1. A multi-temporal multi-polarization SAR landslide extraction method is characterized by comprising the following steps:

s1, acquiring time sequence backscattering coefficients of SAR data in different time phases and different polarization modes of a landslide and a background area and analyzing the time sequence backscattering coefficients;

s2, combining SAR data of different time phases and different polarization modes, wherein the characteristic difference of backscattering coefficients of the landslide and the background area reaches a set value, and forming a composite image of the landslide and the background area;

s3, acquiring backscattering coefficient characteristic values of the landslide and the background area based on the backscattering coefficient characteristic statistical result of the landslide and the background area in the composite image, determining the ground object type in the composite image and segmenting the ground object type;

and S4, eliminating false objects or false background information existing in the result after the segmentation processing by adopting a binary morphology method.

2. The multi-temporal multi-polarization SAR landslide extraction method of claim 1, further comprising before step S1: and performing multi-view, registration, geocoding and radiometric calibration pretreatment on the time sequence complete polarization SAR image.

3. The method for extracting the multi-temporal multi-polarization SAR landslide according to claim 1, wherein the specific step of analyzing the time sequence backscattering coefficients of SAR data of different time phases and different polarization modes of the landslide and the background area in step S1 comprises: and acquiring a time sequence change curve of the backscattering coefficient of the landslide area and backscattering coefficient change curves of surrounding ground object image element points, and acquiring backscattering coefficient difference curves of the landslide and the background area in different polarization modes.

4. The multi-temporal multi-polarization SAR landslide extraction method according to claim 1, wherein the step S2 specifically comprises: firstly, taking an SAR image as a gray image, converting the gray image into a color image, then adopting SAR data as a data source to carry out RGB color synthesis operation, and combining information of all connected regions to form a composite image.

5. The multi-temporal multi-polarization SAR landslide extraction method of claim 1, wherein in step S4, a SVM-based segmentation algorithm is adopted to segment the composite image.

6. The multi-temporal multi-polarization SAR landslide extraction method of claim 5, wherein the specific step of segmenting the composite image based on the SVM segmentation algorithm comprises:

generating training samples according to the heterogeneity of pixels in the composite image, wherein the training samples comprise four categories of mountain bare-leaking areas, landslide areas, vegetation and water bodies;

extracting attribute features of the training samples to train the SVM classifier;

and inputting the obtained characteristic values of the backscattering coefficients of the landslide and the background area into an SVM classifier to obtain a target recognition result.

7. The multi-temporal multi-polarization SAR landslide extraction method according to claim 5, wherein the step S4 is specifically an operation of performing expansion-first and corrosion-later on the segmentation result by using a closed operation in a binary morphology method.

Background

The development of the SAR sensor greatly increases the terrain feature information contained in SAR data, and drives the deep revolution of the SAR application field. The polarized SAR (Polarimetric SAR, polarisar) technology is mature, and the development of a technology for realizing the rapid extraction of a target region based on the electromagnetic scattering mechanism modeling of a typical feature is receiving wide attention. In addition to the advantage of single polarization, PolSAR also contains more ground object information, such as backscattering coefficient, polarization degree, homopolarity ratio, cross-polarization ratio, scattering entropy, polarization phase difference and the like in any polarization state, the same ground object and different polarized electromagnetic waves can generate different reactions, and multi-polarization SAR data can record detailed ground object scattering difference. Therefore, the identification capability of the SAR image to the information is greatly improved by introducing the polarization information, and the polarization SAR data shows better application potential in the fields of ship identification, land classification, sea ice detection and the like.

The landslide is a natural geological disaster with strong burst property, high concealment property and large uncertainty factor, and how to acquire information such as position, form and area of the landslide disaster by using SAR image data is a precondition for carrying out accurate disaster assessment on the landslide and taking correct countermeasures. In the process of landslide inoculation, the ground objects in the landslide area gradually change until the landslide occurs, and the objects in the area change greatly or even absolutely. So far, the following are mainly included in the SAR landslide disaster identification: 1. identifying and extracting a target area by using the intensity information of the SAR image; 2. identifying a landslide area by utilizing an SAR and optical image fusion technology; 3. and (4) extracting the landslide area by combining a time sequence InSAR technology with a coherence coefficient. From the analysis, in the past, the combination of SAR intensity information and optical images or coherence coefficient information is mainly focused on SAR landslide extraction, and landslide extraction by adopting a ground feature classification method by utilizing a polarization decomposition technology is also realized, however, the polarization characteristics of a landslide region are complicated due to the complexity and uncertainty of ground features of the landslide region, so that the selection of effective and representative polarization characteristics for identifying the landslide region is an important precondition for realizing the correct extraction of the landslide region.

Most landslides need to be subjected to long-time deformation accumulation, surface ground objects of the landslides can be changed in the same way in the landslide forming process, before the landslide area collapses, a disaster area and a peripheral area of the disaster area are often covered by vegetation, the difference in polarized SAR images is not obvious, after the landslide occurs, plants in the landslide area drop off, rock gravels drop off, the ground objects are changed greatly, the change of the ground objects in areas except the edge of the landslide is not obvious and still covered by the vegetation, and therefore the difference between a background (the vegetation covered area except the edge) and a target (the landslide area) is formed after the landslide. Meanwhile, the vegetation in different periods has different effects on radar waves, and is reflected in the SAR images as the difference of attributes such as intensity, phase, backscattering coefficient and the like. The method is an effective means for acquiring images containing more comprehensive information of the ground features by efficiently and comprehensively utilizing the detailed features of the ground features in different SAR images and combining complementary information in each image.

Based on the above factors, it is necessary to develop a method for extracting a landslide region by using polarization features in SAR images of different time phases and different polarization modes before and after occurrence of a landslide region disaster.

Disclosure of Invention

The invention aims to provide a multi-temporal multi-polarization SAR landslide extraction method to overcome the defects in the prior art.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a multi-temporal multi-polarization SAR landslide extraction method comprises the following steps:

s1, acquiring time sequence backscattering coefficients of SAR data in different time phases and different polarization modes of a landslide and a background area and analyzing the time sequence backscattering coefficients;

s2, combining SAR data of different time phases and different polarization modes, wherein the characteristic difference of backscattering coefficients of the landslide and the background area reaches a set value, and forming a composite image of the landslide and the background area;

s3, acquiring backscattering coefficient characteristic values of the landslide and the background area based on the backscattering coefficient characteristic statistical result of the landslide and the background area in the composite image, determining the ground object type in the composite image and segmenting the ground object type;

and S4, eliminating false objects or false background information existing in the result after the segmentation processing by adopting a binary morphology method.

Further, before step S1, the method further includes: and performing multi-view, registration, geocoding and radiometric calibration pretreatment on the time sequence complete polarization SAR image.

Further, the specific step of analyzing the time-series backscattering coefficient of the SAR data of different polarization modes in different time phases of the landslide and the background area in step S1 includes: and acquiring a time sequence change curve of the backscattering coefficient of the landslide area and backscattering coefficient change curves of surrounding ground object image element points, and acquiring backscattering coefficient difference curves of the landslide and the background area in different polarization modes.

Further, the step S2 specifically includes: firstly, taking an SAR image as a gray image, converting the gray image into a color image, then adopting SAR data as a data source to carry out RGB color synthesis operation, and combining information of all connected regions to form a composite image.

Further, in step S4, an SVM-based segmentation algorithm is used to segment the composite image.

Further, the specific step of segmenting the composite image by the SVM-based segmentation algorithm includes:

generating training samples according to the heterogeneity of pixels in the composite image, wherein the training samples comprise four categories of mountain bare-leaking areas, landslide areas, vegetation and water bodies;

extracting attribute features of the training samples to train the SVM classifier;

and inputting the obtained characteristic values of the backscattering coefficients of the landslide and the background area into an SVM classifier to obtain a target recognition result.

Further, the step S4 is specifically to perform an operation of expanding the segmentation result first and then performing the etching operation by using a closed operation in a binary morphology method.

Compared with the prior art, the invention has the advantages that: the invention provides a method for extracting landslide by using landslide polarization characteristics of different polarization SAR image data, aiming at the problem of difficulty in landslide extraction caused by the characteristics of burstiness, complexity and concealment of landslide disaster. Firstly, in order to increase detail changes of a landslide and surrounding areas thereof in an SAR image, firstly, acquiring different time-phase multi-polarization SAR image data with the shortest time interval after the landslide occurs; secondly, in order to explain the effectiveness of landslide extraction according to polarization characteristics, the method provided by the invention performs morphological processing on the target area extracted by the SVM segmentation algorithm, and extracts a relatively complete landslide area.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a specific flowchart of the multi-temporal multi-polarization SAR landslide extraction method of the present invention.

Fig. 2 is a characteristic diagram of backscattering coefficients of landslide and background regions in SAR images with different time phases and different polarizations in the present invention.

Fig. 3a is a composite image of SAR images with different time phases and different polarizations in the present invention, and fig. 3b is a diagram of the form and position of a landslide.

FIG. 4 is a flowchart of an image segmentation algorithm based on a support vector machine according to the present invention.

FIG. 5 is a diagram of binary morphology results in the present invention.

FIG. 6 is a diagram showing the result of extracting a landslide area according to the present invention.

Detailed Description

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, and the scope of the present invention will be more clearly and clearly defined.

Referring to fig. 1, the embodiment discloses a multi-temporal multi-polarization SAR landslide extraction method, including: firstly, obtaining a time sequence backscattering coefficient of a target ground object in SAR original image data which is preprocessed through multi-vision, registration, geocoding, radiometric calibration and the like, and then analyzing backscattering coefficient characteristics of a landslide and a background area in different time phase multi-polarization SAR images, wherein the time sequence backscattering coefficient specifically comprises the following steps: analyzing backscattering coefficients of landslide and background area regions in SAR images in different time phases and different polarization modes, firstly obtaining a time sequence change curve of the backscattering coefficients of the landslide area and backscattering coefficient change curves of surrounding ground object image element points, and then obtaining backscattering coefficient difference curves of the landslide and the background area regions in different polarization modes; secondly, acquiring SAR images with different time phases and different polarization modes which can highlight differences of the landslide and the background region through combined operation according to the backscattering coefficient characteristics of the landslide region and the background region, and performing combined operation on the SAR images; then, based on SVM image segmentation, based on the backscattering coefficient characteristic statistical result of the background and the target area in the composite image, obtaining the backscattering coefficient characteristic value of the target in the background, determining the ground object type in the composite image and carrying out segmentation processing on the ground object type; and finally, performing binary morphological processing, performing closed operation and open operation on respective topological information and structure information of pseudo target information in a target hole and a background region in a threshold segmentation result, and filling and eliminating pseudo information pixel points.

The present invention is further described below by specific embodiments, the sample data of the present embodiment adopts landslide disasters in southwest mountainous areas, and the backward scattering coefficient of the present embodiment is used as a basis for landslide identification, mainly because after landslide, along with the falling off of other loosely covered ground objects such as vegetation, the loose covered ground objects are mostly replaced by bare rocks, gravels, and the like, the backward scattering coefficient of the landslide area in the SAR image is greatly changed, and the ground objects around the landslide are only changed in a limited manner, and are mostly covered by vegetation, the backward scattering coefficients of rocks and gravels are higher, and the backward scattering coefficient of vegetation is lower, so that an obvious background is formed and the attribute comparison of the target is formed. According to the backscattering coefficient characteristics, which are the difference results generated by the action of different ground objects and different polarized radar waves, different time phases and different polarized SAR data capable of enhancing attribute characteristic contrast are selected for backscattering combined operation, so that the difference between a landslide area and a surrounding background area is further enhanced.

The method comprises the steps of firstly, obtaining backscattering coefficients of SAR data of different polarization modes of all time phases of a research area. The ground target is formed by a plurality of scatterers, the scatterers carry out backscattering in all directions generated by radar incident waves, and the backscattering waves of the scatterers are coherently superposed to generate speckle noise in the SAR image. The existence of speckle noise inevitably causes noise interference and the correctness of feature extraction of the feature backscattering coefficient of the ground object. In order to improve the accuracy of the subsequent landslide and background region backscatter coefficient extraction and threshold segmentation, before the operation process of the invention is carried out, preprocessing such as multi-view, registration, geocoding, radiometric calibration and the like is firstly carried out on the time sequence full polarization SAR image, so that the time sequence multi-polarization SAR image of the same pixel point corresponding to the same ground object resolution unit is obtained. Sequencing according to the time sequence of the images, and performing time sequence analysis by combining the backscattering coefficients of the pixels of each scene image data to obtain the backscattering coefficients of SAR data of different polarization modes of each time phase of the object in the research area.

And secondly, analyzing the time sequence backscattering coefficient of the landslide and the background area. Selecting multi-polarization SAR data in different time phases according to the backscattering coefficient characteristics of the landslide and the background area acquired in the first step, wherein the landslide area selected in the embodiment is a high-speed high-level remote landslide-debris flow pattern. The height above sea level of the crack at the rear edge of the landslide is as high as 3400m, the horizontal distance of the rear edge collapsing and sliding forwards is as long as 3000m, and the long distance sliding forwards enables the mountain collapsing at the rear edge to form debris flow which is accumulated in a new village. Therefore, the landslide of the present embodiment is integrally divided into three segments: the back edge collapse area, the middle debris flow area and the debris accumulation area are basically bare leaked rocks, gravels and the like. According to the time sequence backscattering coefficients of the background and the target of the research area, after the landslide occurs, the difference between the backscattering coefficients of the background and the target area is gradually reduced along with the time extension, as shown in fig. 2, the backscattering coefficients are integrally distributed between-20 db and 5db, wherein the backscattering coefficient of the target area in the VV polarized SAR data of 19 days in 6 months and 26 days in 5 months is the lowest, the value is distributed between-15 db and-10 db, and the backscattering coefficient is lower mainly because vegetation in the landslide area grows vigorously before the landslide occurs. In 11 days 7 and 7 months in 2017, backscattering coefficients of landslides and background areas are obviously different, the backscattering coefficients of the landslide areas are distributed between 6db and 10db, surrounding background areas are still distributed between-15 db and-10 db, the backscattering coefficient of a VH polarization SAR data target area is 10db at most, and the backscattering coefficients are mainly caused by the fact that vegetation in the landslide areas fall off, land objects obviously change, and new vegetation does not appear after the landslides occur, so that the backscattering coefficients of the landslide areas are greatly increased. According to the characteristics of each polarization and each phase backscattering coefficient in fig. 2, the present invention selects VV and VH polarization SAR data corresponding to phases of 26 days in 5 months, 19 days in 6 months, and 11 days in 7 months as the source data of the third step.

And thirdly, combining and processing SAR images with different time phases and different polarization modes. According to the difference of backscattering coefficients of a target area and a background area in the multiple polarized SAR data in each time period acquired in the first step, selecting VV and VH polarized SAR data corresponding to 0526, 0619 and 0711 phases as a multiband image to be analyzed, namely, a method for highlighting the target area or weakening the background area based on the difference of the same ground object in different phases and different polarized SAR images. The invention makes the landslide and the background area present specific colors through the false color synthesis of SAR images with different polarizations and different time phases, thereby achieving the purpose of highlighting the target information. The specific implementation steps are as follows: firstly, taking the SAR image as a gray image, and converting the gray image into a color image according to the following formula:

(i，j)＝α₁R(i，j)+α₂G(i，j)+α₃B(i，j) (1)

wherein R (I, j), G (I, j), B (I, j) respectively represent the brightness values of red, green and blue, alpha, of three channels of I (I, j) corresponding to the color composite image₁、α₂、α₃Respectively, the contribution of each luminance value in the color image.

And adopting VV and VH polarization SAR data which are screened out in the first step and correspond to 0526, 0619 and 0711 phases as a data source in the step to perform RGB color synthesis operation, and combining information of all connected regions. Since each image is registered in advance, the R, G, B channel designation is performed only according to the backscatter coefficient characteristics of the target and the feature in each image to be synthesized, wherein the backscatter coefficient of the target region in 7 months is the largest, and is displayed in red, and the images in 5 months and 6 months are respectively corresponding to blue and green. The composite image is shown in fig. 3a, and fig. 3b is a diagram showing the form and position of a landslide.

And fourthly, carrying out image segmentation based on the SVM. And 4, image segmentation is to segment the landslide and the background region information in the image, and the invention classifies the pixel points in the composite image obtained in the step three by using a support vector machine. The SVM belongs to supervised classification, and the specific classification flow is shown in FIG. 4. Firstly, a training sample is selected, the training sample category is set to 4 types according to the heterogeneity of pixels in the composite image, the training sample category corresponds to a mountain bare leakage area, a landslide area, vegetation and a water body respectively, and the final classification result is shown in fig. 5. The classification of different attribute pixel points in the composite image is realized based on SVM algorithm, and the hyperplane between each category needs to be determined and distinguishedThe function needs to determine the parameter alpha between variables_iWherein α is_iThe solution formula of (2) is generally:

wherein x is_i，x_jFor inputting variable values, K (x)_i，x_j) Is a kernel function, y_iThe class flag for the input variable value is a feature that a certain class has when it is +1, and represents a non-class feature when it is-1. Then, on the basis, a classification decision function is obtained:

the specific operation is as follows:

input training set T { (x)₁，y₁)，(x₂，y₂)，...，(x_n，y_n) In which x_iIs the i-th feature vector, y_iThe class mark represents a feature of a certain class when the class mark is +1, and represents a non-class feature when the class mark is-1. The hyperplane and the classification decision tree are separated by these input variables. The invention selects the kernel function K (x, z) type as the Gaussian kernel function type in the training process:the penalty function is set to 10, and the optimal parameters in the classifier are constructed and solved according to the formula:

wherein, 0 is less than or equal to alpha_iC, i ═ 1, 2, N, and the optimal parameter solution is obtained according to the above formulaSelecting alpha on the basis of the steps^*A positive component ofCalculation of b^*：

On the basis of determining the kernel function type and obtaining the parameters, a classification decision function is determined, and the classification decision function adopted in the embodiment is as follows:

and fifthly, processing the binary morphology of the segmentation result. Because the segmentation result obtained in the third step contains a large amount of pseudo target or pseudo background pixel point information, and meanwhile, a large amount of missing pixels exist in a landslide region, the embodiment firstly adopts closed operation in a binary morphology method to perform expansion-first and corrosion-later operation on the segmentation result, the method can make up for narrower gaps and slender ravines, the segmentation result has partially missing pixels, and the expansion operation method can fill the missing pixels in the target region, eliminate small holes among objects, and fill up the fracture in the contour line segment. The corrosion operation is opposite to the expansion operation, the contour of a target object in an image can be thinned while discrete points are eliminated by corrosion, the image contour is smoothed, partial protruding areas are removed, and then redundant areas of the target area are reduced. The specific operation expression is as follows:

dst＝close(src，elemennt)＝erode(src，element) (8)

wherein, src is a segmentation image based on SVM, element is a structural element, and each pixel of srt is scanned by scattering of the structural element; on the basis, performing OR operation by using the reflection of the structural element and the binary image covered by the structural element; whether the pixel is 0 or 255 is judged according to the pixel value. After the expansion operation is finished, carrying out corrosion operation, and scanning each element of the image by using the structural element; carrying out AND operation by using the structural elements and the binary image covered by the structural elements; when the pixels are all 1, the pixel value is 1, otherwise, the pixel value is 0. The binary morphology operation result of the present invention is shown in fig. 5. The target region is superimposed in the composite image of the investigation region as shown in fig. 6.

The main advantages of the invention are: the backscattering coefficient is used as the basis for identifying the landslide, and is mainly because after the landslide, along with the falling of other loose covering ground objects such as vegetation and the like, the loose covering ground objects such as bare leaked rocks, gravels and the like are replaced, the backscattering coefficient of the landslide area in an SAR image is greatly changed, the ground objects around the landslide are only changed in a limited way and are still mostly covered by the vegetation, the backscattering coefficient of the rocks and the gravels is higher, and the backscattering coefficient of the vegetation is lower, so that obvious attribute comparison between a background and a target is formed. According to the backscattering coefficient characteristics, which are the difference results generated by the action of different ground objects and different polarized radar waves, different time phases and different polarized SAR data capable of enhancing attribute characteristic contrast are selected for backscattering combined operation, so that the difference between a landslide area and a surrounding background area is further enhanced.

The invention adopts SVM segmentation algorithm to obtain respective backscattering coefficient characteristic values of the background and the target by counting backscattering coefficients of the target and the background in the composite image, and further determines the category to be segmented in the composite image so as to segment the target and the background area. According to the invention, the topological information and the structural information of the target and background areas are considered, and the interference of the pseudo target and the pseudo background pixel point on the threshold segmentation precision is greatly eliminated by adopting a binary morphology processing method. Pixel points with similar backscattering coefficients of the target area and the background area can be effectively separated under the binary morphology processing, so that the integrity of the target area is improved, and the extraction result is fuller.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, various changes or modifications may be made by the patentees within the scope of the appended claims, and within the scope of the invention, as long as they do not exceed the scope of the invention described in the claims.