Method and system for monitoring landslide in vegetation dense area based on multi-source InSAR data fusion and deep learning

Abstract

The invention provides a vegetation dense area landslide monitoring method and monitoring system based on multi-source InSAR data fusion and deep learning, and the method comprises the steps: a data obtaining step: obtaining multi-source InSAR data of a target area; a deformation interpretation step: performing deformation interpretation processing on the target area of the multi-source InSAR data by using a preset MT-InSAR method to obtain deformation rate data; a region deformation feature identification step: performing fusion analysis processing on the deformation rate data by using a preset multi-source InSAR fusion optimization monitoring strategy to obtain region deformation detection data in a first time period, and performing region boundary sketching processing on the region deformation detection data to obtain region deformation feature data; a region morphological feature recognition step: inputting the optical remote sensing image to be recognized into a landslide classifier model constructed based on deep learning to output landslide morphological recognition data; and a landslide monitoring grading step: obtaining landslide monitoring data according to the regional deformation characteristic data and the landslide form identification data.

Description

Technical Field

[0001] The present invention relates to the field of landslide monitoring technology, and in particular to a landslide monitoring method and monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning. Background Art

[0002] Landslides are a common geological hazard in high mountain valleys. Real-time monitoring of landslides is particularly challenging in densely vegetated areas characterized by frequent clouds, fog, deep valleys, and dense trees. Traditional landslide monitoring methods are limited by cloud cover in optical remote sensing and the decoherence of Synthetic Aperture Radar (InSAR) technology, resulting in poor landslide monitoring and identification. Traditional monitoring methods typically employ single InSAR data for identification. This is particularly true because single InSAR data (such as the C-band) is susceptible to signal decoherence due to its short wavelength in areas of high vegetation coverage, making it difficult to obtain continuous deformation information. Furthermore, traditional manual interpretation of the obtained identification information is inefficient and subjective, making it unable to meet the needs of rapid identification and risk grading of large-scale landslide risk points.

[0003] With the development of monitoring technology, existing technologies have incorporated multi-source InSAR data. However, these methods lack zoning monitoring strategies tailored to vegetation cover differences, resulting in insufficient monitoring accuracy. To effectively monitor landslides in densely vegetated areas, a landslide monitoring method that can account for vegetation cover differences is needed to improve monitoring accuracy. Summary of the Invention

[0004] The purpose of the present invention is to address the defects of the existing technology and provide a landslide monitoring method that can target vegetation cover differences to improve the accuracy of landslide monitoring and provide intelligent landslide risk classification. To achieve the above objectives, a first aspect of the present invention provides a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning, comprising: a data acquisition step of acquiring multi-source InSAR data of the target area; wherein the multi-source InSAR data includes Sentinel-1A data of the first period, ALOS-2 PALSAR-2 ScanSAR data of the first period, and LiCSAR data of the first period; a deformation interpretation step of performing deformation interpretation processing on a target area of ​​the multi-source InSAR data using a preset MT-InSAR method to obtain deformation rate data; wherein the deformation rate data includes first deformation rate data, second deformation rate data, and third deformation rate data corresponding to the Sentinel-1A data, the ALOS-2PALSAR-2 ScanSAR data, and the LiCSAR data, respectively; a regional deformation feature identification step, using a preset multi-source InSAR fusion optimization monitoring strategy to perform fusion analysis processing on the first deformation rate data, the second deformation rate data, and the third deformation rate data to obtain regional deformation detection data for the first period, and performing regional boundary delineation processing on the regional deformation detection data to obtain regional deformation feature data; In the regional morphological feature recognition step, the optical remote sensing image to be identified is input into the landslide classifier model built based on deep learning to output landslide morphological recognition data; The landslide monitoring classification step obtains landslide monitoring data based on the regional deformation characteristic data and the landslide morphology identification data; wherein the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data.

[0005] Preferably, the deformation interpretation step specifically includes: Extracting GCP point data from the Sentinel-1A data and the ALOS-2 PALSAR-2ScanSAR data using a preset PS-InSAR method; Using the GCP point data as input to a preset SBAS-InSAR method to perform phase correction processing to obtain the first deformation rate data and the second deformation rate data; The LiCSAR data are processed using a preset LICSBAS method to obtain third deformation rate data.

[0006] Preferably, the preset multi-source InSAR fusion optimization monitoring strategy is specifically: Based on the visual interpretation of optical remote sensing images, low vegetation cover areas and high vegetation cover areas were divided; The low vegetation coverage area is monitored using C-band data; The high vegetation coverage area is monitored using L-band data; The LiCSAR data are used as supplementary monitoring result data.

[0007] Preferably, the method further comprises a landslide classifier model construction step, comprising: a data set construction step of acquiring optical remote sensing images within a first period, constructing a landslide identification sample data set based on the optical remote sensing images, and performing data enhancement processing on the landslide identification sample data set to obtain a landslide identification data set; The model building step uses a preset deep learning model to perform model training on the landslide identification dataset to obtain the landslide classifier model.

[0008] Further preferably, performing data enhancement processing on the landslide identification sample dataset to obtain the landslide identification dataset specifically includes: labeling the landslide area in the optical remote sensing image based on the labelme labeling tool to obtain a labeled optical remote sensing image; Performing transformation processing on the annotated optical remote sensing image to obtain a transformed annotated optical remote sensing image; wherein the transformation processing includes at least four of rotating by 90 degrees, rotating by 180 degrees, rotating by 270 degrees, horizontal mirroring, and vertical mirroring; Binary mask data corresponding to the annotated optical remote sensing image and the transformed annotated optical remote sensing image are constructed; and landslide sample values ​​in the binary mask data are set to 0, and non-landslide sample values ​​are set to 1.

[0009] Preferably, the landslide monitoring classification steps are specifically as follows: All regional deformation feature data and landslide morphology identification data are superimposed and intersected, and the overlapping areas are identified as high-risk landslide point data; The data outside the overlapping area are identified as medium-risk landslide point data.

[0010] Preferably, the landslide monitoring and classification step further comprises obtaining landslide monitoring data based on the regional deformation characteristic data, landslide morphology identification data and historical landslide point data.

[0011] Further preferably, the landslide monitoring data obtained according to the regional deformation characteristic data, the landslide morphology identification data and the historical landslide point data is specifically: All regional deformation feature data, landslide morphology identification data and historical landslide point data are superimposed and intersected, and the overlapping areas are identified as high-risk landslide point data; The data outside the overlapping area are identified as medium-risk landslide data.

[0012] The second aspect of the present invention provides a landslide monitoring system for densely vegetated areas based on multi-source InSAR data fusion and deep learning, the system includes a data acquisition module, a deformation interpretation module, a regional deformation feature recognition module, a regional morphological feature recognition module and a landslide monitoring classification module, wherein The data acquisition module is configured to acquire multi-source InSAR data of a target area; wherein the multi-source InSAR data includes Sentinel-1A data of a first period, ALOS-2 PALSAR-2 ScanSAR data of a first period, and LiCSAR data of a first period; The deformation interpretation module is configured to perform deformation interpretation processing on the target area of ​​the multi-source InSAR data using a preset MT-InSAR method to obtain deformation rate data; wherein the deformation rate data includes first deformation rate data, second deformation rate data, and third deformation rate data corresponding to the Sentinel-1A data, the ALOS-2 PALSAR-2 ScanSAR data, and the LiCSAR data, respectively; The regional deformation feature identification module is configured to perform fusion analysis processing on the first deformation rate data, the second deformation rate data, and the third deformation rate data using a preset multi-source InSAR fusion optimization monitoring strategy to obtain regional deformation detection data for the first period, and to perform regional boundary delineation processing on the regional deformation detection data to obtain regional deformation feature data; The regional morphological feature recognition module is used to input the optical remote sensing image to be identified into the landslide classifier model constructed based on deep learning to output landslide morphological recognition data; The landslide monitoring classification module is used to obtain landslide monitoring data based on the regional deformation characteristic data and the landslide morphology identification data; wherein the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data.

[0013] Preferably, the system further comprises a landslide classifier model building module, which comprises a data set building module and a model building module, wherein The data set construction module is configured to obtain optical remote sensing images within a first period of time, construct a landslide identification sample data set based on the optical remote sensing images, and perform data enhancement processing on the landslide identification sample data set to obtain a landslide identification data set; The model building module is used to use a preset deep learning model to perform model training on the landslide identification dataset to obtain the landslide classifier model.

[0014] The embodiments of the present invention provide a landslide monitoring method and system for densely vegetated areas based on multi-source InSAR data fusion and deep learning. This method fuses multi-source InSAR data to identify regional deformation characteristics of a target area. The obtained regional deformation feature data is then combined with the target area morphology recognition using a landslide classifier model built based on deep learning. The method monitors and predicts landslide risks in the target area and classifies them into different levels. Compared with existing technologies, this method has at least the following significant benefits: 1. The present invention can effectively improve the detection rate in areas with high vegetation coverage.

[0015] 2. This invention combines multi-source InSAR with deep learning technology to effectively improve the overall accuracy of identifying landslide hazard points.

[0016] 3. The present invention proposes a grading standard that integrates the “deformation-morphology” elements, effectively improving the reliability of the landslide hazard level.

[0017] 4. The present invention also proposes a grading standard that integrates the three elements of "deformation-morphology-history" to further improve the reliability of landslide hazard levels.

[0018] 5. The present invention is based on high-quality regional deformation rate results and high-resolution optical images, combined with a deep learning model for landslide detection, replacing manual detection to achieve automatic detection, significantly improving recognition accuracy and work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning provided by the first embodiment of the present invention; Figure 2 Schematic diagram of deformation monitoring using Sentinel-1A data after deformation interpretation processing provided by an embodiment of the present invention; Figure 3 Schematic diagram of deformation monitoring using ALOS-2 PALSAR-2 ScanSAR data after deformation interpretation processing provided by an embodiment of the present invention; Figure 4 Schematic diagram of deformation monitoring of LiCSAR data after deformation interpretation processing provided by an embodiment of the present invention; Figure 5 Schematic diagram of multi-source InSAR fusion optimization deformation monitoring after fusion analysis and processing provided by an embodiment of the present invention; Figure 6 A schematic diagram of a high deformation area after regional deformation feature identification provided by an embodiment of the present invention; Figure 7 A schematic diagram of the recognition result after regional morphological feature recognition provided by an embodiment of the present invention; Figure 8 A schematic diagram of the classification of landslide hazard points after landslide monitoring classification processing provided by the first embodiment of the present invention; Figure 9 A system block diagram of a landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning provided by the first embodiment of the present invention; Figure 10A flow chart of a method for monitoring landslides in densely vegetated areas based on multi-source InSAR data fusion and deep learning, provided in accordance with the second embodiment of the present invention; Figure 11 A schematic diagram of sample enhancement processing and corresponding masks in a landslide identification sample dataset provided by an embodiment of the present invention; Figure 12 A system block diagram of a landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning, provided in accordance with the second embodiment of the present invention; Figure 13 A system block diagram of another landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning, provided in accordance with a second embodiment of the present invention; Figure 14 A flowchart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning provided in the third embodiment of the present invention; Figure 15 A flowchart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided for the fourth embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It is apparent that the embodiments described are only some, not all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0021] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0022] An embodiment of the present invention provides a landslide monitoring method for densely vegetated areas based on multi-source InSAR data fusion and deep learning. The method applies multi-source InSAR data and deep learning technology to landslide monitoring technology, changes the traditional monitoring method of identification based only on deformation or morphology, realizes a landslide point identification method that integrates deformation, morphology and / or historical point data elements, and realizes a grading standard for element fusion.

[0023] [First embodiment] [A landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning] Figure 1 A flow chart of a landslide monitoring method in dense vegetation areas based on multi-source InSAR data fusion and deep learning is provided in the first embodiment of the present invention. Figure 1, the technical solution of the present invention is described with specific embodiments.

[0024] like Figure 1 As shown, the present invention provides a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning, which includes the following steps: S1100, data acquisition step, acquiring multi-source InSAR data of the target area.

[0025] Specifically, in the present invention, the target area refers to the landslide risk monitoring area to which the method of the embodiments of the present invention is applied, that is, the specific area where landslide risk needs to be confirmed. Multi-source InSAR data is obtained externally, for example, by reading from a proprietary database or searching for corresponding data from publicly available online. The multi-source InSAR data includes Sentinel-1A data from the first period, Advanced Land Observing Satellite-2 Phased Array L-band Synthetic Aperture Radar 2 ScanSAR (ALOS-2 PALSAR-2 ScanSAR) data from the first period, and Looking into Continents from Space with Synthetic Aperture Radar (LiCSAR) data from the first period. The first period refers to a specific time period, for example, from January 1, 2020 to December 30, 2024, the last 5 years from the date of using the method, the last 10 years, etc. The specific time period can be set according to the specific circumstances of using the method and the available data. The data types include Sentinel-1A data, ALOS-2 PALSAR-2 ScanSAR data, and LiCSAR data. Each data type can also include ascending orbit images and / or descending orbit images. In the embodiment of the present invention, when acquiring InSAR data, there is no restriction on the number of each data type, and it can be adjusted and set according to the actual situation of using the present invention.

[0026] In a specific example of an embodiment of the present invention, the acquired multi-source InSAR data specifically includes the following: Sentinel-1A orbit-raising images, totaling 60, from January 12, 2020, to December 16, 2024; Sentinel-1A deorbit images, totaling 60, from January 19, 2020, to December 23, 2024; 33 ALOS-2PALSAR-2ScanSAR down-orbit images from January 3, 2020, to March 24, 2023; From January 2020 to December 2024, LiCSAR conducted 1,082 ascending and 500 descending interferometry pairs and their untangling results.

[0027] S1200, deformation interpretation step, using a preset MT-InSAR method to perform deformation interpretation processing on the target area of ​​the multi-source InSAR data to obtain deformation rate data.

[0028] Specifically, Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) is an extension of InSAR technology that utilizes multi-temporal SAR image data to generate time series information on surface deformation. This technology is commonly used to monitor subtle surface deformation changes, such as earthquakes, volcanic activity, landslides, or ground subsidence. The MT-InSAR method in this invention is a pre-defined method, selected from existing technologies that are applicable to the multi-source InSAR data of this invention.

[0029] The present invention uses different methods to interpret deformation for three different types of InSAR data, obtaining deformation rate data corresponding to each type of data. For example, Sentinel-1A and ALOS-2 PALSAR-2 ScanSAR data are processed using the Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) method. Reference points from the Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) method are used as Ground Control Points (GCPs) for phase optimization in SBAS-InSAR. Regional LiCSAR data are processed using the Looking into Continents with Small Baseline Subset (LICSBAS) method. The SBAS-InSAR, PS-InSAR, and LICSBAS methods are commonly available methods in the prior art, as well as their improved methods.

[0030] After processing, the obtained deformation rate data include first deformation rate data corresponding to Sentinel-1A data, second deformation rate data corresponding to ALOS-2 PALSAR-2 ScanSAR data, and third deformation rate data corresponding to LiCSAR data.

[0031] In a specific example of the embodiment of the present invention, the step further includes the following steps: S1201, using the preset PS-InSAR method to extract GCP point data from Sentinel-1A data and ALOS-2 PALSAR-2ScanSAR data.

[0032] Specifically, PS-InSAR is an InSAR technology based on persistent scatterers (PS). It extracts high-precision time series of surface deformation by identifying and analyzing the interferometric phase of PS points. It is suitable for millimeter-scale deformation monitoring in cities, areas with dense artificial infrastructure, or areas with stable terrain. The present invention uses a preset PS-InSAR method to extract GCP point data from Sentinel-1A data and ALOS-2 PALSAR-2 ScanSAR data. In this embodiment of the present invention, the preset PS-InSAR method is selected from commonly available PS-InSAR methods and improved methods.

[0033] S1202: Using the GCP point data as input to a preset SBAS-InSAR method to perform phase correction processing to obtain first deformation rate data and second deformation rate data.

[0034] Specifically, the SBAS-InSAR method combines the InSAR technology of the small baseline set strategy to generate high-density, high-reliability deformation time series by screening high-quality image pairs. It is suitable for large-scale surface deformation monitoring, such as earthquakes, volcanoes, landslides, urban subsidence, etc., and performs particularly well in areas with vegetation cover or complex terrain. In an embodiment of the present invention, GCP point data is used as input to the preset SBAS-InSAR method for phase correction processing to obtain first deformation rate data and second deformation rate data. In an embodiment of the present invention, the preset SBAS-InSAR method is selected from the general SBAS-InSAR methods available in the prior art and methods for improving them.

[0035] S1203: Process the LiCSAR data using a preset LICSBAS method to obtain third deformation rate data.

[0036] Specifically, LICSBAS is an open-source InSAR time series analysis package integrated with LiCSAR. This method can be used to monitor and analyze landslides using SAR data. The present invention uses a pre-defined LICSBAS to process LiCSAR data and obtain the corresponding third deformation rate data.

[0037] Regarding steps S1201 to S203 of the present invention, a specific implementation process in a specific example of an embodiment of the present invention includes: First, the reference points in PS-InSAR were applied to SBAS-InSAR as GCP control points for phase optimization. A SAR image in the time series was selected as the primary image, and all auxiliary images were registered with the primary image, obtaining a total of 59 relative positions (32 relative positions for ALOS-2 PALSAR-2 ScanSAR data). All auxiliary images were interferometrically processed with the primary image, and the interferometric results were atmospherically corrected using GACOS data. The multi-look ratio was set to 4:1 (2:12 for ALOS-2 PALSAR-2 ScanSAR data), and the Shuttle Radar Topography Mission Version 1 Digital Elevation Model 30-meter Resolution (SRTMV1 30m DEM) data was used for registration. All interferometric processing results were subjected to the first inversion, with the maximum reference point area set to 25sqkm, the sub-area overlap ratio to 30%, and the number of candidates to 300. The reference GCP point files in the first inversion results were converted into GCP files in the SAR coordinate system. The GCP files were applied to the first SBAS-InSAR inversion for phase optimization and re-flattening.

[0038] Secondly, the regional Sentinel-1A data and ALOS-2 PALSAR-2 ScanSAR data were processed using the SBAS-InSAR method, including: selecting a SAR image in the time series as the main image, and checking the temporal and spatial baseline distributions, setting the maximum spatial baseline to 4.6% (300m), the maximum temporal baseline to 120d, and the minimum number of connections to 10; aligning all auxiliary images with the main image, obtaining a total of 215 relative positions for ascending orbits and 204 relative positions for descending orbits (85 relative positions for ALOS-2 PALSAR-2 ScanSAR data); interfering all auxiliary images with the main image, applying the GCP point file generated in step S211 for phase optimization and re-leveling, using GACOS data for atmospheric correction of the interferometric results, with a multi-look ratio of 4:1 (2:12 for ALOS-2 PALSAR-2 ScanSAR data), using SRTMV1 30m DEM data for registration, and selecting the minimum cost flow (MCF) unwrapping method. Flow, MCF), the unwrapping decomposition level is level 1, the correlation threshold is 0.3, and the filtering method is Goldstein; all interference processing results are inverted for the first time, and the GCP point file generated in step S211 is used for phase optimization and re-flattening. The product correlation threshold is 0.3, the deformation model is Liner, the spatial wavelet size is 120m, the unwrapping method is Minimum Cost Flow, the unwrapping decomposition level is level 1, the correlation threshold is 0.3, and the filtering method is Goldstein; based on the first inversion result, a second inversion is performed, and the GCP point file generated in step S211 is used for phase optimization and re-flattening. The product correlation threshold is 0.3, the atmospheric low-pass filter is 1600m, and the atmospheric high-pass filter is 365d; the second inversion result is geocoded and output as the result of the WGS-84 geographic coordinate system with a resolution of 30m, and the preliminary deformation rate is obtained, including the first deformation rate data corresponding to the Sentinel-1A data and the first deformation rate data corresponding to the ALOS-2 PALSAR-2 The second deformation rate data corresponding to the ScanSAR data.

[0039] Then, the regional LiCSAR data were processed using the LICSBAS method, including: preprocessing all downloaded LiCSAR data products and applying GACOS data for atmospheric correction with a mask threshold of 0.2; quality checking all preprocessed results to identify and remove bad interferograms with an unwrapping threshold of 0.3 and a correlation threshold of 0.2; loop closure checking all quality-checked results to identify and remove bad interferograms with a ring error parameter of 3 rad; performing small baseline inversion on all quality-checked results; calculating the standard deviation of deformation rates for the inversion results; masking and filtering the time series, and outputting the filtered deformation rate, which is the third deformation rate data corresponding to the LiCSAR data.

[0040] In the above specific example of the present invention, after processing through the above steps, first deformation rate data corresponding to Sentinel-1A data, second deformation rate data corresponding to ALOS-2 PALSAR-2 ScanSAR data, and third deformation rate data corresponding to LiCSAR data are obtained. Figure 2 The following is a schematic diagram of deformation monitoring of Sentinel-1A data after deformation interpretation processing provided by an embodiment of the present invention, which shows a schematic diagram of deformation monitoring of Sentinel-1A data after deformation interpretation processing of Sentinel-1A down-orbit images, as shown in FIG. Figure 2 As shown, it includes first deformation rate data. Figure 3 The schematic diagram of deformation monitoring of ALOS-2 PALSAR-2 ScanSAR data including first deformation rate data provided by an embodiment of the present invention is a schematic diagram of deformation monitoring of ALOS-2 PALSAR-2 ScanSAR data after deformation interpretation processing of ALOS-2 PALSAR-2 ScanSAR down-orbit images, as shown in FIG. Figure 3 As shown, it includes second deformation rate data. Figure 4 The schematic diagram of deformation monitoring of LiCSAR data after deformation interpretation processing provided by an embodiment of the present invention is a schematic diagram of deformation monitoring of LiCSAR data after deformation interpretation processing of LiCSAR down-orbit images, as shown in FIG. Figure 4 As shown, it contains third deformation rate data.

[0041] S1300, regional deformation feature identification step, uses a preset multi-source InSAR fusion optimization monitoring strategy to fuse and analyze the first deformation rate data, the second deformation rate data, and the third deformation rate data to obtain regional deformation detection data for the first period, and then performs regional boundary delineation processing on the regional deformation detection data to obtain regional deformation feature data.

[0042] Specifically, in the embodiment of the present invention, the preset multi-source InSAR fusion optimization monitoring strategy is a strategy preset before the implementation of the present invention, which specifically includes: Based on the visual interpretation of optical remote sensing images, low vegetation coverage areas and high vegetation coverage areas are divided. Specifically, the differences in the impact of vegetation on SAR signals include: low vegetation coverage areas: vegetation height is low and density is low, and the backscatter of C-band SAR mainly comes from the surface, which is suitable for monitoring terrain deformation or surface changes. High vegetation coverage areas: vegetation is dense, and C-band signals are easily reflected or attenuated by vegetation, while L-band has strong penetration ability, which can reduce vegetation interference and obtain surface information under the forest. The low vegetation coverage areas and high vegetation coverage areas are divided based on the visual interpretation of optical remote sensing images through the GIS platform. It can realize the selection of appropriate bands for different vegetation coverage types, avoid data redundancy, and improve monitoring accuracy and efficiency.

[0043] C-band data is used to monitor areas with low vegetation cover. Specifically, the C-band has a high spatial resolution (e.g., 10-20 m for Sentinel-1), making it suitable for monitoring subtle surface changes (such as soil cracks and small-scale subsidence) in low vegetation areas. It is sensitive to short-lived vegetation (such as grasslands and crops), enabling time series analysis of vegetation growth or land use change. In low vegetation cover areas, C-band (Sentinel-1A) data is used for monitoring. C-band data accurately captures surface deformation with an accuracy of millimeters to centimeters.

[0044] L-band data is used to monitor areas with high vegetation cover. Specifically, the L-band has strong penetrating power, capable of penetrating dense vegetation (such as forest canopies) to obtain information on understory topography or surface deformation. It is sensitive to soil moisture and surface roughness, making it suitable for monitoring hydrological or geological changes (such as landslides and permafrost thaw) in areas with high vegetation cover. L-band (ALOS-2 PALSAR-2 ScanSAR) data is used for monitoring areas with high vegetation cover. L-band data effectively penetrates vegetation, reduces canopy interference, and extracts understory topography or deformation information with vertical accuracy ranging from a few decimeters to meters.

[0045] LiCSAR data can be used as a supplement to monitoring results. Specifically, LiCSAR data can provide lightweight, fast-processing monitoring results, compensating for the shortcomings of C / L-band data in temporal resolution or coverage, and enhancing data diversity. In areas with missing data or complex terrain (such as mountainous areas and the polar regions), LiCSAR data can be used as a supplement to improve monitoring robustness and its adaptability to specific scenarios. As a supplement to monitoring results, LiCSAR data can enhance rapid response capabilities to extreme environments (such as heavy rain and earthquakes) and assist in the rapid identification of disaster events (such as earthquakes and volcanoes) or temporary surface changes.

[0046] The multi-source InSAR fusion optimization monitoring strategy provided in the embodiments of the present invention uses visual interpretation of optical images to classify vegetation cover types. Combining the complementary advantages of C-band and L-band SAR data, it achieves precise monitoring of low and high vegetation areas. Furthermore, the inclusion of LiCSAR data complements this approach, enhancing the comprehensiveness of monitoring. This multi-band collaborative strategy offers significant advantages in improving accuracy, adapting to complex environments, and increasing efficiency, and is therefore well-suited to the geological morphology and natural conditions of the target area of ​​the present invention.

[0047] In the embodiment of the present invention, this step specifically includes the following steps: S1301: Using a preset multi-source InSAR fusion optimization monitoring strategy, a fusion analysis is performed on the first deformation rate data, the second deformation rate data, and the third deformation rate data to obtain regional deformation detection data for the first time period. Specifically, after performing the fusion analysis using the preset multi-source InSAR fusion optimization monitoring strategy, corresponding regional deformation detection data is obtained. Figure 5 Schematic diagram of multi-source InSAR fusion optimization deformation monitoring after fusion analysis and processing provided by an embodiment of the present invention, as shown in FIG. Figure 5 As shown, it includes regional deformation detection data.

[0048] S1302: Delineate the regional boundaries of the regional deformation detection data to obtain regional deformation feature data. Specifically, a preset deep learning method, such as a deep learning model based on "Unet-Resnet50", is used to delineate the regional boundaries of the regional deformation detection data to obtain regional deformation feature data.

[0049] In an embodiment of the present invention, this step can also be performed in a manner including the following steps, specifically including: after fusing and analyzing the first deformation rate data, the second deformation rate data, and the third deformation rate data using a preset InSAR fusion optimization monitoring strategy, the areas where the annual average deformation rate is greater than 20 mm / y in the ascending and descending orbit deformation results are respectively drawn according to the deformation monitoring results to obtain high deformation areas. Figure 6 A schematic diagram of a high deformation area after regional deformation feature recognition provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the high deformation area indicated by the high deformation area data is included.

[0050] S1400, a regional morphological feature recognition step, inputs the optical remote sensing image to be recognized into a landslide classifier model constructed based on deep learning to output landslide morphological recognition data.

[0051] Specifically, in the embodiments of the present invention, the optical remote sensing image to be identified refers to an optical remote sensing image that requires landslide monitoring and identification. The landslide classifier model constructed based on deep learning provided in the embodiments of the present invention can be a pre-trained landslide classifier model that is built into the method of the present invention. Alternatively, it can be a landslide classifier model constructed based on deep learning using the method provided by the present invention. In a second embodiment of the present invention, the method for constructing the landslide classifier model provided by the present invention will be described in detail.

[0052] In this step, the optical remote sensing image to be identified is input into the landslide classifier model built based on deep learning, and the model can directly output landslide morphology recognition data. Figure 7 A schematic diagram of the recognition result after regional morphological feature recognition provided by an embodiment of the present invention is shown in FIG. Figure 7 As shown, it includes landslide morphology identification data.

[0053] S1500, a landslide monitoring classification step, obtains landslide monitoring data based on regional deformation characteristic data and landslide morphology identification data.

[0054] Specifically, in this embodiment of the present invention, this step is implemented by superimposing and intersecting all regional deformation feature data and landslide morphology identification data, identifying the overlapping areas as high-risk landslide point data, and identifying the data outside the overlapping areas as medium-risk landslide point data. In other words, the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data. Figure 8 This is a schematic diagram of the classification of landslide hazard points after landslide monitoring classification processing provided by the first embodiment of the present invention, as shown in FIG. Figure 8 As shown, these include high-risk landslide points formed by high-risk landslide point data and medium-risk landslide points formed by medium-risk landslide point data, which are also landslide monitoring data.

[0055] The above describes in detail, in conjunction with the accompanying drawings, a method for monitoring landslides in densely vegetated areas based on multi-source InSAR data fusion and deep learning, provided by the first embodiment of the present invention. The present invention also provides a system for monitoring landslides in densely vegetated areas based on multi-source InSAR data fusion and deep learning, which is used to implement the method for monitoring landslides in densely vegetated areas based on multi-source InSAR data fusion and deep learning provided by the present invention. The following describes in detail a system for monitoring landslides in densely vegetated areas based on multi-source InSAR data fusion and deep learning, provided by the present invention, in conjunction with the accompanying drawings.

[0056] [A landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning] Figure 9The system block diagram of a landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning provided by the first embodiment of the present invention is as follows: Figure 9 As shown, the monitoring system 1000 provided by the present invention includes a data acquisition module 1001, a deformation interpretation module 1002, a regional deformation feature recognition module 1003, a regional morphological feature recognition module 1004 and a landslide monitoring classification module 1005, wherein A data acquisition module 1001 is configured to acquire multi-source InSAR data of a target area, wherein the multi-source InSAR data includes Sentinel-1A data of a first period, ALOS-2 PALSAR-2 ScanSAR data of a first period, and LiCSAR data of a first period. The deformation interpretation module 1002 is configured to perform deformation interpretation processing on the target area of ​​the multi-source InSAR data using a preset MT-InSAR method to obtain deformation rate data; wherein the deformation rate data includes first deformation rate data, second deformation rate data, and third deformation rate data corresponding to the Sentinel-1A data, ALOS-2PALSAR-2 ScanSAR data, and LiCSAR data, respectively; The regional deformation feature identification module 1003 is configured to use a preset multi-source InSAR fusion optimization monitoring strategy to perform fusion analysis on the first deformation rate data, the second deformation rate data, and the third deformation rate data to obtain regional deformation detection data for the first period, and to perform regional boundary delineation on the regional deformation detection data to obtain regional deformation feature data. The regional morphological feature recognition module 1004 is used to input the optical remote sensing image to be identified into the landslide classifier model constructed based on deep learning to output landslide morphological recognition data; The landslide monitoring classification module 1005 is used to obtain landslide monitoring data based on regional deformation characteristic data and landslide morphology identification data; wherein the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data.

[0057] The method and system provided by the first embodiment of the present invention are described in detail above by combining the drawings with the embodiments. Next, the second embodiment of the present invention will be described in detail with reference to the drawings.

[0058] [Second embodiment] [A landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning] In addition to providing the relevant method steps in the first embodiment, the second embodiment provided by the present invention also provides a method for constructing a landslide classifier model based on deep learning. The method provided by the second embodiment of the present invention is described in detail below with reference to the accompanying drawings. Figure 10 A flow chart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided for the second embodiment of the present invention, as shown in FIG. Figure 10 As shown, in addition to steps S2100 to S2500 corresponding to steps S1110 to S1500 in the first embodiment of the present invention, the following steps are also included: S2600, a landslide classifier model construction step, constructing a data set to train a preset deep learning model to construct a landslide classifier model.

[0059] Specifically, in an embodiment of the present invention, a method for constructing a landslide classifier model is provided, which specifically includes the following steps: In the data set construction step S2601, optical remote sensing images within a first period are acquired, a landslide identification sample data set is constructed based on the optical remote sensing images, and data enhancement processing is performed on the landslide identification sample data set to obtain a landslide identification data set.

[0060] Specifically, in an embodiment of the present invention, optical remote sensing images within a first time period are acquired externally. The specific data source for acquiring the optical remote sensing images is determined based on the specific target scenario and available data for the method of the present invention; the present invention does not impose any restrictions on the data source. Optical remote sensing images that coincide with the multi-source InSAR data of the target area are acquired externally. Based on the optical remote sensing images, a landslide identification sample dataset is generated and data enhancement processing is performed. The dataset is then divided into training and test sets to construct a landslide identification dataset.

[0061] In the embodiment of the present invention, data enhancement processing is performed on the landslide identification sample dataset to obtain the landslide identification dataset, which specifically includes: Firstly, the landslide area in the optical remote sensing image is labeled based on the labelme labeling tool to obtain the labeled optical remote sensing image.

[0062] Specifically, in one example provided by an embodiment of the present invention, the optical remote sensing image is a high-resolution optical satellite image. Using the LabelMe annotation tool, typical landslide areas in the high-resolution optical satellite image are annotated, thereby constructing a labeled optical remote sensing image. After annotating 90 optical remote sensing image data, a custom landslide dataset totaling 90 data points is obtained. Next, the annotated optical remote sensing image is transformed to obtain a transformed annotated optical remote sensing image.

[0063] Specifically, in the embodiments of the present invention, the transformation processing for annotating the optical remote sensing image includes at least four of the following: 90° rotation, 180° rotation, 270° rotation, horizontal mirroring, and vertical mirroring. The selection of the transformation methods can be based on the size of the dataset and sample characteristics, and is not restricted herein.

[0064] In this embodiment of the present invention, an existing landslide dataset is also provided. The self-made landslide dataset obtained after the annotation process in the previous step together constitutes the dataset required by the present invention. The annotated optical remote sensing images (samples) in the existing and self-made landslide datasets were randomly rotated 90°, 180°, and 270°, and vertically mirrored. A total of 3440 image data items were obtained, including 3080 existing data items and 360 self-made data items.

[0065] As can be seen from the above examples, the number of samples in the existing and self-made landslide datasets has been expanded through enhancement processing. Finally, binary mask data corresponding to the annotated optical remote sensing image and the transformed annotated optical remote sensing image are constructed; the landslide sample values ​​in the binary mask data are set to 0, and the non-landslide sample values ​​are set to 1. Figure 11 The enhanced processing of samples in the landslide identification sample dataset provided by an embodiment of the present invention and the corresponding mask schematic diagram are shown in the figure, which includes a sample and its sample data obtained by random rotation of 90°, 180°, 270° and vertical mirroring operations and the corresponding mask schematic diagram.

[0066] In the embodiment of the present invention, after respectively processing the samples in the existing landslide data set and the self-made landslide data set in the data set, the obtained data sets constitute the landslide identification data set of the present invention.

[0067] In an embodiment of the present invention, the constructed landslide identification dataset is divided into a training set and a validation set, which are used for training and validating a preset deep learning model, respectively.

[0068] In the model building step S2602, a preset deep learning model is used to perform model training on the landslide identification dataset to obtain a landslide classifier model.

[0069] Specifically, the preset deep learning model provided in this embodiment of the present invention is the "Unet-Resnet50" deep learning model, which is built and trained based on the deep learning framework of the open source Python machine learning library PyTorch. This model is trained using a landslide identification dataset to obtain the landslide classifier model provided by this invention. In this embodiment of the present invention, the preset deep learning model is not limited to the "Unet-Resnet50" deep learning model; other suitable general deep learning models may also be used. The selection of the preset deep learning model can be based on the regional and natural characteristics of the target area for which the present invention is being used.

[0070] The landslide classifier model provided by the embodiment of the present invention is constructed, in a specific example, including the following steps: First, the samples in the constructed landslide identification dataset are cropped to 256×256 pixels.

[0071] Secondly, the training set and validation set are randomly divided into 7:3, with 2408 training set data and 1032 validation set data.

[0072] Again, the “Unet-Resnet50” deep learning model was trained using the training set for 2000 iterations.

[0073] Next, the trained deep learning model is used to predict the data of the test set to verify the accuracy of the deep learning model. The training model test results are train_loss: 0.0106, val_loss: 0.0537, lr: 0.000000, Accuracy (PA): 98.7, Precision (CPA): ['99.1', '94.6'], Recall: ['99.5', '90.9'], IoU: ['98.6', '86.4'], MPA: 96.9, MIoU: 92.5, which meets the preset requirements. At this time, the landslide classifier model construction of the present invention is completed.

[0074] It should be noted that the landslide classifier model construction step S2600 provided in the embodiment of the present invention can be executed before step S2000 or before step S2400, and the specific execution order is not restricted here. Figure 10 As shown, the landslide classifier model construction step S2600 is placed before step S2400. The specific implementation of steps S2100 to S2500 has been introduced in the above embodiment 1 and will not be repeated here.

[0075] [A landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning] Correspondingly, the second embodiment of the present invention also provides another densely vegetated area landslide monitoring system 2000 based on multi-source InSAR data fusion and deep learning, which is used to implement the densely vegetated area landslide monitoring method based on multi-source InSAR data fusion and deep learning provided by the present invention. Figure 12 The system block diagram of a landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided in the second embodiment of the present invention. Figure 12As shown, the system 2000 includes, in addition to the data acquisition module 2001, deformation interpretation module 1002, regional deformation feature recognition module 1003, regional morphological feature recognition module 1004 and landslide monitoring and classification module 1005 corresponding to the data acquisition module 1001, deformation interpretation module 1002, regional deformation feature recognition module 2003, regional morphological feature recognition module 2004 and landslide monitoring and classification module 1005 of the first embodiment, a landslide classifier model construction module 2006 for constructing a landslide classifier model.

[0076] Figure 13 A system block diagram of another landslide monitoring system in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided for the second embodiment of the present invention. Figure 13 As shown, in another specific example of the embodiment of the present invention, the landslide classifier model construction module 2006 further includes a data set construction module 20061 and a model construction module 20062, wherein The data set construction module 20061 is used to obtain optical remote sensing images within the first period, construct a landslide identification sample data set based on the optical remote sensing images, and perform data enhancement processing on the landslide identification sample data set to obtain a landslide identification data set; The model building module 20062 is used to use a preset deep learning model to perform model training on the landslide identification dataset to obtain a landslide classifier model.

[0077] Above, a detailed introduction is given to a landslide monitoring method and monitoring system for densely vegetated areas based on multi-source InSAR data fusion and deep learning provided by the second embodiment of the present invention. The second embodiment of the present invention provides a landslide monitoring method and monitoring system for densely vegetated areas based on multi-source InSAR data fusion and deep learning. The landslide classifier model constructed by the method is highly targeted. By utilizing the method, which is based on multi-source InSAR data fusion and deep learning, the accuracy of landslide monitoring in the target area of ​​the present invention is improved.

[0078] Next, the third embodiment of the present invention will be introduced with reference to the accompanying drawings.

[0079] [Third embodiment] [A landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning] The third embodiment provided by the present invention provides a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning, and proposes a classification standard for the fusion of the three elements of "deformation-morphology-history" to further improve the reliability of the landslide hazard level. The method provided by the third embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0080] Figure 14 A flow chart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided for the third embodiment of the present invention. Figure 14 As shown, the third embodiment of the present invention provides a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning, including steps S3100 to S3500, wherein steps S3100 to S3400 respectively correspond to steps S1100 to S1400 provided in the first embodiment. The specific implementation methods of each step have been introduced in the above-mentioned embodiment one and will not be repeated here. In this embodiment, step S3500 replaces S1500 in embodiment one to realize the "deformation-morphology-history" three-factor fusion grading standard.

[0081] S3500, a landslide monitoring classification step, obtains landslide monitoring data based on regional deformation characteristic data, landslide morphology identification data, and historical landslide point data.

[0082] Specifically, in this embodiment of the present invention, historical landslide point data is pre-acquired, retained landslide point data indicating a previous landslide in the target area. Landslide monitoring data is obtained based on regional deformation feature data, landslide morphology identification data, and historical landslide point data by superimposing and intersecting all regional deformation feature data, landslide morphology identification data, and historical landslide point data, and identifying the overlapping areas as high-risk landslide point data. Data outside of the overlapping areas is identified as medium-risk landslide data.

[0083] [Fourth embodiment] [A landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning] The fourth embodiment provided by the present invention is provided as a variation of the second embodiment. Figure 15 A flowchart of a landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning is provided for the fourth embodiment of the present invention. Figure 15 As shown, in addition to steps S4100 to S4400 and S4600 corresponding to steps S2100 to S2400 and S2600 provided in the second embodiment, step S4500 is also provided to replace S2500 in the second embodiment. The specific implementation method of step S4500 corresponds to step S3500 in the third embodiment and is not repeated here.

[0084] The embodiments of the present invention are described in detail above with reference to the accompanying drawings.

[0085] Embodiments of the present invention provide a landslide monitoring method and system for densely vegetated areas based on multi-source InSAR data fusion and deep learning. This method utilizes a zoning optimization monitoring strategy that fuses multi-source InSAR data (C-band Sentinel-1A for low-vegetation areas, L-band ALOS-2 for high-vegetation areas, and LICSAR as supplementary data). This method combines MT-InSAR phase optimization processing technology (using reference points from PS-InSAR as GCP control points in SBAS-InSAR for phase optimization, reducing manual point selection errors) with the automated LICSBAS process. This method effectively overcomes the decoherence problem in densely vegetated areas of high mountain valleys and produces high-precision regional deformation rate maps. This method also provides a deep learning-based landslide classifier model constructed based on a pre-set deep learning model. This model utilizes an enhanced landslide sample dataset to automatically identify landslide morphology with an accuracy rate of 98.7%. By establishing a "deformation-morphology-history" three-factor fusion classification standard (matching deformation characteristics, landslide morphology, and historical disaster locations), it enables intelligent classification of high- and medium-risk landslide sites. This method significantly improves the accuracy and efficiency of identifying landslide hazards in densely vegetated areas, overcoming the limitations of traditional monitoring methods such as cloud cover and manual interpretation. It provides reliable technical support for early disaster warning and risk prevention and control, effectively protecting people's lives and property. Compared with existing technologies, it has at least the following significant effects: 1. The present invention can effectively improve the detection rate in areas with high vegetation coverage.

[0086] 2. This invention combines multi-source InSAR with deep learning technology to effectively improve the overall accuracy of identifying landslide hazard points.

[0087] 3. The present invention proposes a grading standard that integrates the “deformation-morphology” elements, effectively improving the reliability of the landslide hazard level.

[0088] 4. The present invention also proposes a grading standard that integrates the three elements of "deformation-morphology-history" to further improve the reliability of landslide hazard levels.

[0089] 5. The present invention is based on high-quality regional deformation rate results and high-resolution optical images, combined with a deep learning model for landslide detection, replacing manual detection to achieve automatic detection, significantly improving recognition accuracy and work efficiency.

[0090] Professionals should also be further aware that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0091] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0092] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A landslide monitoring method in densely vegetated areas based on multi-source InSAR data fusion and deep learning, characterized by: The method comprises: a data acquisition step of acquiring multi-source InSAR data of the target area; wherein the multi-source InSAR data includes Sentinel-1A data of the first period, ALOS-2 PALSAR-2 ScanSAR data of the first period, and LiCSAR data of the first period; a deformation interpretation step of performing deformation interpretation processing on a target area of ​​the multi-source InSAR data using a preset MT-InSAR method to obtain deformation rate data; wherein the deformation rate data includes first deformation rate data, second deformation rate data, and third deformation rate data corresponding to the Sentinel-1A data, the ALOS-2PALSAR-2 ScanSAR data, and the LiCSAR data, respectively; a regional deformation feature identification step, using a preset multi-source InSAR fusion optimization monitoring strategy to perform fusion analysis processing on the first deformation rate data, the second deformation rate data, and the third deformation rate data to obtain regional deformation detection data for the first period, and performing regional boundary delineation processing on the regional deformation detection data to obtain regional deformation feature data; In the regional morphological feature recognition step, the optical remote sensing image to be identified is input into the landslide classifier model built based on deep learning to output landslide morphological recognition data; The landslide monitoring classification step obtains landslide monitoring data based on the regional deformation characteristic data and the landslide morphology identification data; wherein the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data.

2. The method according to claim 1, characterized in that The deformation interpretation step specifically includes: Extracting GCP point data from the Sentinel-1A data and the ALOS-2 PALSAR-2 ScanSAR data using a preset PS-InSAR method; Using the GCP point data as input to a preset SBAS-InSAR method to perform phase correction processing to obtain the first deformation rate data and the second deformation rate data; The LiCSAR data are processed using a preset LICSBAS method to obtain third deformation rate data.

3. The method according to claim 1, characterized in that The preset multi-source InSAR fusion optimization monitoring strategy is specifically as follows: Based on the visual interpretation of optical remote sensing images, low vegetation cover areas and high vegetation cover areas were divided; The low vegetation coverage area is monitored using C-band data; The high vegetation coverage area is monitored using L-band data; The LiCSAR data are used as supplementary monitoring result data.

4. The method according to claim 1, wherein The method further comprises a landslide classifier model construction step, comprising: a data set construction step of acquiring optical remote sensing images within a first period, constructing a landslide identification sample data set based on the optical remote sensing images, and performing data enhancement processing on the landslide identification sample data set to obtain a landslide identification data set; The model building step uses a preset deep learning model to perform model training on the landslide identification dataset to obtain the landslide classifier model.

5. The method according to claim 4, characterized in that The performing data enhancement processing on the landslide identification sample dataset to obtain the landslide identification dataset specifically includes: labeling the landslide area in the optical remote sensing image based on the labelme labeling tool to obtain a labeled optical remote sensing image; Performing transformation processing on the annotated optical remote sensing image to obtain a transformed annotated optical remote sensing image; wherein the transformation processing includes at least four of rotating by 90 degrees, rotating by 180 degrees, rotating by 270 degrees, horizontal mirroring, and vertical mirroring; Binary mask data corresponding to the annotated optical remote sensing image and the transformed annotated optical remote sensing image are constructed; and landslide sample values ​​in the binary mask data are set to 0, and non-landslide sample values ​​are set to 1.

6. The method according to claim 1, characterized in that The landslide monitoring classification steps are specifically as follows: All regional deformation feature data and landslide morphology identification data are superimposed and intersected, and the overlapping areas are identified as high-risk landslide point data; The data outside the overlapping area are identified as medium-risk landslide point data.

7. The method according to claim 1, characterized in that The landslide monitoring classification step further includes obtaining landslide monitoring data based on the regional deformation characteristic data, landslide morphology identification data and historical landslide point data.

8. The method according to claim 7, characterized in that The landslide monitoring data obtained based on the regional deformation characteristic data, landslide morphology identification data and historical landslide point data are as follows: All regional deformation feature data, landslide morphology identification data and historical landslide point data are superimposed and intersected, and the overlapping areas are identified as high-risk landslide point data; The data outside the overlapping area are identified as medium-risk landslide data.

9. A landslide monitoring system for densely vegetated areas based on multi-source InSAR data fusion and deep learning, characterized by: The system includes a data acquisition module, a deformation interpretation module, a regional deformation feature recognition module, a regional morphological feature recognition module and a landslide monitoring and classification module, wherein The data acquisition module is configured to acquire multi-source InSAR data of a target area; wherein the multi-source InSAR data includes Sentinel-1A data of a first period, ALOS-2 PALSAR-2 ScanSAR data of a first period, and LiCSAR data of a first period; The deformation interpretation module is configured to perform deformation interpretation processing on the target area of ​​the multi-source InSAR data using a preset MT-InSAR method to obtain deformation rate data; wherein the deformation rate data includes first deformation rate data, second deformation rate data, and third deformation rate data corresponding to the Sentinel-1A data, the ALOS-2 PALSAR-2 ScanSAR data, and the LiCSAR data, respectively; The regional deformation feature identification module is configured to perform fusion analysis processing on the first deformation rate data, the second deformation rate data, and the third deformation rate data using a preset multi-source InSAR fusion optimization monitoring strategy to obtain regional deformation detection data for the first period, and to perform regional boundary delineation processing on the regional deformation detection data to obtain regional deformation feature data; The regional morphological feature recognition module is used to input the optical remote sensing image to be identified into the landslide classifier model constructed based on deep learning to output landslide morphological recognition data; The landslide monitoring classification module is used to obtain landslide monitoring data based on the regional deformation characteristic data and the landslide morphology identification data; wherein the landslide monitoring data includes high-risk landslide point data and medium-risk landslide point data.

10. The system according to claim 9, characterized in that The system also includes a landslide classifier model building module, which includes a data set building module and a model building module, wherein The data set construction module is configured to obtain optical remote sensing images within a first period of time, construct a landslide identification sample data set based on the optical remote sensing images, and perform data enhancement processing on the landslide identification sample data set to obtain a landslide identification data set; The model building module is used to use a preset deep learning model to perform model training on the landslide identification dataset to obtain the landslide classifier model.

Images