A Method for Landslide Detection in Full Polarimetric SAR Images

Abstract

The present invention discloses a method for detecting landslides in full-polarization SAR images, which relates to the technical field of landslide detection and includes the following steps: collecting a single full-polarization SAR image of a landslide target area and performing data preprocessing; based on the data preprocessing results, using the depolarization degree as a constraint and combining double-sided angle scattering and surface scattering to construct a landslide identification detector; using the generalized gamma distribution model to characterize the statistical information of the landslide identification detector; using a morphological operator to denoise the binary image obtained from the characterized landslide identification detector to obtain the landslide detection result, thereby completing the landslide detection in full-polarization SAR images. The present invention solves the technical bottleneck in the prior art that it is difficult to quickly and accurately obtain the location distribution of landslides based on a single full-polarization SAR image.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide detection, and particularly to a method for detecting landslides in full-polarization SAR images. Background Art

[0002] Landslides usually occur after major triggering events such as earthquakes or heavy rainfall. Accurate and timely landslide mapping is of great significance for post-disaster rescue, disaster assessment, and ensuring personal and property safety. To achieve this goal, high-resolution optical images have become an important means for extracting landslide information after disasters. However, high-resolution optical images are often blocked by clouds and fog, making it difficult to quickly provide reliable information after disasters. Synthetic aperture radar (SAR) has the ability to observe the earth all-weather and all-time, and has gradually become the mainstream tool for landslide identification research. Existing SAR image-based landslide identification methods usually rely on multiple SAR images as inputs and require a large number of ground truth samples for training, which limits their application in scenarios with scarce data or high timeliness requirements. In addition, most studies have not fully explored the potential of extracting landslide information from single-temporal SAR images.

[0003] Full-polarization SAR images show advantages in landslide mapping. By alternately transmitting and receiving microwaves with horizontal (H) and vertical (V) polarizations, they make up for the lack of information in single-polarization SAR data. However, most current studies based on polarimetric SAR images use methods such as change detection and coherence analysis, which require high-resolution data before the event, such as pre-event optical images, SAR images, and digital elevation models (DEMs). However, pre-event high-resolution data is not easy to obtain, which has become the main obstacle to timely obtaining landslide distribution information. A few studies have tried to use supervised classification algorithms to process full-polarization SAR images to extract landslide distribution information. This method can, to a certain extent, use a single full-polarization SAR image for landslide identification. However, it requires a large amount of sample point data for training, and the model training process takes a long time, and it cannot quickly respond to post-disaster mapping requirements. Therefore, there is still much room for improvement in the existing technology in terms of efficient, accurate, and pre-event data-independent landslide mapping capabilities. Summary of the Invention

[0004] In view of the above deficiencies in the prior art, the present invention provides a method for detecting landslides in full-polarization SAR images to solve the problems existing in the prior art.

[0005] To achieve the above invention objective, the technical solution adopted by the present invention is: A method for detecting landslides in full-polarization SAR images, comprising the following steps:

[0006] S1: Collect a single full-polarization SAR image of the landslide target area and perform data preprocessing;

[0007] S2: Based on the data preprocessing results, taking the depolarization degree as a constraint, and combining double-sided angle scattering and surface scattering, construct a landslide recognition detector;

[0008] S4: Using the generalized gamma distribution model to characterize the statistical information of the landslide recognition detector;

[0009] S4: Use morphological operators to denoise the binary image obtained from the characterized landslide recognition detector to obtain the landslide detection result, and complete the landslide detection of the fully polarized SAR image.

[0010] Furthermore, the data preprocessing in S1 includes multi-look processing, geocoding operation, and primary denoising processing;

[0011] The multi-look processing is: reducing random noise by superimposing multiple views of the SAR image to improve the signal-to-noise ratio of the SAR image. The pixel value after multi-look processing is:

[0012]

[0013] where, is the number of views in the azimuth direction, is the number of views in the range direction, is the azimuth index and the range index corresponding pixel value;

[0014] The geocoding operation is: converting the SAR image from the range-azimuth coordinate system to the geographic coordinate system for subsequent geographic positioning and analysis of the landslide area;

[0015] The primary denoising processing is: using a non-local filter to remove speckle noise from a single fully polarized SAR image of the landslide target area.

[0016] Furthermore, the landslide recognition detector in S2 is:

[0017]

[0018]

[0019] where, represents the landslide recognition detector, is the double-sided angle scattering, is the surface scattering, is the depolarization degree, is the maximum value in the Stokes vector, , , and are the Stokes vectors.

[0020] Further, the probability density function of the generalized gamma distribution model in S3 is as follows:

[0021]

[0022] wherein, is the power parameter, is the shape parameter, is the scale parameter, is the variable in the generalized gamma distribution model, is the Gamma function.

[0023] Further, the detection threshold of the generalized gamma distribution model in S3 is as follows:

[0024]

[0025] wherein, is the inverse incomplete Gamma function, is the given false alarm probability value.

[0026] Further, the morphological operator in S4 is as follows:

[0027]

[0028] wherein, is the opening operation, is the closing operation.

[0029] The beneficial effects of the present invention are as follows: 1) Using mathematical formulas to enhance the difference between landslide targets and their backgrounds, significantly improving the distinguishability of landslide features, thereby enhancing the recognition effect; 2) Combining scattering mechanism analysis and polarization anisotropy features, the present invention can effectively overcome the technical bottleneck that it is difficult to quickly and accurately obtain the landslide position distribution based on a single full-polarization SAR image in the prior art; 3) It can meet the rapid response requirements in landslide disaster monitoring and recognition, providing real-time support for disaster emergency handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flowchart of a method for detecting landslides in full-polarization SAR images.

[0031] Figure 2 is a structural block diagram of a method for detecting landslides in full-polarization SAR images.

[0032] Figure 3 is a landslide detection result diagram of this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] As Figure 1 and Figure 2 shown, a method for detecting landslides in fully polarized SAR images includes the following steps:

[0035] S1: Collect a single fully polarized SAR image of the landslide target area and perform data preprocessing;

[0036] S2: Based on the data preprocessing results, use the depolarization degree as a constraint, and combine double-sided angle scattering and surface scattering to construct a landslide recognition detector;

[0037] S3: Use the generalized gamma distribution model to characterize the statistical information of the landslide recognition detector;

[0038] S4: Use morphological operators to denoise the binary image obtained from the characterized landslide recognition detector to obtain the landslide detection result, and complete the landslide detection in the fully polarized SAR image.

[0039] The basic idea of the present invention is based on the polarization characteristics of landslides in fully polarized SAR images, capture the performance of landslide targets in terms of scattering mechanisms and polarization anisotropy characteristics, use mathematical basic formulas to enhance the significant difference between the target and the background, and achieve precise acquisition of landslide targets by introducing the generalized gamma distribution model ( ), thereby solving the problem that it is difficult to quickly capture landslide targets in fully polarized SAR images.

[0040] The data preprocessing in S1 includes multi-look processing, geocoding operation, and primary denoising processing;

[0041] The multi-look processing is: reduce random noise by superimposing multiple views of the SAR image to improve the signal-to-noise ratio of the SAR image. The pixel value after multi-look processing is:

[0042]

[0043] wherein, is the number of views in the azimuth direction, is the number of views in the range direction, is the azimuth index and the range index corresponding pixel value;

[0044] The geocoding operation is: convert the SAR image from the range-azimuth coordinate system to the geographic coordinate system for subsequent geographic positioning and analysis of the landslide area;

[0045] The initial denoising process is as follows: a non-local filter is used to remove speckle noise from a single full-polarization SAR image of the landslide target area.

[0046] In this embodiment, first, multi-look processing is performed according to the characteristics of the full-polarization SAR image to reduce random noise in the image and improve the signal-to-noise ratio; geocoding operation is carried out to accurately correspond the image data with the actual ground geographical state; finally, a non-local filter is used to denoise the image, the filtering window is set to 11×11, and the image quality is further improved by removing speckle noise.

[0047] Based on the relatively strong scattering characteristics of the landslide shown in the full-polarization SAR image, the present invention introduces the depolarization degree ( ) as a constraint, combines the dihedral angle scattering and the surface scattering to construct a landslide recognition detector.

[0048] The landslide recognition detector in S2 is:

[0049]

[0050]

[0051] Among them, represents the landslide recognition detector, is the dihedral angle scattering, is the surface scattering, is the depolarization degree, is the maximum value in the Stokes vector, , , and are the Stokes vectors.

[0052] The normalized value of is between 0 and 1. When the polarized wave interacts with the random medium, the polarization state of the backscattered wave will change, and the change of the purity of the scattered wave with the change of the polarization of the transmitted wave causes the change of the wave entropy. This method overcomes the interference of strong scatterers such as buildings, reduces the computational complexity, and thus effectively improves the efficiency of the algorithm.

[0053] After defining the new detector, it is necessary to map the statistical information of the detector to characterize the adaptive detection of the landslide in different scenarios. In the present invention, is used to characterize the statistical behavior of the landslide detector.

[0054] The probability density function of the generalized gamma distribution model in S3 is:

[0055]

[0056] Among them, is the power parameter, is the shape parameter, is the scale parameter, is the variable in the generalized gamma distribution model, is the Gamma function.

[0057] The detection threshold of the generalized gamma distribution model in S3 is:

[0058]

[0059] Among them, is the inverse incomplete Gamma function, is the given false alarm probability value.

[0060] The above steps effectively utilize the detection framework to detect landslides on fully polarized SAR images. However, since the binary images obtained by the landslide detector are often distorted by salt-and-pepper noise, especially in fully polarized SAR images, where the speckle noise is independently distributed in the image and causes significant changes in reflectivity. The present invention adopts morphological image processing to eliminate defects by considering the form and structure of the image.

[0061] That is, morphological operators are introduced (i.e., opening operation and closing operation ) to solve problems such as internal holes and blurred boundaries and overcome the limitations of fixed structural elements at the edges of the target area. In the binary image, the opening operation removes small white objects (such as salt-and-pepper noise) from the black background, while the closing operation fills small black holes or connects isolated points within the white target area to smooth its edges.

[0062] The morphological operator in S4

[0063]

[0064] Among them, is the opening operation, is the closing operation.

[0065] These operations do not significantly change the area of the target region, but can effectively reduce noise and produce better landslide detection output. Through the implementation of the above content, the accurate extraction of landslide targets can be efficiently completed based on full-polarimetric SAR images, breaking through the technical bottleneck of existing polarimetric SAR landslide identification methods. This method not only improves the accuracy of landslide identification, but also significantly speeds up the response speed of landslide detection, thus forming a technical approach for rapid landslide identification based on full-polarimetric SAR images, providing a reliable and efficient solution, and providing strong technical support for landslide disaster monitoring and emergency response.

[0066] In one embodiment of the present invention, as Figure 3 shown, the landslide area is detected by the method of the present invention. The white area in the figure represents the detected landslide area, while the black area represents the non-landslide area. It can be seen that this method can accurately distinguish the landslide area from the background area and has a high recognition accuracy, fully reflecting the effectiveness and applicability of this method in landslide detection.

[0067] The present invention specifically addresses the technical difficulties that landslide targets are widely distributed, have complex shapes and are difficult to effectively identify in actual scenarios. It fully considers the scattering mechanism and polarization anisotropy characteristics of landslide targets in full-polarimetric SAR images, constructs basic mathematical formulas for feature calculation, and finally realizes the accurate acquisition of landslide targets through the generalized gamma distribution model ( ). This method does not require the introduction of machine learning technology to obtain sample data and perform long-term training, thus further realizing the rapid and automatic identification of landslide targets.

[0068] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the invention.

Claims

1. A method for detecting landslides in fully polarimetric SAR images, characterized in that, It includes the following steps: S1: Collect a single full-polarization SAR image of the landslide target area and perform data preprocessing; S2: Based on the data preprocessing results, use the depolarization degree as a constraint, and combine double-sided angle scattering and surface scattering to construct a landslide recognition detector; The landslide recognition detector in S2 is: Among them, represents a landslide recognition detector, is double-sided angular scattering, is surface scattering, is the depolarization degree, is the maximum value in the Stokes vector, and and and are the Stokes vectors; S3: Use the generalized gamma distribution model to characterize the statistical information of the landslide recognition detector; The probability density function of the generalized gamma distribution model in S3 is as follows: Among them, is the power parameter, is the shape parameter, is the scale parameter, is the variable in the generalized gamma distribution model, is the Gamma function; The detection threshold of the generalized gamma distribution model in S3 is as follows: wherein, is the inverse incomplete Gamma function, is the given false alarm probability value; S4: Use morphological operators to denoise the binary image obtained from the characterized landslide recognition detector to obtain the landslide detection result, and complete the landslide detection of the full-polarization SAR image.

2. The landslide detection method for full-polarization SAR images according to claim 1, characterized in that, The data preprocessing in S1 includes multi-look processing, geocoding operation, and primary denoising processing; The multi-look processing is as follows: reducing random noise and increasing the signal-to-noise ratio of the SAR image by superimposing multiple views of the SAR image. The pixel value after multi-look processing is as follows: Among them, is the number of views in the azimuth direction, is the number of views in the range direction, is the azimuth index and the range index corresponding pixel value; The geocoding operation is: Convert the SAR image from the range-azimuth coordinate system to the geographic coordinate system for subsequent geographic positioning and analysis of the landslide area; The primary denoising processing is: Use a non-local filter to remove speckle noise from the single full-polarization SAR image of the landslide target area.

3. A method for landslide detection in fully polarized SAR images according to claim 1, characterized in that The morphological operator in S4 is as follows: Among them, is the opening operation, is the closing operation.

Images