A method, device, medium and product for dynamic identification of landslide hazards

Abstract

The present application discloses a method, equipment, medium and product for dynamic identification of landslide hazards, which relates to the field of geological disaster identification. The method comprises: obtaining the extraction result of InSAR deformation anomaly area based on sentinel data of the study area; inputting the extraction result of InSAR deformation anomaly area into the constructed landslide image identifier to obtain the identification result of the suspected landslide hazard target area in the study area; using the results of InSAR deformation anomaly area and suspected landslide hazard target area, conducting landslide hazard identification research integrating InSAR surface deformation characteristics and optical remote sensing geomorphological characteristics, and obtaining the dynamic identification result of landslide hazard. The present application solves the problems of the existing easy to pay a large workload in feature selection and hyperparameter tuning, the limited receptive field of CNN and its weak global modeling ability, and realizes the accurate identification of landslide hazards, while improving the extraction efficiency of InSAR deformation anomaly area and the identification efficiency of landslide hazards.

Description

Technical Field

[0001] The present application relates to the field of geological disaster identification, and in particular to a method, equipment, medium and product for dynamic identification of landslide hazards. Background Art

[0002] Landslide refers to the phenomenon that the materials that make up the slope (rock, soil, artificial fill, etc.) move downward or outward along the slope under the influence of gravity and other external conditions. In recent years, the frequency and diversity of landslide disasters have shown an upward trend due to the intensification of global warming, the increase in the intensity of human activities and the sensitive response to climate change. Therefore, geological disaster prevention and control work has always been a long-term major challenge that requires continuous research and efforts.

[0003] Identification of hidden dangers of landslide disasters is an important prerequisite for prevention, control and early warning. However, about 80% of the catastrophic geological disasters that have occurred in recent years are not in the discovered hidden danger database. The main reason is that the source areas of catastrophic geological disasters are often located at high altitudes and are highly concealed. Traditional artificial ground survey methods are difficult to reach the site for effective identification. Usually, the source area of ​​landslides will undergo spatiotemporal deformation evolution before the disaster occurs. Deformation detection and investigation of the study area is conducive to the development of landslide prevention and control work. Based on this, it is extremely important to quickly and accurately identify landslide hazards.

[0004] In the early stage of landslide hazard identification, it mainly depends on the expert visual interpretation of high-resolution remote sensing images. With the continuous advancement of computer technology and image processing technology, the identification method of using artificially designed thresholds to extract landslide information has been developed. In recent years, the rapid development of earth observation technology and artificial intelligence technology has promoted the increasingly widespread application of remote sensing technology in large-scale landslide disaster investigation. Therefore, the identification of landslide hazards based on multi-source and multi-temporal remote sensing images based on intelligent algorithms has become an important technical means for landslide disaster investigation and prevention. Scholars around the world have achieved fruitful results in the field of landslide hazard identification using advanced technical means such as high-resolution remote sensing image analysis, synthetic aperture radar interferometry, and deep learning.

[0005] In summary, the existing methods for identifying landslide hazards can be mainly divided into optical remote sensing visual interpretation method, feature threshold division method, object-oriented information extraction method and machine learning method. Among them, the optical remote sensing visual interpretation method is to identify landslide hazards by expert visual recognition and spatial analysis based on the established interpretation. The visual interpretation method using optical remote sensing images is generally more accurate, but the disadvantages are high workload and low efficiency. It is necessary to manually find landslide bodies from massive remote sensing images, and the interpretation results are also affected by the knowledge of experts in the field of geology and remote sensing. There are problems such as large workload and easy confusion between landslides and other landforms. The feature threshold division method is to set one or more thresholds to achieve automatic identification of landslides by statistically analyzing the different features of the landslide area. The feature threshold division method can improve the efficiency of landslide interpretation to a certain extent. The difficulty lies in how to establish effective and accurate landslide feature extraction rules. Therefore, it has a small scope of application and weak universality. Object-oriented information extraction is a method based on image attribute features, which uses image segmentation algorithms to group pixels with similar features into consistent objects. It places high demands on the selection and preprocessing of remote sensing images. Images need to have similar acquisition angles, lighting conditions, and seasonal characteristics. There is little research on large-scale landslide identification. Although machine learning methods can solve the above problems to a certain extent, it is also necessary to extract various relevant features of the data used in the data preparation stage before various machine learning classifiers can be used for landslide identification. Based on the above description, these methods all require the design of interpretation logic or the extraction of features in images, which is easy to spend a lot of work in feature selection and hyperparameter tuning, limiting the potential for algorithm improvement. In addition, most machine learning methods are semantic segmentation networks based on CNN. Although they can reduce the time spent on landslide feature selection during the research process and have been proven to have strong extraction capabilities, there is still a lack of systematic research on the limited receptive field of CNN and its weak global modeling capabilities. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present application provides a method, equipment, medium and product for dynamically identifying landslide hazards.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for dynamically identifying landslide hazards, comprising:

[0009] The InSAR deformation anomaly area extraction result is obtained based on the sentinel data of the study area; the sentinel data includes SAR data and DEM data;

[0010] Constructing a landslide recognition model, training and testing the landslide recognition model using a landslide enhanced sample data set, and using the tested landslide recognition model as a landslide image recognizer; the landslide recognition model is a neural network model combining a multi-scale feature fusion module and a Transformer segmentation module;

[0011] Inputting the InSAR deformation anomaly area extraction result into the landslide image identifier to obtain the identification result of the suspected landslide hazard target area in the study area;

[0012] By using the results of InSAR deformation anomaly areas and suspected landslide hazard target areas, a study on landslide hazard identification was carried out by integrating InSAR surface deformation characteristics and optical remote sensing geomorphological characteristics, and dynamic identification results of landslide hazards were obtained; the optical remote sensing geomorphological characteristics include optical remote sensing geomorphology and optical morphological characteristics.

[0013] Optionally, the InSAR deformation anomaly area extraction results are obtained based on the sentinel data of the study area, including:

[0014] Using SBAS-InSAR technology to perform orbit-raising mode inversion on the SAR data to obtain a first deformation inversion result; the first deformation inversion result includes a deformation rate and a cumulative deformation amount;

[0015] Using D-InSAR technology to perform descending track mode inversion on the DEM data to obtain a second deformation inversion result;

[0016] Based on the first deformation inversion result and in combination with the deformation rate threshold, extracting the deformation abnormal area;

[0017] Using the second deformation inversion result to supplement the abnormal deformation area and determine the deformation response period;

[0018] The supplemented deformation anomaly area whose deformation response period exceeds the set period is taken as the InSAR deformation anomaly area, and the InSAR deformation anomaly area extraction result is obtained.

[0019] Optionally, the SBAS-InSAR technology is used to perform orbit-raising mode inversion on the SAR data to obtain a first deformation inversion result, specifically including:

[0020] A SAR image is randomly selected from the SAR data as a main image for image registration, and a temporal and spatial baseline threshold is set to perform interference pair combination to obtain a combined image pair;

[0021] Performing differential interference processing on each combined image pair respectively to generate multiple differential interference images;

[0022] Determine the interference phase at each pixel point in the differential interference pattern to obtain a deformation phase value of the differential interference pattern;

[0023] Perform phase unwrapping on the deformation phase value of the differential interference pattern to obtain a phase matrix;

[0024] The generalized inverse matrix of the phase matrix is ​​determined by the singular value decomposition method, and the deformation rate or the cumulative displacement time series is determined based on the generalized inverse matrix to obtain the first deformation inversion result; the cumulative deformation amount is determined based on the cumulative displacement time series.

[0025] Optionally, the D-InSAR technology adopted is a two-track method.

[0026] Optionally, the InSAR deformation anomaly area extraction result is input into the landslide image identifier to obtain the identification result of the suspected landslide hazard target area in the study area, specifically including:

[0027] The InSAR deformation anomaly area extraction result is input into the multi-scale feature fusion module; in the multi-scale feature fusion module, the original features of the InSAR deformation anomaly area extraction result are extracted through the ResNet network, and the original features are pooled into features of four different scales using four different convolution kernels; the features of the four different scales are fused through the upsampling layer; the fused image features are obtained, and the fused image features are processed using a 1×1 convolution layer to obtain processed features;

[0028] The original features are input into the Transformer segmentation module of the encoder-decoder structure after the 1×1 convolution layer. At the same time, the processed features are also input into the Transformer segmentation module of the encoder-decoder structure. The multi-head attention module of the Transformer segmentation module is used to distinguish the target area, and the identification result of the suspected landslide hazard target area in the study area is obtained.

[0029] Optionally, using the results of InSAR deformation anomaly areas and suspected landslide hazard target areas, a study on landslide hazard identification that integrates InSAR surface deformation characteristics and optical remote sensing geomorphological characteristics is carried out to obtain dynamic identification results of landslide hazards, including:

[0030] Taking the disaster-bearing body as the center, a buffer zone is established in the InSAR deformation anomaly zone at a set distance;

[0031] Acquiring an optical remote sensing image of the buffer zone, and determining an optical remote sensing landform based on the optical remote sensing image, and determining whether the buffer zone has the terrain conditions for landslide development based on the optical remote sensing landform;

[0032] Determine whether there is a suspected landslide hazard in the buffer zone based on the suspected landslide hazard target area result;

[0033] When there is a suspected landslide hazard in the buffer zone and the buffer zone has the terrain conditions for landslide development, it is determined that there is a landslide hazard in the buffer zone;

[0034] When there is a suspected landslide hazard in the buffer zone and the buffer zone does not have the terrain conditions for landslide development, determining that there is no landslide hazard in the buffer zone;

[0035] When there is no suspected landslide hazard in the buffer zone and the buffer zone does not have the terrain conditions for landslide development, it is determined that there is no landslide hazard in the buffer zone;

[0036] When there is no suspected landslide hazard in the buffer zone and the buffer zone has the terrain conditions for landslide development, it is determined whether there are optical deformation features that meet the set conditions in the optical remote sensing image; when there are optical deformation features that meet the set conditions in the optical remote sensing image, it is determined that there is a landslide hazard in the buffer zone; when there are no optical deformation features that meet the set conditions in the optical remote sensing image, it is determined that there is no landslide hazard in the buffer zone.

[0037] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned methods for dynamically identifying landslide hazards.

[0038] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-described methods for dynamically identifying landslide hazards.

[0039] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for dynamically identifying landslide hazards.

[0040] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0041] The present application provides a method, device, equipment, medium and product for dynamic identification of landslide hazards. By obtaining the InSAR deformation anomaly area extraction result based on the sentinel data of the study area, the problem of the existing high workload in feature selection and hyperparameter tuning is solved, and the accurate extraction of the InSAR deformation anomaly area is achieved while improving the feature extraction efficiency; by constructing a landslide recognition model, the landslide recognition model is trained and tested using a landslide enhanced sample data set, and the tested landslide recognition model is used as a landslide image recognizer, and the InSAR deformation anomaly area extraction result is input into the landslide image recognizer, the recognition result of the suspected landslide hazard target area in the study area is obtained, and the problem of limited receptive field and weak global modeling ability of CNN is solved, and the accurate and rapid identification of landslide hazards is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 This is an application environment diagram of a method for dynamically identifying landslide hazards in one embodiment of the present application;

[0044] Figure 2 An implementation architecture diagram of a method for dynamically identifying landslide hazards provided in one embodiment of the present application;

[0045] Figure 3 SBAS-InSAR technology processing flow chart provided in one embodiment of the present application;

[0046] Figure 4 A D-InSAR technology processing flow chart provided for another embodiment of the present application;

[0047] Figure 5 A flowchart of InSAR deformation anomaly area extraction provided in an embodiment of the present application;

[0048] Figure 6 A schematic diagram of the landslide identification model structure provided in one embodiment of the present application;

[0049] Figure 7 A schematic diagram of landslide optical remote sensing interpretation marks provided in another embodiment of the present application.

[0050] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0053] The method for dynamically identifying landslide hazards provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the sentinel data of the study area to the server 104. After the server 104 receives the sentinel data of the study area, for the sentinel data of the study area, the server 104 obtains the InSAR deformation anomaly area extraction result based on the sentinel data of the study area; and constructs a landslide recognition model, uses the landslide enhanced sample data set to train and test the landslide recognition model, and uses the tested landslide recognition model as a landslide image recognizer; the InSAR deformation anomaly area extraction result is input into the landslide image recognizer to obtain the recognition result of the suspected landslide hazard target area in the study area; using the InSAR deformation anomaly area and the suspected landslide hazard target area results, a landslide hazard identification study integrating the InSAR surface deformation characteristics and the optical remote sensing geomorphological characteristics is carried out to obtain the dynamic identification result of the landslide hazard. The server 104 may feed back the obtained landslide hazard dynamic identification result to the terminal 102. In addition, in some embodiments, the landslide hazard dynamic identification method may also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 may directly perform dynamic identification of landslide hazards for the sentinel data of the study area, or the server 104 may obtain the sentinel data of the study area from the data storage system and perform dynamic identification of landslide hazards for the sentinel data of the study area.

[0054] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0055] In an exemplary embodiment, Figure 2 As shown, a method for dynamically identifying landslide hazards is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps 201 to 204. Among them:

[0056] Step 201, obtaining the InSAR deformation anomaly area extraction result based on the sentinel data of the study area. The sentinel data includes SAR data and DEM data.

[0057] Step 202, construct a landslide recognition model (Multi-scale Feature Aggregation Transformer Segmentation-Landslide Recognition, MFATrans-LR), use the landslide enhanced sample data set to train and test the landslide recognition model, and use the tested landslide recognition model as a landslide image recognizer. The landslide recognition model is a neural network model that combines a multi-scale feature fusion module and a Transformer segmentation module.

[0058] Step 203, input the InSAR deformation anomaly area extraction result into the landslide image recognizer to obtain the recognition result of the suspected landslide hazard target area in the study area.

[0059] Step 204, using the results of the InSAR deformation anomaly area and the suspected landslide hazard target area, a landslide hazard identification study integrating the InSAR surface deformation characteristics and the optical remote sensing geomorphic morphological characteristics is carried out to obtain a dynamic identification result of the landslide hazard. The optical remote sensing geomorphic morphological characteristics include optical remote sensing geomorphology and optical morphological characteristics.

[0060] By implementing the above steps 201 to 204, the scope of landslide monitoring is expanded through the complementarity of the ascending and descending orbit data of the Sentinel-1 satellite and the superposition of shadow areas. The combined ascending and descending orbit InSAR technology ensures the comprehensiveness and reliability of the deformation inversion results in the study area, providing support for the extraction of abnormal deformation areas. Secondly, in view of the structural differences between remote sensing image data and the shortcomings of the existing standard deep learning model framework in describing remote sensing image features, the landslide identification MFATrans-LR model combining multi-scale feature fusion and Transformer segmentation is designed to realize the automatic identification of suspected landslide hazard target areas in the study area. Finally, a study on the confirmation and identification of landslide hazards combining optical remote sensing with suspected landslide hazard data was carried out.

[0061] In addition, the effectiveness of landslide disaster prevention depends to a large extent on the accurate identification of landslide hazards at an early stage. In view of the problems of cumbersome work, much manual intervention, high missed detection and false detection rate, and easy confusion between landslides and ground objects in the traditional method of identifying landslide hazards based only on computer image detection technology, this application first focuses on the advantages of remote sensing space, and focuses on strengthening the network's modeling ability of global contextual semantic information in the design of landslide identification models. Through the "three-level" progressive approach of deformation anomaly area detection and extraction - suspected landslide hazard target area identification - landslide hazard confirmation and identification, a three-level landslide hazard identification method based on image-based InSAR confirmation is established to find the area where landslide hazards exist (identification of location and size). In terms of model construction, its core task is to ensure that the network is more focused and deeply learns the essential characteristics of landslides. It is necessary to propose an efficient and accurate landslide hazard identification method. Based on this, this application essentially proposes a method such as Figure 2 The three-level driving framework of the landslide hazard identification system is shown in the figure. First, in the first level, the deformation anomaly detection index is used to extract the InSAR deformation anomaly area in the study area in combination with the ascending and descending InSAR deformation inversion results. Secondly, in the second level, a landslide recognition model (Multi-scale Feature Aggregation Transformer Segmentation-Landslide Recognition, MFATrans-LR) combining multi-scale feature fusion and Transformer segmentation is developed. The model is trained and verified using the landslide enhanced sample data set. The MFATrans-LR model is adjusted by using the transfer learning method by expanding the landslide image annotation samples in the study area to achieve automatic identification of suspected landslide hazard target areas in the study area. Finally, in the third level, the results of InSAR deformation anomaly areas and suspected landslide hazard target areas are used to carry out landslide hazard identification research by integrating InSAR surface deformation characteristics and optical remote sensing geomorphological features. This method aims to combine the advantages of the two remote sensing technologies to effectively improve the efficiency of landslide hazard identification. This hierarchical landslide hazard identification method not only helps to reduce manual workload and efficiently utilize data, but also reduces the missed detection rate, thereby improving the automation level of landslide hazard monitoring. It has important practical application value for the rapid investigation of large-scale landslide hazards.

[0062] In another exemplary embodiment of the present application, when using synthetic aperture radar interferometry (InSAR) related technology to identify and monitor landslides, it is necessary to take into account the characteristics of SAR satellites, which is an active microwave side-view imaging technology. In areas with large terrain undulations, shadows, overlapping and perspective contraction often occur. In addition, dense vegetation and cloudy and rainy climate conditions may also have a negative impact on the quality of SAR data. In order to overcome these problems, methods such as multi-view observation, increasing the number of observations and improving time resolution can be used, which are effective strategies for InSAR deformation inversion in low-coherence areas. Therefore, the present application uses SBAS-InSAR and D-InSAR combined technology to process Sentinel-1A data. Taking advantage of the lifting and track data, especially in the shielded area, the coverage of landslide monitoring is increased, and the accuracy of InSAR deformation inversion is improved. This method can guarantee the comprehensiveness and reliability of deformation results to the greatest extent, and provides strong support for extracting abnormal deformation areas. Among them, the SBAS-InSAR and D-InSAR combined technology is used to process the Sentinel data. This processing process includes:

[0063] (1) Using SBAS-InSAR technology, the SAR data are inverted in ascending orbit mode to obtain the first deformation inversion result. The first deformation inversion result includes deformation rate and cumulative deformation.

[0064] Among them, the Small Baseline Subsets InSAR (SBAS-InSAR) technology was proposed by Italian scholar Berardino and others. It is a time series InSAR analysis method based on multiple main images and short space-time baseline criteria. It mainly generates interference combinations through multiple main images and short space-time baseline criteria to form as many interference pair sets as possible. The deformation time series of each small set is solved by the least squares method, and multiple small sets are combined and solved by the singular value decomposition method, so as to obtain the time series surface deformation information during the entire observation time. SBAS-InSAR technology has alleviated the decoherence phenomenon caused by the long space-time baseline to a certain extent, and also increased the temporal resolution of deformation monitoring. Figure 3 As shown in the figure, the main principles and processes of SBAS-InSAR technology are as follows:

[0065] First, suppose that according to a certain time series (t0, t1...t N) Obtain N+1 SAR images covering the same area, arbitrarily select one SAR image as the main image for registration, set an appropriate (determined according to actual needs) time and space baseline threshold to perform interference pair combination, and then perform differential interference processing on each combined image pair to generate M differential interference images. At this time, M satisfies:

[0066]

[0067] Assume that the jth untangled interference pattern is formed by t A and t B Time (t B >t A ) is obtained by differential interference of two SAR images obtained by , then the interference phase at a certain pixel point in this differential interference image can be expressed as:

[0068]

[0069] In the formula, is the interference phase at a certain pixel point. and They are the pixels at t A and t B The time corresponds to t o The phase of the moment. t A ~t B The deformation phase of the line of sight between moments. is the terrain phase error. Atmospheric phase error. is the phase error caused by noise. d(t B ) and d(t A ) represent t A and t B The cumulative deformation in the line of sight relative to the reference time t0 (d(t0) = 0). λ is the radar wavelength. Δh is the elevation error. θ is the incident angle. R is the slant distance from the radar to the ground observation object. B ⊥ is the vertical baseline. t A and t B The atmospheric delay phase corresponding to the time.

[0070] Then, the deformation phase value of the jth interference pattern can be written as:

[0071]

[0072] All the unwrapped differential interference phases are combined into a matrix form, expressed as: Where Bv is an M×N matrix.

[0073] The singular value decomposition method can be used to obtain the generalized inverse matrix of the matrix, calculate the final deformation rate or cumulative displacement time series, and then obtain the cumulative deformation.

[0074] (2) The D-InSAR technology is used to perform down-orbit mode inversion on the DEM data to obtain the second deformation inversion result.

[0075] Among them, Differential InSAR (DInSAR) technology is an extension of the traditional InSAR technology. This technology uses multi-period SAR single-view complex images or combined DEM to measure small surface deformations. The DInSAR technology used in this application is a two-track method, such as Figure 4 As shown in the figure, the main processing steps include image clipping, image registration and resampling, interference and filtering, phase difference, phase unwrapping and geocoding, and finally the surface deformation data of the study area are obtained.

[0076] In the above steps (1) and (2), the advantages of lift and track data are utilized, especially in the shielded area, to increase the coverage of landslide monitoring and improve the accuracy of InSAR deformation inversion. This method can guarantee the comprehensiveness and reliability of deformation results to the greatest extent, and provide strong support for extracting abnormal deformation areas. InSAR deformation is an important indicator to characterize the magnitude of surface deformation in the SAR radar line of sight (LOS) direction. The two quantities of "deformation rate and cumulative deformation" within the slope range are used as responses to regional deformation conditions.

[0077] (3) Based on the first deformation inversion result and combined with the deformation rate threshold, the deformation anomaly area is extracted.

[0078] (4) The second deformation inversion result is used to supplement the deformation anomaly area and determine the deformation response period.

[0079] (5) The supplemented deformation anomaly area whose deformation response period exceeds the set period is taken as the InSAR deformation anomaly area to obtain the InSAR deformation anomaly area extraction result.

[0080] InSAR deformation is an important indicator of the magnitude of surface deformation in the SAR radar line of sight (LOS) direction. The two quantities of "deformation rate and cumulative deformation" within the slope range are used as responses to regional deformation. Based on this, the specific implementation method of extracting deformation anomaly areas by combining ascending and descending InSAR technology in the above steps (4) and (5) is as follows: Figure 5As shown, it includes: setting the SBAS-InSAR deformation rate threshold of the ascending SAR data processing to 10 mm / year, and extracting the abnormal deformation area in combination with the cumulative deformation distribution characteristics. The abnormal deformation area is supplemented by using the D-InSAR deformation inversion results of multi-period descending data processing (the deformation response of 2 periods or more is used as the basis for determining the abnormal deformation area). Finally, the InSAR abnormal deformation area extraction of the study area is realized, and then the InSAR abnormal deformation area extraction result is obtained.

[0081] In another exemplary embodiment of the present application, in order to solve the problem of limited receptive field and weak global modeling ability of existing CNN, the present application proposes an MFATrans-LR model combining a multi-scale feature fusion module and a Transformer segmentation module, which aims to enhance the network's modeling ability of global contextual semantic information, alleviate the limitations of convolution in the receptive field size problem, improve the model's generalization ability, and enable the model to identify landslides of different forms. Figure 6 As shown in the figure, the model mainly consists of two modules, namely the Multi-scale Feature Fusion Module (MSFFM) and the Transformer Segmentation Module (TSM). The implementation steps of the MFATrans-LR model are mainly divided into the following points: 1) Input a remote sensing image data, and the Multi-scale Feature Fusion Module extracts the original image features through the ResNet network. 2) Use four different convolution kernels to pool the features into four features of different scales. 3) The four different features are fused through the upsampling layer to obtain the fused image features, which can capture rich contextual information. 4) The fused features and the original features are respectively reduced by the 1×1 convolution layer to reduce the number of feature channels, and then sent to the Transformer segmentation module of the encoder-decoder structure. Finally, the Transformer's multi-head attention module is used to distinguish the target area and output the final predicted landslide area segmentation map, thereby realizing the landslide area segmentation.

[0082] Based on the description in this embodiment, the implementation process of the above step 203 can be replaced by the following steps 1) and 2), namely:

[0083] 1) The InSAR deformation anomaly area extraction results are input into the multi-scale feature fusion module. In the multi-scale feature fusion module, the original features of the InSAR deformation anomaly area extraction results are extracted through the ResNet network, and the original features are pooled into four different scale features using four different convolution kernels. The four different scale features are fused through the upsampling layer. The fused image features are obtained, and the fused image features are processed using a 1×1 convolution layer to obtain the processed features.

[0084] 2) The original features are input into the Transformer segmentation module of the encoder-decoder structure after the 1×1 convolution layer. At the same time, the processed features are also input into the Transformer segmentation module of the encoder-decoder structure. The multi-head attention module of the Transformer segmentation module is used to distinguish the target area, and the identification results of the suspected landslide hazard target area in the study area are obtained.

[0085] In another exemplary embodiment of the present application, landslide hazard identification is not an investigation of landslide disasters that have occurred in a general sense, nor is it an assessment after the disaster, but requires a comprehensive assessment of the spatiotemporal evolution characteristics of professional geological knowledge and the disaster development process, and is measured and identified in combination with remote sensing observation elements. Relevant researchers of the disaster-bearing body theory believe that there is no disaster without a disaster-bearing body, and therefore, landslide hazard is relative to landslide disasters. If there is no threat object (disaster-bearing body) and potential hazards (landslide disasters), it cannot be determined as a landslide hazard for a period of time. The abnormal deformation area extracted by InSAR technology may be inconsistent with the landslide hazard. So, in order to solve the problem that the abnormal deformation area extracted by InSAR technology may be inconsistent with the landslide hazard, in this embodiment, it is necessary to screen and determine the landslide hazard through data such as basic geographic information, historical landslide disasters and geological information.

[0086] Based on the above description, when the definition of landslide hazard identification is to identify landslide areas that may cause damage to the hazard-bearing body (houses, roads, urban infrastructure, reservoirs), the implementation process of step 204 can be: first, with the hazard-bearing body distribution as the center, buffer zones are set up in the InSAR deformation anomaly area at set distances (for example, 1 km), and the suspected landslide hazard target areas located in these buffer zones are analyzed to determine whether they have landslide hazards. Secondly, the suspected landslide hazards are identified and confirmed by the following three methods: (1) If there is a suspected landslide hazard in the buffer zone, then, and it is judged according to the optical satellite image that it has the terrain conditions for the development of landslides, but there are no obvious optical deformation characteristics, then the area can be determined to have a landslide hazard. (2) If the buffer zone is a suspected landslide hazard, and it is judged according to the optical satellite image that it has the terrain conditions for the development of landslides and has obvious optical deformation characteristics, then the buffer zone can be determined to have a landslide hazard. (3) The buffer zone is not a suspected landslide hazard, but according to the suspected landslide hazard target area and the comprehensive judgment of optical satellite images, the buffer zone has the terrain conditions for landslide development and has obvious optical deformation characteristics. Therefore, the buffer zone can be determined to have landslide hazards. Among them, the optical deformation characteristics of landslide hazards refer to the chair-shaped landslide, the local sliding characteristics with obvious color difference, as well as large-scale cracks, staggered ridges, etc. Figure 7 shown.

[0087] The present application also provides an application scenario, which applies the above-mentioned method for dynamic identification of landslide hazards. Specifically: the method for dynamic identification of landslide hazards provided in this embodiment can be applied in content distribution scenarios. The internal distribution scenario includes a content production link, a content processing link, and a content distribution link. Sentinel data enters the content processing link from the content production link, obtains corresponding content features through human-computer collaboration, and enters the downstream content distribution link. The method for dynamic identification of landslide hazards provided in this embodiment belongs to the dynamic identification of landslide hazards in the content processing link. Specifically, in the content processing link process for sentinel data, sentinel data can be labeled based on the collaborative method of dynamic identification of machine landslide hazards and dynamic identification of artificial landslide hazards, that is, corresponding landslide hazard labels can be added to the sentinel data.

[0088] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the dynamic identification results of landslide hazards. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for dynamic identification of landslide hazards is implemented.

[0089] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0090] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0091] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0092] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0094] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0095] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0096] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for dynamically identifying landslide hazards, characterized in that: The landslide hazard dynamic identification method comprises: The InSAR deformation anomaly area extraction result is obtained based on the sentinel data of the study area; the sentinel data includes SAR data and DEM data; Constructing a landslide recognition model, training and testing the landslide recognition model using a landslide enhanced sample data set, and using the tested landslide recognition model as a landslide image recognizer; the landslide recognition model is a neural network model combining a multi-scale feature fusion module and a Transformer segmentation module; Inputting the InSAR deformation anomaly area extraction result into the landslide image identifier to obtain the identification result of the suspected landslide hazard target area in the study area; Using the results of InSAR deformation anomaly areas and suspected landslide hazard target areas, a study on landslide hazard identification was carried out by integrating InSAR surface deformation characteristics and optical remote sensing geomorphological features, and the dynamic identification results of landslide hazards were obtained; the optical remote sensing geomorphological features include optical remote sensing geomorphology and optical morphological features, specifically including: Taking the disaster-bearing body as the center, a buffer zone is established in the InSAR deformation anomaly zone at a set distance; Acquiring an optical remote sensing image of the buffer zone, and determining an optical remote sensing landform based on the optical remote sensing image, and determining whether the buffer zone has the terrain conditions for landslide development based on the optical remote sensing landform; Determine whether there is a suspected landslide hazard in the buffer zone based on the suspected landslide hazard target area result; When there is a suspected landslide hazard in the buffer zone and the buffer zone has the terrain conditions for landslide development, it is determined that there is a landslide hazard in the buffer zone; When there is a suspected landslide hazard in the buffer zone and the buffer zone does not have the terrain conditions for landslide development, determining that there is no landslide hazard in the buffer zone; When there is no suspected landslide hazard in the buffer zone and the buffer zone does not have the terrain conditions for landslide development, it is determined that there is no landslide hazard in the buffer zone; When there is no suspected landslide hazard in the buffer zone and the buffer zone has the terrain conditions for landslide development, it is determined whether there are optical deformation features that meet the set conditions in the optical remote sensing image; when there are optical deformation features that meet the set conditions in the optical remote sensing image, it is determined that there is a landslide hazard in the buffer zone; when there are no optical deformation features that meet the set conditions in the optical remote sensing image, it is determined that there is no landslide hazard in the buffer zone.

2. The method for dynamic identification of landslide hazards according to claim 1, characterized in that: Based on the sentinel data of the study area, the InSAR deformation anomaly area extraction results are obtained, including: Using SBAS-InSAR technology to perform orbit-raising mode inversion on the SAR data to obtain a first deformation inversion result; the first deformation inversion result includes a deformation rate and a cumulative deformation amount; Using D-InSAR technology to perform descending track mode inversion on the DEM data to obtain a second deformation inversion result; Based on the first deformation inversion result and in combination with the deformation rate threshold, extracting the deformation abnormal area; Using the second deformation inversion result to supplement the abnormal deformation area and determine the deformation response period; The supplemented deformation anomaly area whose deformation response period exceeds the set period is taken as the InSAR deformation anomaly area, and the InSAR deformation anomaly area extraction result is obtained.

3. The method for dynamic identification of landslide hazards according to claim 2, characterized in that: The SAR data is inverted in an ascending orbit mode using the SBAS-InSAR technology to obtain a first deformation inversion result, which specifically includes: Selecting any one SAR image from the SAR data as a main image for image registration, and setting a temporal and spatial baseline threshold to perform interference pair combination to obtain a combined image pair; Performing differential interference processing on each combined image pair respectively to generate multiple differential interference images; Determine the interference phase at each pixel point in the differential interference pattern to obtain a deformation phase value of the differential interference pattern; Perform phase unwrapping on the deformation phase value of the differential interference pattern to obtain a phase matrix; The generalized inverse matrix of the phase matrix is ​​determined by the singular value decomposition method, and the deformation rate or the cumulative displacement time series is determined based on the generalized inverse matrix to obtain the first deformation inversion result; the cumulative deformation amount is determined based on the cumulative displacement time series.

4. The method for dynamic identification of landslide hazards according to claim 2 is characterized in that: The D-InSAR technology used is a two-track method.

5. The method for dynamic identification of landslide hazards according to claim 1 is characterized in that: The InSAR deformation anomaly area extraction result is input into the landslide image identifier to obtain the identification result of the suspected landslide hazard target area in the study area, which specifically includes: The InSAR deformation anomaly area extraction result is input into the multi-scale feature fusion module; in the multi-scale feature fusion module, the original features of the InSAR deformation anomaly area extraction result are extracted through the ResNet network, and the original features are pooled into features of four different scales using four different convolution kernels; the features of the four different scales are fused through the upsampling layer; the fused image features are obtained, and the fused image features are processed using a 1×1 convolution layer to obtain processed features; The original features are input into the Transformer segmentation module of the encoder-decoder structure after the 1×1 convolution layer. At the same time, the processed features are also input into the Transformer segmentation module of the encoder-decoder structure. The multi-head attention module of the Transformer segmentation module is used to distinguish the target area, and the identification result of the suspected landslide hazard target area in the study area is obtained.

6. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for dynamically identifying landslide hazards according to any one of claims 1 to 5.

7. A 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 dynamically identifying landslide hazards described in any one of claims 1 to 5 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for dynamically identifying landslide hazards described in any one of claims 1 to 5 are implemented.

Images