:

[0010]

[0011] Where E[] represents the mathematical expectation, S1 is the SAR image before the landslide, S2 is an image selected from the PolSAR images after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, Represents the conjugate complex number of S2. The value range of coherence |γ| is [0,1]. The larger the |γ|, the lower the degree of change in a region, and the smaller the |γ|, the more obvious the change in the region. The coherence |γ| is used to perform threshold segmentation and morphological processing on the coherence map to obtain the significant change area.

[0012] Furthermore, the surface scattered power p is obtained by Yamaguchi decomposition: s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h ; The polarization entropy H, the average scattering angle α and the anti-entropy A are obtained by H / A / α decomposition; the real part of the correlation coefficient of the co-polarization component 0e(ρ hh-vv ).

[0013] Furthermore, the power corresponding to each scattering component is extracted from the polarization coherence matrix T through Yamaguchi decomposition, that is, the surface scattering power P s, secondary scattering power P d , volume scattering power P v and the helical scattering power P h ; The polarization coherence matrix T is expressed as a weighted combination of the scattered powers:

[0014] T=f s T s +f d T d +f v T v +f h T h

[0015] Among them, T s , T d , T v , Τ h are the polarization coherence matrices corresponding to the surface scattering, secondary scattering, volume scattering and helical volume scattering components respectively; f s 、f d 、f v 、f h P s , P d , P v and P h The decomposition coefficient of .

[0016] Furthermore, the polarization entropy H, the average scattering angle α and the anti-entropy A are obtained by H / A / α decomposition, including: under the single-station reciprocity condition, the polarization coherence matrix T is a 3×3 semi-positive Hermitian matrix, so the following eigenvalue decomposition is performed:

[0017]

[0018] Among them, the eigenvalue λ i , 1≤i≤3, satisfying λ1≥λ2≥λ3>0, λ i The corresponding eigenvector u i Mutually orthogonal, H represents the conjugate transpose; then the polarization entropy H, the average scattering angle α and the anti-entropy A are given by the eigenvalue λ i and the eigenvector u i The calculation results are:

[0019]

[0020] In the formula, is the probability of each scattering mechanism occurring, is the feature vector u i The corresponding scattering angle.

[0021] Furthermore, the co-polarization component correlation coefficient ρ hh-vvis the correlation coefficient between the two co-polarized scattering vectors of the target. The larger the value, the more uniform the scattering type of the target. Re(ρ hh-vv ) is as follows:

[0022]

[0023] Where Re(·) represents the real part, T 11 , T 12 and T 22 represents the corresponding element in the polarization coherence matrix T.

[0024] Furthermore, the method of using weighted TOPSIS to fully integrate polarization characteristic parameters for landslide detection includes the following steps:

[0025] Firstly, the polarization characteristic parameters are normalized and trended;

[0026] The eight polarization characteristic parameters are used as evaluation indicators, each pixel point of the PolSAR image is used as an evaluation object, and an evaluation matrix is ​​obtained;

[0027] Determine that the positive ideal sample of the evaluation matrix is ​​1, and the negative ideal sample is 0;

[0028] Calculate the distance between each pixel and the positive and negative ideal samples;

[0029] Calculate the relative closeness of each pixel to the positive ideal sample;

[0030] Determine the appropriate threshold to perform threshold segmentation on the relative proximity of each pixel in the PolSAR image, determine the pixels with relative proximity greater than the threshold as suspected landslide pixels, and then perform morphological processing to obtain the suspected landslide area.

[0031] Furthermore, the normalization and trending of the polarization characteristic parameters include:

[0032] For the scattered power p s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h Perform normalization:

[0033]

[0034] where the subscript x∈{s,d,v,h}, s, d, v and h represent surface scattering, secondary scattering, volume scattering and helical scattering, respectively; the normalized p d ,p v ,p h Perform positive trend processing:

[0035] p x =1-p x

[0036] The interval normalization processing is performed on the polarization entropy H and the average scattering angle α:

[0037]

[0038] For H, For α,

[0039] For the real part of the correlation coefficient of the co-polarized component Re(ρ hh-vv ), Re(ρ hh-vv ) is assigned a negative value of 0.

[0040] Furthermore, the weights of polarization characteristic parameters were determined using the AHP method. According to the geomorphological characteristics after the landslide, p s The surface scattering characteristics of landslides can be best reflected. H and α can better distinguish landslides from other landforms. Re(ρ hh-vv ) is often used to reflect the surface scattering characteristics of landforms, p d 、p v 、p h and A have little effect on identifying landslides; therefore, p s As the most important evaluation index, H and α are relatively important evaluation indicators, Re(ρ hh-vv ) is a generally important evaluation index, p d 、p v 、p h and A as relatively unimportant evaluation indicators; then, the importance comparison values ​​between different evaluation indicators are determined according to the pairwise comparison matrix importance scale table, and a judgment matrix is ​​constructed.

[0041] The present invention also provides a device for identifying landslides in complex backgrounds by integrating polarization information, comprising:

[0042] The change detection module is used to register the single-polarization SAR image before the landslide and the PolSAR image after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, construct a dual-phase SAR image pair, and generate a coherence map through coherent change detection to extract the change area;

[0043] The suspected landslide area extraction module is used to perform radiation calibration, multi-view and terrain correction preprocessing on the post-landslide PolSAR images in sequence; multiple polarization feature parameters are extracted from the preprocessed post-landslide PolSAR images, and the weighted TOPSIS is used to fully integrate the above polarization feature parameters for landslide detection. The area with the same scattering characteristics as the landslide surface is the suspected landslide area;

[0044] The landslide area determination module is used to fuse the change area extracted from the coherence map and the suspected landslide area through a logical "AND" operation, and suppress the false alarm caused by the objects in the suspected landslide area with similar scattering characteristics to the real landslide area through the change area extracted from the coherence map, so as to obtain the final landslide area.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] 1. Aiming at the problems of low accuracy and high false alarm rate of landslide detection in complex background by existing methods, the present invention proposes a landslide identification method in complex background by integrating polarization information. First, the change area is obtained by coherent change detection, and then the polarization information is integrated by TOPSIS. By integrating multiple polarization feature parameters, the suspected landslide area is detected based on the scattering characteristics of different landforms. The change area and the suspected landslide area are further integrated, and the final landslide area is extracted from the change area. A clear landslide area range can be obtained, the accuracy of landslide detection is greatly improved, and the false alarms caused by most landforms can be effectively suppressed.

[0047] 2. With the continuous development of PolSAR technology, the use of PolSAR images for disaster detection will play an increasingly important role. The method of the present invention can accurately and quickly extract the information of the disaster area after the landslide occurs without relying on prior information, providing a specific and feasible solution for landslide detection under complex backgrounds, and providing a reliable basis for post-disaster rescue, secondary disaster warning and post-disaster reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a flowchart of a method for identifying landslides in a complex background by integrating polarization information according to an embodiment of the present invention;

[0050] Figure 2 is a flow chart of determining a suspected landslide area by using TOPSIS according to an embodiment of the present invention;

[0051] Figure 31 is a Yamaguchi decomposition false color image and an optical image before and after a landslide according to an embodiment of the present invention, wherein (a) is a Yamaguchi decomposition false color image before the landslide, (b) is a Yamaguchi decomposition false color image after the landslide, (c) is an optical image before the landslide, and (d) is an optical image after the landslide;

[0052] Figure 4 It is a comparison diagram of the visual interpretation result of the embodiment of the present invention and the method of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] How to accurately identify landslide areas in a complex background with multiple types of landforms while effectively suppressing false alarms caused by other landforms is a key issue in landslide detection based on polarimetric synthetic aperture radar (PolSAR) images. Figure 1 As shown, this embodiment proposes a landslide identification method under complex background by integrating polarization information, introduces the idea of ​​integrating multiple polarization features and detecting coherent changes, and includes the following steps:

[0055] Step S11, registering the single-polarization SAR image before the landslide and the PolSAR image after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, constructing a dual-phase SAR image pair, and generating a coherence map through coherence change detection to extract the change area.

[0056] Step S12, the post-landslide PolSAR image is preprocessed by radiation calibration, multi-viewing and terrain correction in sequence; multiple polarization feature parameters are extracted from the preprocessed post-landslide PolSAR image, and the weighted TOPSIS is used to fully fuse the above polarization feature parameters for landslide detection, and the area with the same scattering characteristics as the landslide surface is the suspected landslide area.

[0057] Step S13, the change area extracted from the coherence map and the suspected landslide area are merged through a logical "AND" operation, and the false alarm caused by the objects in the suspected landslide area with scattering characteristics similar to those of the real landslide area are suppressed by the change area extracted from the coherence map to obtain the final landslide area, so as to achieve the purpose of accurately identifying the landslide area from a complex background with multiple types of objects, and at the same time better suppressing the false alarm caused by other objects.

[0058] Landslide is a sudden natural disaster, and the disaster area has very significant mutation, while unchanged landforms generally do not have this feature. Therefore, the changed area can be detected by analyzing the mutation of landforms. Significant changes in topography usually cause serious decoherence in SAR interferometric processing, that is, low coherence, while unchanged areas will maintain high coherence. Therefore, the coherence map can be used to determine whether a certain area has changed, thereby suppressing false alarms caused by unchanged landforms and making landslide detection results more accurate.

[0059] The coherence map is obtained using the dual-phase SAR image pair. The grayscale value in the coherence map indicates the degree of coherence. The coherence |γ| is calculated as follows:

[0060]

[0061] Where E[] represents the mathematical expectation, S1 is the SAR image before the landslide, S2 is an image selected from the PolSAR images after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, Represents the conjugate complex number of S2. The value range of coherence |γ| is [0,1]. The larger the |γ|, the lower the degree of change in a region, and the smaller the |γ|, the more obvious the change in the region. The coherence |γ| is used to perform threshold segmentation and morphological processing on the coherence map to obtain the significant change area.

[0062] In SAR images of different polarization combinations, vertical polarization backscatter can better reflect the landforms dominated by horizontal structures. Therefore, this example uses VV polarization combination images that can highlight the characteristics of bare soil and bare rock areas to generate the coherence map of the study area.

[0063] Specifically, the surface scattering power p is obtained by Yamaguchi decomposition: s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h ; The polarization entropy H, the average scattering angle α and the anti-entropy A are obtained by H / A / α decomposition; the real part Re(ρ hh-vv ).

[0064] (1)Yamaguchi decomposition

[0065] Yamaguchi decomposition is an incoherent decomposition algorithm based on a scattering model, which can well match the basic scattering mechanism of ground objects. This decomposition method extracts the power corresponding to each scattering component from the polarization coherence matrix T, that is, the surface scattering power P s , secondary scattering power P d , volume scattering power Pv and the helical scattering power P h ; The polarization coherence matrix T can be expressed as a weighted combination of the scattered powers:

[0066] T=f s T s +f d T d +f v T v +f h T h

[0067] Among them, T s , T d , T v , Τ h are the polarization coherence matrices corresponding to the surface scattering, secondary scattering, volume scattering and helical volume scattering components respectively; f s 、f d 、f v 、f h P s , P d , P v and P h The decomposition coefficient of .

[0068] (2) H / A / α decomposition

[0069] H / A / α decomposition is an incoherent decomposition algorithm that uses the eigenvalues ​​and eigenvectors of the covariance matrix or polarization coherence matrix to define polarization characteristics. Under the single-station reciprocity condition, the polarization coherence matrix T is a 3×3 semi-positive Hermitian matrix, so the following eigenvalue decomposition is performed:

[0070]

[0071] Among them, the eigenvalue λ i (1≤i≤3) satisfies λ1≥λ2≥λ3>0, λ i The corresponding eigenvector u i Mutually orthogonal, H represents the conjugate transpose; then the polarization entropy H, the average scattering angle α and the anti-entropy A are given by the eigenvalue λ i and the eigenvector u i The calculation results are:

[0072]

[0073] In the formula, is the probability of each scattering mechanism occurring, is the feature vector u i The corresponding scattering angle.

[0074] (3) Calculation of correlation coefficient

[0075] Co-polarization component correlation coefficient ρ hh-vv is the correlation coefficient between the two co-polarized scattering vectors of the target. The larger the value, the more uniform the scattering type of the target. For example, Re(ρ hh-vv ) is often used to reflect the surface scattering characteristics of ground objects. Re(ρ hh-vv ) is as follows:

[0076]

[0077] Where Re(·) represents the real part, T 11 , T 12 and T 22 represents the corresponding element in the polarization coherence matrix T.

[0078] TOPSIS is a commonly used method in multi-attribute decision analysis, also known as the approximate ideal method. It has no strict restrictions on data distribution and sample content, can make full use of original data information, and the evaluation results are relatively accurate and reliable. This method ranks the degree of proximity between a limited number of evaluation objects and ideal samples according to a series of attribute conditions (i.e., evaluation indicators). In step S12, weighted TOPSIS is used to fully integrate the above polarization characteristic parameters for landslide detection, such as Figure 2 As shown, the following steps are included:

[0079] Step S121, normalizing and isotrending the polarization characteristic parameters.

[0080] Step S122, assuming that M evaluation indicators are used to evaluate N evaluation objects respectively, where the value of the i-th (i=1,2,3,…,N) evaluation object on the j-th (j=1,2,3,…,M) evaluation indicator is z NM , the evaluation matrix can be obtained as follows:

[0081]

[0082] In this example, each pixel point of the PolSAR image is taken as an evaluation object, each polarization characteristic parameter is taken as an evaluation index of the evaluation object score, and each pixel point of the PolSAR image is evaluated respectively using the eight polarization characteristic parameters.

[0083] Step S123, define the positive and negative ideal samples of the evaluation matrix as:

[0084]

[0085]

[0086] In order to ensure the applicability of TOPSIS to different PolSAR image data, the positive ideal sample is determined to be 1 and the negative ideal sample is determined to be 0.

[0087] Step S124: The distances between the ith evaluation object (i.e., the ith pixel point) and the positive and negative ideal samples are:

[0088]

[0089]

[0090] Where w i The weight representing each evaluation index (ie, each polarization characteristic parameter) can be determined by different methods according to different requirements.

[0091] Step S125, calculate the final score of the evaluation object, that is, its relative closeness to the positive ideal sample is:

[0092]

[0093] Relative closeness S i The size of S reflects the gap between the evaluation object and the ideal sample. i The larger the value, the closer the evaluation object is to the positive ideal sample. i The smaller it is, the farther the evaluation object is from the ideal sample. The similarity between each pixel point in the PolSAR image and the landslide, that is, the relative closeness, is accurately calculated by the above formula.

[0094] Step S126, determine a suitable threshold to perform threshold segmentation on the relative proximity of each pixel point in the PolSAR image, determine the pixel points with relative proximity greater than the threshold as suspected landslide pixels, and then perform morphological processing to obtain the suspected landslide area.

[0095] Morphological processing: The binary image of the suspected landslide area is processed by first "opening" and then "closing" operations to eliminate the interference of irrelevant structures in the image. The white spots (such as salt and pepper noise) in the black background are first removed by opening operation, and then the black holes in the white area are eliminated by closing operation, the adjacent white areas are connected and their boundaries are smoothed, so that the boundary features of the suspected landslide area are more obvious.

[0096] Because the eight polarization characteristic parameters (p s 、p d 、p v 、p h 、H、A、α、Re(ρ hh-vv)) As evaluation indicators, differences in directionality, measurement units, numerical evaluation methods, and evaluation value ranges will have an adverse effect on the evaluation results. Therefore, it is necessary to normalize and trend these polarization characteristic parameters before constructing the evaluation matrix, so that the values ​​of each parameter are within the interval [0,1] and are positively correlated with the scattering characteristics of the landslide, that is, the larger the value, the more it can reflect the scattering characteristics of the landslide. The normalization and trending of the polarization characteristic parameters in step S121 include:

[0097] (1) For the scattered power p s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h Perform normalization:

[0098]

[0099] Wherein, the subscript x∈{s,d,v,h}, s, d, v and h represent surface scattering, secondary scattering, volume scattering and spiral volume scattering respectively; since the exposed soil covered by the surface after the landslide usually presents in the form of surface scattering, which is very different from other landforms, so p s The larger the value, the d ,p v ,p h The smaller the value of is, the more it can reflect the surface scattering type of the landslide landform. Therefore, the normalized p d ,p v ,p h Perform positive trend processing:

[0100] p x =1-p x

[0101] (2) For the polarization entropy H and the average scattering angle α, when H is in the range of 0.52-0.63 and α is in the range of 29-37, the scattering characteristics of the landslide can be better reflected. The polarization entropy H and the average scattering angle α are normalized in intervals:

[0102]

[0103] For H, For α,

[0104] (3)Re(ρ hh-vv ) is often used to identify the type of surface scattering objects. The larger the value, the more significant the surface scattering type of the object. hh-vv ) is assigned a negative value of 0.

[0105] In step S124, i represents the weight of each polarization characteristic parameter. Different polarization characteristic parameters have different effects on the final evaluation, so they need to be assigned different weights. Here, this embodiment uses the AHP method to determine the weight of the polarization characteristic parameter. According to the geomorphological characteristics after the landslide disaster, p s The surface scattering characteristics of landslides can be best reflected. H and α can better distinguish landslides from other landforms. Re(ρ hh-vv ) is often used to reflect the surface scattering characteristics of landforms, p d 、p v 、p h and A have little effect on identifying landslides; therefore, p s As the most important evaluation index, H and α are relatively important evaluation indicators, Re(ρ hh-vv ) is a generally important evaluation index, p d 、p v 、p h and A as relatively unimportant evaluation indicators; then the importance comparison values ​​between different evaluation indicators are determined according to the pairwise comparison matrix importance scale table, and the judgment matrix is ​​constructed as shown in Table 1. In order to eliminate the subjectivity in the judgment matrix as much as possible, it is necessary to introduce the consistency ratio (CR) as a reference. When CR < 0.1, the judgment matrix is ​​considered reasonable. After calculation, the judgment matrix CR shown in Table 1 is 0, indicating that the weight distribution of each polarization feature parameter is reasonable.

[0106] Table 1 Judgment matrix for determining weights using the AHP method

[0107]

[0108] A specific experiment is given below to better illustrate the present invention.

[0109] (1) Research area and data source

[0110] Study area: On December 20, 2015, a landslide caused by heavy rain occurred in Shenzhen Hengtaiyu Industrial Park (22.7065°~22.7289°N, 113.9220°~113.9467°E). The landslide occurred in a complex geomorphic background (including five basic features: landslide, water system, forest, residential area, and field), and there were other changes caused by natural or human activities. Therefore, the use of this area as the research object is representative. The satellite data are shown in Table 2.

[0111] Table 2 Basic parameters of satellite data

[0112]

[0113] Yamaguchi decomposition false color image and optical image of the study area are shown in Figure 3 As shown, the red color of the false color image represents secondary scattering, green represents volume scattering, and blue represents surface scattering. There is a certain difference in the acquisition time of the optical images and PolSAR images before and after the disaster, but they can still be used as a reference for analyzing and discussing the changes in topography caused by landslides. After visual interpretation, five types of ground feature samples, including landslides, water systems, forests, residential areas, and fields, were selected using white boxes on the optical images and Yamaguchi decomposition maps to facilitate the analysis of the detection accuracy below.

[0114] (2) Quantitative analysis of landslide detection performance

[0115] In order to quantitatively analyze the effectiveness of the method of the present invention, Figure 3 The correct rate and false alarm rate of landslide detection (i.e., the total number of pixels detected as landslides in the sample area of ​​ground objects other than landslides divided by the total number of pixels in the sample area) are calculated for different ground object pixel samples in the marked area, as shown in Table 3.

[0116] Table 3. Landslide detection accuracy and false alarm rate

[0117]

[0118] (3) Analysis and discussion of landslide detection result diagram

[0119] In order to analyze the results of landslide detection by different methods more intuitively, the corresponding result graphs in Table 3 are Figure 4 Given in.

[0120] The landslide area is obtained by comparing the Yamaguchi false color image and the optical image visual interpretation. The visual result is the result determined by humans, which is used as the judgment basis for whether the results obtained by each detection method are correct. Area 1 represents the landslide area, area 2 represents the field, and area 3 represents the forest. The method of the present invention can effectively separate the landslide from the residential area dominated by secondary scattering and the forest dominated by body scattering. At the same time, there are small black holes scattered in the landslide area. This is because there are a large number of large-scale machinery (such as excavators, forklifts, earthmoving vehicles, etc.) in the area during disaster relief activities. The mechanical arm, the vehicle body and the ground form a secondary scattering structure, which causes certain interference to the landslide detection, but does not affect the determination of the large-scale landslide range. The landslide area obtained by the method of the present invention is more complete and more connected than the landslide detection method combined with the analytic hierarchy process mentioned in the background technology, and the false alarm of most of the objects in the background is better suppressed. For example, the lower circular frame of area 1 represents the area where the scattering type does not change before and after the landslide. The existing detection method may not detect that the area is a landslide area, but the method of the present invention can detect that the area is a landslide area. However, the method of the present invention still has a small number of false alarms in the field (region 2), which is due to the low coherence of the region caused by the growth of crops. The false alarms caused by the changes here are eliminated by the coherence map. The forest coverage rate of region 3 is low, and there are a small number of bare soil areas in the middle. Because these areas have not changed significantly before and after the landslide, the method of the present invention can better suppress the false alarms here through coherent change detection. Finally, it is verified that the method of the present invention is consistent with the visual results. The existing detection method is likely to mistakenly detect these landforms such as fields and bare soil as landslide areas, but the method of the present invention can eliminate these false alarms through coherence detection, thereby improving the accuracy of landslide detection.

[0121] Corresponding to the above-mentioned method for identifying a landslide in a complex background by integrating polarization information, this embodiment further proposes a device for identifying a landslide in a complex background by integrating polarization information, including:

[0122] The change detection module is used to register the single-polarization SAR image before the landslide and the PolSAR image after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, construct a dual-phase SAR image pair, and generate a coherence map through coherent change detection to extract the change area;

[0123] The suspected landslide area extraction module is used to perform radiation calibration, multi-view and terrain correction preprocessing on the post-landslide PolSAR images in sequence; multiple polarization feature parameters are extracted from the preprocessed post-landslide PolSAR images, and the weighted TOPSIS is used to fully integrate the above polarization feature parameters for landslide detection. The area with the same scattering characteristics as the landslide surface is the suspected landslide area;

[0124] The landslide area determination module is used to fuse the change area extracted from the coherence map and the suspected landslide area through a logical "AND" operation, and suppress the false alarm caused by the objects in the suspected landslide area with similar scattering characteristics to the real landslide area through the change area extracted from the coherence map, so as to obtain the final landslide area.

[0125] It should be noted that, in this article, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0126] Finally, it should be noted that the above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for identifying landslides in complex backgrounds by integrating polarization information, characterized in that: The following steps are involved: Step 1: register the single-polarization SAR image before the landslide and the PolSAR image after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, construct a dual-phase SAR image pair, and generate a coherence map through coherence change detection to extract the change area; Step 2: Perform radiation calibration, multi-view and terrain correction preprocessing on the post-landslide PolSAR image in sequence; extract multiple polarization feature parameters from the preprocessed post-landslide PolSAR image, and use weighted TOPSIS to fully integrate the above polarization feature parameters for landslide detection, and obtain the area with the same scattering characteristics as the landslide surface as the suspected landslide area; Multiple polarization characteristic parameters are extracted from the pre-processed post-landslide PolSAR images, including: The surface scattered power p is obtained by Yamaguchi decomposition s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h ; The polarization entropy H, the average scattering angle α and the anti-entropy A are obtained by H / A / α decomposition; the real part Re(ρ hh-vv ); The landslide detection is carried out by fully integrating polarization characteristic parameters using weighted TOPSIS, which includes the following steps: Firstly, the polarization characteristic parameters are normalized and trended; The eight polarization characteristic parameters are used as evaluation indicators, each pixel point of the PolSAR image is used as an evaluation object, and an evaluation matrix is ​​obtained; Determine that the positive ideal sample of the evaluation matrix is ​​1, and the negative ideal sample is 0; Calculate the distance between each pixel and the positive and negative ideal samples; Calculate the relative closeness of each pixel to the positive ideal sample; Determine the appropriate threshold to perform threshold segmentation on the relative proximity of each pixel in the PolSAR image, determine the pixels with relative proximity greater than the threshold as suspected landslide pixels, and then perform morphological processing to obtain the suspected landslide area; Step 3: The change area extracted from the coherence map and the suspected landslide area are merged through a logical "AND" operation. The false alarm caused by the objects in the suspected landslide area with similar scattering characteristics to the real landslide area is suppressed by the change area extracted from the coherence map to obtain the final landslide area.

2. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: The coherence map is obtained by using the dual-phase SAR image pair. The gray value in the coherence map indicates the degree of coherence. The mutation of the landslide will cause significant changes in the topography and landforms, with low coherence. The area that has not changed will maintain a high coherence. The coherence |γ| is calculated as follows: Where E[] represents the mathematical expectation, S1 is the SAR image before the landslide, S2 is an image selected from the PolSAR images after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, Represents the conjugate complex number of S2. The value range of coherence |γ| is [0,1]. The larger the |γ|, the lower the degree of change in a region, and the smaller the |γ|, the more obvious the change in the region. The coherence |γ| is used to perform threshold segmentation and morphological processing on the coherence map to obtain the significant change area.

3. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: The power corresponding to each scattering component is extracted from the polarization coherence matrix T through Yamaguchi decomposition: surface scattering power P s , secondary scattering power P d , volume scattering power P v and the helical scattering power P h ; The polarization coherence matrix T is expressed as a weighted combination of the scattered powers: T=f s T s +f d T d +f v T v +f h T h Among them, T s , T d , T v , Τ h are the polarization coherence matrices corresponding to the surface scattering, secondary scattering, volume scattering and helical volume scattering components respectively; f s 、f d 、f v 、f h P s , P d , P v and P h The decomposition coefficient of .

4. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: The polarization entropy H, the average scattering angle α and the anti-entropy A are obtained by H / A / α decomposition, including: Under the single-station reciprocity condition, the polarization coherence matrix T is a 3×3 semi-positive Hermitian matrix, so the following eigenvalue decomposition is performed: Among them, the eigenvalue λ i , 1≤i≤3, satisfying λ1≥λ2≥λ3>0, λ i The corresponding eigenvector u i Mutually orthogonal, H represents the conjugate transpose; then the polarization entropy H, the average scattering angle α and the anti-entropy A are given by the eigenvalue λ i and the eigenvector u i The calculation results are: In the formula, is the probability of each scattering mechanism occurring, is the feature vector u i The corresponding scattering angle.

5. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: Co-polarization component correlation coefficient ρ hh-vv is the correlation coefficient between the two co-polarized scattering vectors of the target. The larger the value, the more uniform the scattering type of the target. Re(ρ hh-vv ) is as follows: Where Re(·) represents the real part, T 11 , T 12 and T 22 represents the corresponding element in the polarization coherence matrix T.

6. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: The normalization and trending of the polarization characteristic parameters include: For the scattered power p s , secondary scattering power p d , volume scattering power p v and the helical scattering power p h Perform normalization: where the subscript x∈{s,d,v,h}, s, d, v and h represent surface scattering, secondary scattering, volume scattering and helical scattering, respectively; the normalized p d ,p v ,p h Perform positive trend processing: p x =1-p x The interval normalization processing is performed on the polarization entropy H and the average scattering angle α: For H, For α, For the real part of the correlation coefficient of the co-polarized component Re(ρ hh-vv ), Re(ρ hh-vv ) is assigned a negative value of 0.

7. The method for identifying landslides in complex backgrounds by integrating polarization information according to claim 1, characterized in that: The weights of polarization characteristic parameters were determined by AHP method. According to the geomorphological characteristics after landslide disaster, p s The surface scattering characteristics of landslides can be best reflected. H and α can better distinguish landslides from other landforms. Re(ρ hh-vv ) is often used to reflect the surface scattering characteristics of landforms, p d 、p v 、p h and A have little effect on identifying landslides; therefore, p s As the most important evaluation index, H and α are relatively important evaluation indicators, Re(ρ hh-vv ) is a generally important evaluation index, p d 、p v 、p h and A as relatively unimportant evaluation indicators; then, the importance comparison values ​​between different evaluation indicators are determined according to the pairwise comparison matrix importance scale table, and a judgment matrix is ​​constructed.

8. A device for identifying landslides in complex backgrounds by integrating polarization information, characterized in that: The device is used to implement the method for identifying landslides in complex backgrounds by integrating polarization information as claimed in claim 1, and comprises: The change detection module is used to register the single-polarization SAR image before the landslide and the PolSAR image after the landslide that has the same polarization combination as the single-polarization SAR image before the landslide, construct a dual-phase SAR image pair, and generate a coherence map through coherent change detection to extract the change area; The suspected landslide area extraction module is used to perform radiation calibration, multi-view and terrain correction preprocessing on the post-landslide PolSAR images in sequence; multiple polarization feature parameters are extracted from the preprocessed post-landslide PolSAR images, and the weighted TOPSIS is used to fully integrate the above polarization feature parameters for landslide detection, and the area with the same scattering characteristics as the landslide surface is obtained as the suspected landslide area; The landslide area determination module is used to fuse the change area extracted from the coherence map and the suspected landslide area through a logical "AND" operation. The change area extracted from the coherence map suppresses the false alarm caused by the objects in the suspected landslide area with similar scattering characteristics to the real landslide area, and obtains the final landslide area.

Images