Method, device, electronic device and storage medium for establishing landslide identification knowledge base

Abstract

The present invention relates to a method, device, electronic device and storage medium for establishing a knowledge base for landslide identification. The method comprises: using InSAR technology to analyze SAR image data sets of each target area, and determining a time series deformation map and a cumulative deformation map of each target area; based on the slope map and the slope aspect map of the target area, jointly applying hotspot clustering technology and spatial clustering analysis technology, analyzing deformation points in the time series deformation map of the target area, and determining the landslide boundary of the target area; based on a pre-set interpretation mark, performing remote sensing visual interpretation on the optical image data set, the time series deformation map, the cumulative deformation map and the landslide boundary of the target area, and determining the landslide attribute information of the target area. The present invention takes into account the influence of slope aspect and slope on landslide identification, improves the accuracy of the acquired landslide attribute information, and then improves the data accuracy of the landslide identification knowledge base, and lays a good foundation for subsequent AI automatic identification.

Description

Technical Field

[0001] The invention relates to the technical field of landslide remote sensing image recognition, and in particular to a method, device, electronic equipment and storage medium for establishing a landslide recognition knowledge base. Background Art

[0002] Landslide disaster is a sudden and destructive natural disaster. For the purpose of safety protection, it is necessary to conduct risk investigation, hidden danger investigation, early identification and early warning treatment in areas prone to landslides, so as to reduce or avoid the occurrence of regional landslide disasters. Due to the limitation that manual interpretation of landslides requires a lot of manpower and time, automatic identification is generally used to identify landslides. The automatic identification of landslides requires a large amount of landslide sample data as support. Therefore, the construction of a landslide sample library is of great research significance for the automatic identification of landslides.

[0003] At present, satellite remote sensing technology is widely used in identifying potential landslides due to its large detection area, rich time series, multi-resolution and affordable price. It interprets optical remote sensing images to achieve the purpose of identifying landslides. However, when interpreting optical remote sensing images, it over-relies on the characteristics of image color, texture and other element information, resulting in low accuracy in identifying landslides when the quality of the original images obtained is low. Synthetic aperture radar (SAR) is gradually being used in identifying potential landslides due to its all-day, all-weather, strong ability to penetrate vegetation cover and millimeter-level measurement accuracy of surface deformation. It usually uses D-InSAR, M-SAR, etc. AI, PSI and other interpretation technologies are used to interpret radar images to improve the recognition accuracy of landslides. However, since landslides usually erode the soft debris of the base during movement, the volume of the landslide increases and then causes ultra-long-distance movement. In this case, due to the influence of time decoherence, terrain and atmospheric delay, the above interpretation technologies may only be able to detect local deformation on the surface of the landslide, and cannot accurately determine the landslide boundary, which may cause the emergence of pseudo-landslides and increase the redundancy of the determination results. Therefore, directly using optical remote sensing images marked with landslide information or radar images marked with landslide information to establish a landslide sample library will cause the performance of the landslide sample library to fail to meet the expected requirements. Summary of the invention

[0004] In view of the problems existing in the prior art, the embodiments of the present invention provide a method, device, electronic device and storage medium for establishing a landslide identification knowledge base that overcomes the above problems or at least partially solves the above problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for establishing a landslide identification knowledge base, comprising:

[0006] The SAR image data set of each target area is analyzed by using the InSAR technology to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; and there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set;

[0007] Based on the slope map and aspect map of the target area, the hotspot clustering technology and spatial clustering analysis technology are combined to analyze the deformation points in the time series deformation map of the target area and determine the landslide boundary of the target area.

[0008] Based on the pre-set interpretation marks, remote sensing visual interpretation is performed on the optical image data set, time series deformation map, cumulative deformation map and landslide boundary of the target area to determine the landslide attribute information of the target area;

[0009] The landslide attribute information of the target area is marked on the SAR image of the target area using a marking tool, and a landslide identification knowledge base is generated based on the marked SAR image of the target area.

[0010] According to a landslide identification knowledge base provided by the present invention, based on the slope map and the aspect map of the target area, the hotspot clustering technology and the spatial clustering analysis technology are jointly applied to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area, specifically: the hotspot clustering technology is used to perform hotspot analysis on the deformation points in the time series deformation map of the target area to obtain the landslide hazard points in the target area;

[0011] Based on the P values, Z scores and deformation rates corresponding to the landslide risk points in the target area, the landslide risk points in the target area are preliminarily screened; wherein the P values ​​and Z scores corresponding to the landslide risk points in the target area are determined in the process of hot spot analysis of the deformation points in the time series deformation map of the target area using the hot spot clustering technology; the deformation rates corresponding to the landslide risk points in the target area are determined in the process of analyzing the SAR image data sets of each target area using the InSAR technology;

[0012] Based on the slope map and aspect map of the target area, the spatial clustering analysis technology is used to re-screen the landslide hazard points corresponding to the target area after the preliminary screening, and the landslide boundary of the target area is determined based on the landslide hazard points corresponding to the target area after the re-screening.

[0013] According to a landslide identification knowledge base provided by the present invention, the landslide potential points in the target area are initially screened based on the P value, Z score and deformation rate corresponding to each of the landslide potential points in the target area, specifically:

[0014] Landslide potential points that do not meet any of the following conditions: the absolute value of the annual deformation rate is greater than 2 mm, the P value is less than 0.01, and the absolute value of the Z score is greater than 2.58 are eliminated.

[0015] According to a landslide identification knowledge base provided by the present invention, based on the slope map and the aspect map of the target area, the spatial clustering analysis technology is used to re-screen the landslide hazard points corresponding to the target area after the preliminary screening, specifically:

[0016] Based on the slope map and aspect map of the target area, weights are assigned to the landslide potential points corresponding to the target area after preliminary screening;

[0017] The Z scores of the landslide hazard points corresponding to the target area are weighted by using the weights of the landslide hazard points corresponding to the target area;

[0018] The spatial cluster analysis technique is used to analyze the Z scores of the landslide hazard points corresponding to the weighted target area, and the landslide hazard points that do not constitute landslides in the analysis results are eliminated.

[0019] According to a landslide identification knowledge base provided by the present invention, weighting of landslide hazard points corresponding to the target area after preliminary screening based on the slope map and aspect map of the target area needs to meet the following conditions at the same time:

[0020] When the slope of the landslide potential point is less than 15°, the slope is greater than 65°, or the slope direction of the landslide potential point is inconsistent with the slope direction of the overall landslide trend, the weight of the landslide potential point is assigned to 0;

[0021] In other cases, the weight assigned to the landslide potential point is proportional to the size of the slope of the landslide potential point.

[0022] According to a landslide identification knowledge base provided by the present invention, the detection bands include: C band and L band;

[0023] The landslide attribute information includes: landslide detailed information and radar data information; wherein the landslide detailed information includes: landslide name, landslide type, landslide location, longitude and latitude of the landslide center, landslide sliding direction, landslide impact range, maximum sliding speed, landslide displacement characteristics and landslide boundary; the radar data information includes: data source, band information, resolution, orbit direction, incident angle and polarization mode of SAR image;

[0024] The optical image data set includes: a topographic map of the target area, a regional outline map, a geological disaster point distribution map, a hydrological factor map, a geological map and high-resolution optical remote sensing data;

[0025] The pre-set interpretation marks include: direct interpretation marks and indirect interpretation marks; wherein the direct interpretation marks include: vegetation characteristics, hydrological characteristics and topography; the indirect interpretation marks include: shape characteristics, color tone characteristics and texture characteristics.

[0026] According to a landslide identification knowledge base provided by the present invention, the marking tool includes: a CVAT marking tool.

[0027] In a second aspect, the present invention further provides a device for establishing a landslide identification knowledge base, comprising:

[0028] A SAR image data set analysis module is used to analyze the SAR image data set of each target area using the InSAR technology to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; and there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set;

[0029] The landslide boundary determination module is used to analyze the deformation points in the time series deformation map of the target area based on the slope map and the aspect map of the target area, and to determine the landslide boundary of the target area by combining the hotspot clustering technology and the spatial clustering analysis technology;

[0030] The remote sensing visual interpretation module is used to perform remote sensing visual interpretation of the optical image data set, time series deformation map, cumulative deformation map and landslide boundary of the target area based on the pre-set interpretation marks to determine the landslide attribute information of the target area;

[0031] The marking and generating module is used to mark the landslide attribute information of the target area on the SAR image of the target area by using a marking tool, and to generate a landslide identification knowledge base based on the marked SAR image of the target area.

[0032] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for establishing a landslide identification knowledge base as described in the first aspect are implemented.

[0033] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for establishing a landslide identification knowledge base as described in the first aspect are implemented.

[0034] The method, device, electronic device and storage medium for establishing a landslide identification knowledge base provided by the present invention take into account the effects of slope direction and slope on landslide identification, improve the accuracy of the acquired landslide attribute information, and further improve the data accuracy of the landslide identification knowledge base, laying a good foundation for subsequent AI automatic identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0036] Figure 1 It is a flow chart of the method for establishing a landslide identification knowledge base provided by the present invention;

[0037] Figure 2 It is a flow chart of multi-temporal, multi-viewing angle and multi-band coordinated microwave remote sensing data processing provided by the present invention;

[0038] Figure 3 It is a landslide boundary schematic diagram of the XXX landslide provided by the present invention;

[0039] Figure 4 It is a structural schematic diagram of a device for establishing a landslide identification knowledge base provided by the present invention;

[0040] Figure 5 It is a structural schematic diagram of an electronic device for implementing a method for establishing a landslide identification knowledge base provided by the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0042] Combine the following Figure 1-Figure 5 The invention describes a method, device, electronic device and storage medium for establishing a landslide identification knowledge base.

[0043] First, as Figure 1 As shown, the present invention provides a method for establishing a landslide identification knowledge base, comprising:

[0044] S11, using InSAR technology to analyze the SAR image data set of each target area, and determine the time series deformation map and cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; and there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set;

[0045] It is important to understand that the SAR satellites currently in orbit have three bands: X, C, and L. The wavelength of the L band is 23.5 cm, the wavelength of the C band is 5.6 cm, and the wavelength of the X band is 3.1 cm. Landslides often occur in areas with undulating terrain, dense vegetation, cloudy weather, and heavy rain. Due to the high water vapor content in the atmosphere, path delays will occur in the radar electromagnetic waves during their propagation in the atmosphere. The difference in atmospheric delay between two observations will affect the accuracy of deformation monitoring. Considering that the atmospheric delay is inversely proportional to the radar wavelength, that is, the longer the wavelength, the less affected by the atmospheric delay, the X-band data is most seriously affected by the atmospheric delay, followed by the C-band data, and the L-band is relatively less affected by the atmospheric delay. At the same time, due to the dense vegetation coverage in these areas, the longer the radar wavelength, the stronger its penetration ability to vegetation. The echo of L-band data mainly comes from the stem layer of vegetation, while the echo of X and C-band mainly comes from the blade layer of vegetation. Because the blade layer is more likely to change over time relative to the stem layer, the data of L-band can maintain a higher coherence for a long time in areas with dense vegetation coverage, C-band is second, and X-band is the weakest. Therefore, L-band and C-band data have higher coherence than X-band data, and are suitable for long-term surface deformation monitoring. Therefore, the present invention selects the method of collaborative monitoring of L-band and C-band.

[0046] In addition, the interaction between microwaves and the surface is very complex, and different angles will produce different reflections. Low angles of incidence usually return stronger signals, and as the angle of incidence increases, the return signal gradually weakens. Depending on the height of the radar from the surface, the angle of incidence will change from short distance to long distance, which in turn affects the imaging geometry. Synthetic aperture radar can select a suitable angle of incidence to adjust the radar antenna according to different beam scanning methods to obtain different imaging widths. The intensity of the radar reflection will also decrease as the angle of incidence increases. The backscatter coefficient of the radar image is also affected by the angle of incidence. The angle of incidence is also called the angle of view, which is the angle between the radar beam and the vertical surface line. By adjusting the angle of incidence, the backscatter coefficient of the image can be adjusted. Selecting a suitable angle of incidence can better extract surface information to monitor landslides.

[0047] In the technical field of the present invention, different satellite sensors have different wavelengths and different anti-decoherence capabilities. Jointly solving multi-source sensor data to increase the number of observations and improve the time resolution is the mainstream method for realizing InSAR deformation monitoring in low-coherence areas. Figure 2 The following is an example of a multi-temporal, multi-view and multi-band coordinated microwave remote sensing data processing flow chart: Figure 2 As shown, the present invention conducts multi-phase, multi-band, and multi-angle coordinated monitoring of the target area to improve time coherence, thereby improving the accuracy of landslide identification in the target area.

[0048] It is also necessary to understand that in mountainous areas with complex and changeable terrain and dense vegetation, when using time-series InSAR technology to monitor landslide deformation, multi-view processing can easily lead to pixel aliasing and coherence estimation bias. It is necessary to conduct long-term deformation analysis of the target area in a targeted manner;

[0049] The core idea of ​​PS-InSAR is to obtain stable point targets that are not affected by temporal and spatial incoherence, namely permanent scatterer points (PS points), through multi-view radar images covering the same area. PS points are usually selected on relatively stable objects such as houses, roads, bridges, rocks, etc. However, in areas with complex terrain and dense vegetation, temporal and spatial incoherence is serious, resulting in a small number of selectable PS points, which will bring certain difficulties to phase unwrapping and deformation solution, resulting in the observation results not being able to objectively reflect the overall changes in the study area.

[0050] The solution model based on distributed scatterer points can obtain high-density point information in non-artificial surface areas. Distributed scatterer points correspond to uniform areas distributed on a group of pixels in SAR images (such as pastures, shrubs and bare land). Compared with traditional PS-InSAR technology, DS-InSAR technology significantly increases the density of point targets, especially in areas with sparse vegetation, overcomes the deficiency of small number of permanent scatterer points, and greatly increases the number of observation points.

[0051] The present invention combines DS-InSAR technology with PS-InSAR technology, and uses permanent scatterer points (PS points) and distributed scatterer points (DS points) at the same time, which greatly increases the number of coherent points in the study area, and provides effective monitoring means for landslide monitoring in complex terrain areas. Taking a certain area with complex terrain and dense vegetation as an example, there are 856,514 permanent scatterer points selected using amplitude discreteness alone, which are mainly distributed on buildings in urban areas, and there are 6,774,263 DS points selected using KS test alone, which are evenly distributed in the study area, which is 8 times the number of PS points. It can be seen that only using permanent scatterer points selected using amplitude discreteness based on traditional PS-InSAR technology can no longer meet the actual needs of InSAR landslide deformation monitoring in this area, and more high coherent points are needed to solve the deformation information of the surface. Combining the permanent scatterer points selected by amplitude discreteness and the DS points selected by KS test, the number of high coherent points can be greatly increased.

[0052] It should be noted that when the target area is an area with undulating terrain, dense vegetation, cloudy and rainy, the deformation analysis will often be interfered by time incoherence, terrain influence and atmospheric delay, resulting in the inability to accurately identify the deformation of the landslide body. Therefore, when using InSAR technology, appropriate interference elimination should be performed to ensure the accuracy of the deformation analysis results.

[0053] Among them, incoherence is the main error source that affects the quality of InSAR interferograms and deformation monitoring, and determines the ability to use time-series InSAR related technologies to monitor landslide deformation. Temporal incoherence and spatial incoherence are very important sources of incoherence for InSAR technology. The incoherence caused by the spatial position difference generated when the SAR satellite flies in repeated orbits is called spatial incoherence; and the incoherence caused by the change in the scattering characteristics of surface objects during the satellite's revisit period is called temporal incoherence. The quality of InSAR interferometric phase can be improved from the perspective of improving the quality of parameter inversion, thereby solving the problem of temporal incoherence.

[0054] Areas with complex terrain and large undulations bring two problems to time-series InSAR landslide monitoring: on the one hand, the overlap and shadow effects in the SAR imaging process are increased, making it impossible to observe the deformation of the slope in the overlap and shadow areas; on the other hand, in areas with large undulations, the error of the input DEM (digital elevation model data) will be relatively large. The accuracy of the landslide deformation estimation can be improved by eliminating the error used in the initial DEM.

[0055] Cloudy and rainy areas are affected by atmospheric circulation and terrain, so that when microwave signals emitted by satellite radar systems pass through the atmosphere to reach ground targets, the non-uniform atmosphere will affect the propagation of radar signals. This effect is manifested as atmospheric phase delay, which is one of the most important factors affecting the accuracy of interferometric phase. Therefore, in the process of estimating landslide deformation parameters, it is necessary to consider the effective estimation and suppression of atmospheric delay phase. Atmospheric phase can be filtered in the time domain using a triangular window filter at each scatterer point based on its high-pass characteristics in the time domain and low-pass characteristics in the space domain to extract the high-frequency components in the time domain, and filtered in the space domain on each interference pair to extract the low-frequency components in the space domain to obtain the atmospheric disturbance phase at the scatterer point, thereby estimating the atmospheric disturbance phase at all pixel points on all interference pairs, and removing the calculated atmospheric phase from the differential interferometric phase map, so that the atmospheric effect is effectively suppressed.

[0056] The InSAR technology of the present invention utilizes basic processing algorithms such as image registration, resampling, and differential fringe pattern generation to form the core algorithm for deformation parameter inversion; interferometric measurement is performed using permanent scatterer points (PS points) and distributed scatterer points (DS points), thereby providing an important basis for long-term series deformation analysis of landslide bodies.

[0057] S12. Based on the slope map and aspect map of the target area, the hotspot clustering technology and the spatial clustering analysis technology are combined to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area;

[0058] In the technical field, a time-series deformation map can clearly reflect the area where deformation occurs, and all pixels in the area are regarded as deformation points.

[0059] In addition, it can be understood that spatial clustering analysis can identify spatial clusters with deformable information. The present invention takes into account that the probability of landslide occurring when the landslide body is at different slopes and slope directions is also different (for example, different slope directions will not constitute the same landslide body). Outside a certain slope threshold range, there will be no influence of factors such as landslides. The terrain elements such as slope and slope direction are used as an element to assist in detecting the landslide boundary, thereby improving the accuracy of landslide boundary identification.

[0060] S13, based on the pre-set interpretation marks, performing remote sensing visual interpretation on the optical image data set, the time series deformation map, the cumulative deformation map and the landslide boundary of the target area to determine the landslide attribute information of the target area;

[0061] In the technical field of the present invention, natural landslides vary greatly due to the external forces, topography, lithology and climate environment, so the landslide forms are also various, which adds a lot of difficulty to the interpretation of landslide disasters. In order to better identify the geological disaster of landslide, geologists believe that a fully developed landslide should contain the following elements: landslide body, landslide bed, landslide surface, landslide wall, landslide tongue, landslide step, landslide depression, etc. Table 1 gives a specific explanation of landslide elements; but in fact not all landslides have these elements, but landslide body, landslide surface and landslide wall all have them.

[0062] Table 1

[0063]

[0064] Among them, the landslide body, landslide wall and landslide boundary are the important elements for interpreting remote sensing images. If the resolution of the image is quite high, elements such as the landslide step dividing line and landslide depression can also be seen.

[0065] The present invention utilizes pre-set interpretation marks to conduct comprehensive and in-depth mining of landslide attribute information from radar data information (time series deformation diagram, cumulative deformation diagram and landslide boundary) and optical data information (optical image data set) of the target area, so as to improve the recognition accuracy of landslide attribute information in the target area, thereby achieving the effect of improving the data accuracy of the landslide identification knowledge base.

[0066] S14, using a marking tool to mark the landslide attribute information of the target area on the SAR image of the target area, and generating a landslide identification knowledge base based on the marked SAR image of the target area.

[0067] The present invention adopts Computer Vision Annotation Tool (CVAT) for annotation. CVAT is a free, online and interactive image annotation tool for computer vision, which can be used for annotation of model training data. Compared with other annotation software, CVAT is more suitable for multi-person collaboration, can be used for almost all computer vision (CV) related annotation tasks, and has been updated and optimized. It basically covers the basic annotation methods of images such as boxes, points, polygons, semantic segmentation, etc., and includes interpolation annotation. It integrates a semi-automatic annotation model based on tensorflow, can pre-process part of the data, and has a good open source environment.

[0068] To use the CVAT annotation tool to annotate SAR images and establish a landslide identification knowledge base, you can follow the steps below:

[0069] Step 1: Create a landslide identification knowledge base, create tasks, and set properties;

[0070] Step 2: Landslide library marking;

[0071] Step 3: Add detailed radar data information and landslide information and other attribute information.

[0072] Step 4: Export the landslide identification knowledge base.

[0073] All long-time series SAR images are annotated in CVAT software, and the tasks are exported and provided in the form of compressed packages. If you need to modify and view them, you can import them into CVAT software. The exported CVAT for images 1.1 compressed package includes two files, images and annotations. Images contains annotated images, and annotations is an XML annotation document of the image. It can be opened and viewed with Notepad, which mainly records the annotated task name, the time when the task was created and changed, and attribute information.

[0074] The method for establishing a landslide identification knowledge base provided by the present invention takes into account the effects of slope direction and slope on landslide identification, improves the accuracy of the acquired landslide attribute information, and further improves the data accuracy of the landslide identification knowledge base, thereby laying a good foundation for subsequent AI automatic identification.

[0075] On the basis of the above embodiments, as an optional embodiment, the slope map and aspect map of the target area are combined with the hotspot clustering technology and the spatial clustering analysis technology to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area, specifically:

[0076] Hot spot clustering technology is used to analyze the deformation points in the time series deformation map of the target area to obtain the landslide potential points in the target area;

[0077] Hot spot analysis of deformation points is to initially identify the landslide boundary. The landslide potential points mentioned here are the pixels contained in the initially identified landslide boundary.

[0078] Based on the P value, Z score and deformation rate corresponding to each landslide risk point in the target area, the landslide risk points in the target area are preliminarily screened;

[0079] The P value and Z score corresponding to the landslide potential point in the target area are determined in the process of hot spot analysis of the deformation points in the time series deformation map of the target area using hot spot clustering technology;

[0080] The deformation rate corresponding to the landslide risk point in the target area is determined in the process of analyzing the SAR image data set of each target area using InSAR technology;

[0081] The preliminary screening of landslide potential points by the present invention is actually equivalent to further correcting the landslide boundary. The purpose of the preliminary screening of landslide potential points is to reduce the influence of error values ​​and data redundancy as much as possible to improve the accuracy of landslide boundary recognition.

[0082] Based on the slope map and aspect map of the target area, the spatial clustering analysis technology is used to re-screen the landslide hazard points corresponding to the target area after the preliminary screening, and the landslide boundary of the target area is determined based on the landslide hazard points corresponding to the target area after the re-screening.

[0083] The present invention fully considers the images of landslide recognition when the landslide body is at different slopes and slope directions, and further corrects the previous landslide boundary recognition results based on this, thereby improving the recognition accuracy of the landslide boundary.

[0084] On the basis of the above embodiments, as an optional embodiment, the landslide potential points in the target area are preliminarily screened based on the P value, Z score and deformation rate corresponding to each landslide potential point in the target area, specifically:

[0085] Landslide potential points that do not meet any of the following conditions: the absolute value of the annual deformation rate is greater than 2 mm, the P value is less than 0.01, and the absolute value of the Z score is greater than 2.58 are eliminated.

[0086] This embodiment provides a feasible screening and value selection method based on manual experience, with the aim of reducing recognition error rate and data redundancy.

[0087] On the basis of the above embodiments, as an optional embodiment, the slope map and the aspect map of the target area are based on the spatial clustering analysis technology to re-screen the landslide hazard points corresponding to the target area after the preliminary screening, specifically:

[0088] Based on the slope map and aspect map of the target area, weights are assigned to the landslide potential points corresponding to the target area after preliminary screening;

[0089] The Z scores of the landslide hazard points corresponding to the target area are weighted by using the weights of the landslide hazard points corresponding to the target area;

[0090] The spatial cluster analysis technique is used to analyze the Z scores of the landslide hazard points corresponding to the weighted target area, and the landslide hazard points that do not constitute landslides in the analysis results are eliminated.

[0091] It can be understood that spatial cluster analysis can identify spatial clusters with deformation information, and landslide hazard points that do not constitute landslides, that is, points that are obviously not in spatial clusters.

[0092] The present invention gives different weights to each pixel on the landslide body according to different terrain elements, and performs spatial clustering analysis in combination with the z-score (probability of forming a landslide body) obtained by hotspot analysis, and finally determines the final boundary range of the landslide body. In this process, outliers can be removed, and terrain elements such as slope and slope direction are used as an auxiliary factor for detecting the landslide boundary, thereby improving the accuracy of landslide boundary identification.

[0093] Based on the above embodiments, as an optional embodiment, the weighting of the landslide hazard points corresponding to the target area after preliminary screening based on the slope map and the aspect map of the target area needs to meet the following conditions at the same time:

[0094] When the slope of the landslide potential point is less than 15°, the slope is greater than 65°, or the slope direction of the landslide potential point is inconsistent with the slope direction of the overall landslide trend, the weight of the landslide potential point is assigned to 0;

[0095] In other cases, the weight assigned to the landslide potential point is proportional to the size of the slope of the landslide potential point.

[0096] This embodiment provides a feasible weight assignment method, and experiments have verified that this method of weight assignment has a significant effect on improving landslide boundary identification.

[0097] Based on the above embodiments, as an optional embodiment, the detection band includes: C band and L band;

[0098] Based on the above embodiments, as an optional embodiment, the detection band includes: C band and L band;

[0099] The present invention acquires image data through Sentinel-1, RADARSAT-2 and ALOS-2, wherein Sentinel-1 works in the C band, RADARSAT-2 works in the C band, and ALOS-2 works in the L band.

[0100] The landslide attribute information includes: landslide detailed information and radar data information;

[0101] The detailed information of the landslide includes: the name of the landslide, the type of landslide, the location of the landslide, the longitude and latitude of the center of the landslide, the sliding direction of the landslide, the impact range of the landslide, the maximum sliding speed, the displacement characteristics of the landslide and the landslide boundary;

[0102] The radar data information includes: data source, band information, resolution, orbit direction, incident angle and polarization mode of the SAR image;

[0103] In order to have a more detailed understanding of the landslide attribute information, this embodiment takes XXX landslide as an example for explanation;

[0104] Among them, Table 2 is the landslide attribute information table of XXX landslide; Figure 3 The landslide boundary diagram of XXX landslide is shown as an example.

[0105] Table 2

[0106]

[0107] The landslide attribute information of the present invention defines the landslide from various angles, and constructs a landslide identification knowledge base using SAR images marked with landslide attribute information, which is beneficial to improving the data richness of the landslide identification knowledge base and further improving the accuracy of landslide identification.

[0108] The optical image data set includes: a topographic map of the target area, a regional outline map, a geological disaster point distribution map, a hydrological factor map, a geological map and high-resolution optical remote sensing data;

[0109] The pre-set interpretation marks include: direct interpretation marks and indirect interpretation marks; wherein the direct interpretation marks include: vegetation characteristics, hydrological characteristics and topography; the indirect interpretation marks include: shape characteristics, color tone characteristics and texture characteristics.

[0110] In this embodiment, the direct interpretation of the signs mainly focuses on some characteristic information of the landslide itself in the remote sensing image, such as plane shape, color tone, texture and other characteristics.

[0111] Among them, the shape characteristics are: due to the downward movement of the landslide body, the terrain in the left, right and rear directions of the landslide body is slightly higher. The overall shape looks like a horseshoe, round chair, ox horn and tongue, and the opening of the rear wall is facing the bottom of the slope.

[0112] Color characteristics: New landslides are grayish white or bluish white in light tones due to the destruction of surface vegetation and the fragmentation of the ground surface, and the color distribution is uneven. The landslide wall has a light color due to its strong spectral reflectivity, while the landslide depression has a dark color due to the accumulation of water. For old landslides that occurred earlier, these color characteristics are not obvious due to the recovery of surface vegetation, but they can still be distinguished from the surrounding colors.

[0113] Texture characteristics: Due to the destruction of the integrity of the original stratum, the vegetation has been overturned and the soil is exposed, the surface is broken, the texture is relatively rough and patchy shadow patterns often appear.

[0114] Indirect interpretation signs: The main focus is on the environmental factors around the landslide, such as vegetation distribution, topography, geological structure, hydrological information and ecological landscape.

[0115] Among them, vegetation characteristics: For slow-occurring landslides or old landslides, due to the continuous downward movement of the landslide body and the upward growth of trees, more "saber trees" and "drunkard forests" appear on the landslide body, which is particularly obvious in high-resolution aerial photography.

[0116] Hydrological characteristics: The chaotic distribution pattern of the water system on the landslide body, the sudden change of river flow at the bottom of the slope, the sudden narrowing of the river channel, etc. can all be used to infer the existence of a landslide.

[0117] Topography: The topography has poor continuity, and there is often a unique "steep slope + gentle slope" slope shape. The bottom of the slope will also be uneven due to the squeezing of the landslide body. In addition, remote sensing images of different phases can be used for comparative description to predict the development trend of the landslide.

[0118] The use of interpretation marks for interpretation can improve the accuracy of the landslide attribute information obtained.

[0119] Based on the above embodiments, as an optional embodiment, the marking tool includes: a CVAT marking tool.

[0120] Compared with other annotation software, CVAT annotation tool is more suitable for multi-person collaboration, can be used for almost all computer vision (CV) related annotation tasks, and has been updated and optimized; it basically covers the basic image annotation methods such as boxes, points, polygons, semantic segmentation, etc., and includes interpolation annotation; it integrates a semi-automatic annotation model based on tensorflow, can pre-process part of the data, and has a good open source environment; using CVAT marking tool can improve the accuracy of data marking.

[0121] In a second aspect, the device for establishing a landslide identification knowledge base provided by the present invention is described. The device for establishing a landslide identification knowledge base described below and the method for establishing a landslide identification knowledge base described above can refer to each other. Figure 4 The structural diagram of a device for establishing a landslide identification knowledge base is shown as an example. Figure 4 As shown, the device includes: a SAR image data set analysis module 21, a landslide boundary determination module 22, a remote sensing visual interpretation module 23 and a marking and generation module 24;

[0122] Among them, the SAR image data set analysis module 21 is used to analyze the SAR image data set of each target area using the InSAR technology to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set; the landslide boundary determination module 22 is used to analyze the deformation points in the time series deformation map of the target area based on the slope map and slope aspect map of the target area, and determine the landslide boundary of the target area by combining the hot spot clustering technology and the spatial clustering analysis technology; the remote sensing visual interpretation module 23 is used to perform remote sensing visual interpretation on the optical image data set, time series deformation map, cumulative deformation map and landslide boundary of the target area based on pre-set interpretation marks, and determine the landslide attribute information of the target area; the marking and generation module 24 is used to mark the landslide attribute information of the target area on the SAR image of the target area using a marking tool, and generate a landslide identification knowledge base based on the marked SAR image of the target area.

[0123] The landslide identification knowledge base establishment device provided in the embodiment of the present invention specifically executes the above-mentioned landslide identification knowledge base establishment method embodiment process. For details, please refer to the contents of the above-mentioned landslide identification knowledge base establishment embodiments, which will not be repeated here. The landslide identification knowledge base establishment device provided in the embodiment of the present invention takes into account the image of slope direction and slope on landslide identification, improves the accuracy of the acquired landslide attribute information, and further improves the data accuracy of the landslide identification knowledge base, laying a good foundation for subsequent AI automatic identification.

[0124] Thirdly, Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 530 to execute the method for establishing a landslide identification knowledge base, which includes: using the InSAR technology to analyze the SAR image data set of each target area to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set; based on the slope map and the slope aspect map of the target area, the hot spot clustering technology and the spatial clustering analysis technology are combined to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area; based on the pre-set interpretation mark, the optical image data set, the time series deformation map, the cumulative deformation map and the landslide boundary of the target area are remotely sensed visually interpreted to determine the landslide attribute information of the target area; using a marking tool to mark the landslide attribute information of the target area on the SAR image of the target area, and generating a landslide identification knowledge base based on the marked SAR image of the target area.

[0125] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0126] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when executed by a processor to execute the landslide identification knowledge base establishment method provided by the above-mentioned embodiments, the method comprising: using InSAR technology to analyze the SAR image data set of each target area to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set; based on the slope map and slope aspect map of the target area, the hot spot clustering technology and the spatial clustering analysis technology are combined to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area; based on a pre-set interpretation mark, the optical image data set, the time series deformation map, the cumulative deformation map and the landslide boundary of the target area are remotely sensed visually interpreted to determine the landslide attribute information of the target area; using a marking tool to mark the landslide attribute information of the target area on the SAR image of the target area, and generating a landslide identification knowledge base based on the marked SAR image of the target area.

[0127] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0128] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for establishing a landslide identification knowledge base, characterized in that: include: The SAR image data set of each target area is analyzed by using the InSAR technology to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; and there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set; Based on the slope map and aspect map of the target area, the hotspot clustering technology and the spatial clustering analysis technology are jointly applied to analyze the deformation points in the time series deformation map of the target area to determine the landslide boundary of the target area, specifically: the hotspot clustering technology is used to perform hotspot analysis on the deformation points in the time series deformation map of the target area to obtain the landslide risk points in the target area; based on the P value, Z score and deformation rate corresponding to the landslide risk points in the target area, the landslide risk points in the target area are preliminarily screened; wherein, the P value and Z score corresponding to the landslide risk points in the target area are determined in the process of hotspot analysis of the deformation points in the time series deformation map of the target area using the hotspot clustering technology; the deformation rate corresponding to the landslide risk points in the target area is determined in the process of analyzing the SAR image data set of each target area using the InSAR technology; based on the slope map and aspect map of the target area, the spatial clustering analysis technology is used to re-screen the landslide risk points corresponding to the target area after the preliminary screening, and the landslide boundary of the target area is determined based on the landslide risk points corresponding to the target area after the re-screening; Based on the pre-set interpretation marks, remote sensing visual interpretation is performed on the optical image data set, time series deformation map, cumulative deformation map and landslide boundary of the target area to determine the landslide attribute information of the target area; The landslide attribute information of the target area is marked on the SAR image of the target area using a marking tool, and a landslide identification knowledge base is generated based on the marked SAR image of the target area.

2. The method for establishing a landslide identification knowledge base according to claim 1, characterized in that: The landslide potential points in the target area are initially screened based on the P values, Z scores and deformation rates corresponding to the landslide potential points in the target area, specifically: The landslide potential points that do not meet any of the following conditions: the absolute value of the annual deformation rate is greater than 2 mm, the P value is less than 0.01, and the absolute value of the Z score is greater than 2.58 are eliminated.

3. The method for establishing a landslide identification knowledge base according to claim 1, characterized in that: Based on the slope map and aspect map of the target area, the spatial cluster analysis technology is used to re-screen the landslide hazard points corresponding to the target area after the preliminary screening, specifically: Based on the slope map and aspect map of the target area, weights are assigned to the landslide potential points corresponding to the target area after preliminary screening; The Z scores of the landslide hazard points corresponding to the target area are weighted by using the weights of the landslide hazard points corresponding to the target area; The spatial cluster analysis technique is used to analyze the Z scores of the landslide potential points corresponding to the weighted target area, and the landslide potential points that do not constitute landslides in the analysis results are eliminated.

4. The method for establishing a landslide identification knowledge base according to claim 3, characterized in that: The weight assignment of the landslide potential points corresponding to the target area after preliminary screening based on the slope map and aspect map of the target area needs to meet the following conditions at the same time: When the slope of the landslide potential point is less than 15°, the slope is greater than 65°, or the slope direction of the landslide potential point is inconsistent with the slope direction of the overall landslide trend, the weight of the landslide potential point is assigned to 0; In other cases, the weight assigned to the landslide potential point is proportional to the size of the slope of the landslide potential point.

5. The method for establishing a landslide identification knowledge base according to claim 1, characterized in that: The detection bands include: C band and L band; The landslide attribute information includes: landslide detailed information and radar data information; wherein the landslide detailed information includes: landslide name, landslide type, landslide location, longitude and latitude of the landslide center, landslide sliding direction, landslide impact range, maximum sliding speed, landslide displacement characteristics and landslide boundary; the radar data information includes: data source, band information, resolution, orbit direction, incident angle and polarization mode of SAR image; The optical image data set includes: a topographic map of the target area, a regional outline map, a geological disaster point distribution map, a hydrological factor map, a geological map and high-resolution optical remote sensing data; The pre-set interpretation marks include: direct interpretation marks and indirect interpretation marks; wherein the direct interpretation marks include: vegetation characteristics, hydrological characteristics and topography; the indirect interpretation marks include: shape characteristics, color tone characteristics and texture characteristics.

6. The method for establishing a landslide identification knowledge base according to claim 1, characterized in that: The marking tool includes: a CVAT marking tool.

7. A device for establishing a landslide identification knowledge base, characterized in that: include: A SAR image data set analysis module is used to analyze the SAR image data set of each target area using the InSAR technology to determine the time series deformation map and the cumulative deformation map of each target area; wherein the coherent measurement points analyzed by the InSAR technology include permanent scatterer points and distributed scatterer points; and there are differences in detection phase, detection band and / or detection viewing angle between different SAR images in the data set; The landslide boundary determination module is used to analyze the deformation points in the time series deformation map of the target area based on the slope map and the aspect map of the target area, and determine the landslide boundary of the target area by combining the hot spot clustering technology and the spatial clustering analysis technology. Specifically, the hot spot clustering technology is used to perform hot spot analysis on the deformation points in the time series deformation map of the target area to obtain the landslide potential points in the target area; based on the P value, Z score and deformation rate corresponding to the landslide potential points in the target area, the landslide potential points in the target area are preliminarily screened; wherein, the P value and Z score corresponding to the landslide potential points in the target area are determined in the process of hot spot analysis of the deformation points in the time series deformation map of the target area using the hot spot clustering technology; the deformation rate corresponding to the landslide potential points in the target area is determined in the process of analyzing the SAR image data set of each target area using the InSAR technology; based on the slope map and the aspect map of the target area, the spatial clustering analysis technology is used to re-screen the landslide potential points corresponding to the target area after the preliminary screening, and the landslide boundary of the target area is determined based on the landslide potential points corresponding to the target area after the re-screening; The remote sensing visual interpretation module is used to perform remote sensing visual interpretation of the optical image data set, time series deformation map, cumulative deformation map and landslide boundary of the target area based on the pre-set interpretation marks to determine the landslide attribute information of the target area; The marking and generating module is used to mark the landslide attribute information of the target area on the SAR image of the target area by using a marking tool, and to generate a landslide identification knowledge base based on the marked SAR image of the target area.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for establishing a landslide identification knowledge base as described in any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for establishing a landslide identification knowledge base as claimed in any one of claims 1 to 6 are implemented.

Images