Potential landslide identification method and system based on surface deformation cause analysis

Abstract

The invention discloses a potential landslide identification method and system based on surface deformation cause analysis. The method comprises the following steps: acquiring radar data and preprocessing the radar data to obtain surface deformation phase data; an IDANet-1 network is constructed and trained, and an earth surface deformation abnormal area is obtained through prediction; respectively processing the terrain DEM data, the multi-temporal MODIS data and the optical image data to obtain ground surface settlement type, glacier snow melting type and mine collapse type ground surface deformation areas, performing union set operation, and performing difference set operation on the union set operation and a ground surface deformation abnormal area to obtain a landslide type ground surface deformation target; training an IDANet-2 network by using the landslide type surface deformation identification data set; and inputting the wide-area comprehensive remote sensing data into the IDANet-2 network model for reasoning to obtain wide-area potential landslide hidden danger distribution information. Through the technical scheme of the invention, multi-level combination of multi-source and multi-mode remote sensing data is realized, accurate identification of earth surface deformation information is realized, and the identification efficiency of large-area potential landslide hidden dangers is effectively improved.

Description

Technical Field

[0001] The invention relates to the field of remote sensing technology, and in particular to a potential landslide identification method based on surface deformation cause analysis and a potential landslide identification system based on surface deformation cause analysis. Background Art

[0002] my country has complex geological conditions and frequent geological tectonic activities. As a major geological disaster, landslides often cause heavy casualties and economic and property losses. Landslides have always been the focus and difficulty of geological disaster prevention due to their wide distribution, great harm, and complex causes. In recent years, with the continuous development of satellite remote sensing technology, the use of "air-ground" integrated remote sensing technology to carry out geological disaster hazard investigations has achieved outstanding results. InSAR (Synthetic Aperture Radar Interferometry) technology has become one of the important means of identifying and monitoring landslide hazards due to its high precision, large-scale monitoring capabilities, and its characteristics of not being restricted by weather conditions.

[0003] However, this technology still faces two difficulties in the process of wide-area identification applications: first, the extraction of useful surface deformation information from data with a large amount of noise and interference information still mainly relies on manual visual interpretation, which is difficult to meet the needs of wide-area identification; second, the causes of surface deformation information are complex, and how to effectively distinguish surface deformation information related to landslides still needs further research.

[0004] Generally speaking, the causes of surface deformation can be roughly summarized into four categories: mine collapse, surface subsidence, collapse and landslide, and glacier snowmelt. Among the four types, collapse and landslide surface deformation overlaps with the other three types, so it is the most difficult to distinguish. In previous studies, the focus was usually on analyzing surface deformation of a certain type of cause, and there were few studies on the classification of the four types of surface deformation at the same time. In addition, the distinction between different types of surface deformation often involves the application of different remote sensing data. How to fuse multi-source remote sensing data and extract key features for the classification of surface deformation causes is also a difficult point. Summary of the invention

[0005] In view of the above problems, the present invention provides a method and system for identifying potential landslides based on the analysis of the causes of surface deformation. Through a new remote sensing intelligent recognition application mode of multi-model hierarchical processing, multi-level combination of multi-source and multi-modal remote sensing data is realized, overcoming the problem that traditional methods are limited by single-modal data. Through two deep learning networks based on large visual models and taking into account the complexity of multi-modal remote sensing data, accurate recognition of surface deformation information is achieved, effectively avoiding the problems of insufficient data information mining and insufficient feature extraction capabilities caused by differences in data calculation complexity in a single encoder network. It is suitable for intelligent classification of surface deformation causes in wide-area scenarios, effectively improving the efficiency of identifying potential landslide hazards in large areas, and is of great significance for improving the intelligent level of landslide disaster monitoring and early warning.

[0006] To achieve the above object, the present invention provides a potential landslide identification method based on surface deformation cause analysis, comprising:

[0007] Obtain radar data of a preset demonstration area for preprocessing, and calculate the surface deformation phase data;

[0008] Constructing and training an IDANet-1 network, and inputting the surface deformation phase data into the IDANet-1 network for prediction to obtain a surface deformation anomaly area;

[0009] The terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation and analysis respectively, and the preset required area is intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area;

[0010] After performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, a difference operation is performed with the surface deformation anomaly area to obtain a landslide type surface deformation target;

[0011] Resampling and channel-dimensionally superimposing the data of the abnormal surface deformation area to obtain comprehensive remote sensing data, and superimposing the data with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set;

[0012] The landslide-type surface deformation identification dataset is used to train the IDANet-2 network, and the wide-area integrated remote sensing data is input into the trained IDANet-2 network model for reasoning to obtain the wide-area potential landslide hazard distribution information.

[0013] In the above technical solution, preferably, the radar data of the preset demonstration area is obtained and preprocessed, and the surface deformation phase data is calculated, and the specific process includes:

[0014] Acquire Sentinel-1 radar data of the demonstration area and stitch and register sub-strip data;

[0015] By using Stacking-InSAR multi-temporal analysis technology, appropriate image pairs were selected for differential interferometry calculation, phase unwrapping, phase stacking processing and trend fringes removal to obtain the surface deformation phase data of the demonstration area.

[0016] In the above technical solution, preferably, the construction and training of the IDANet-1 network, and inputting the surface deformation phase data into the IDANet-1 network for prediction to obtain the surface deformation anomaly area, specifically includes:

[0017] An IDANet-1 network is constructed through an encoder and a decoder, wherein the encoder includes a preset number of cross-arranged Trans modules and a dynamic adjustment layer, the dynamic adjustment layer includes three multi-layer perceptrons and two GeLU activation layers, which are used to migrate the model to the deformation phase recognition scenario, and the decoder includes a Trans module and two multi-layer perceptrons, which are used to restore the features to label data consistent with the size of the input image;

[0018] Initialize all layers except the dynamic adjustment layer in the IDANet-1 network using the SAM network parameters trained on the sa-1b data set;

[0019] The IDANet-1 network is trained using surface deformation phase data and true value labels. During the training process, all layers except the dynamic adjustment layer are frozen, and only the dynamic adjustment layer is trained.

[0020] The surface deformation phase data of the demonstration area is input into the trained IDANet-1 network for prediction to obtain the surface deformation anomaly area of ​​the demonstration area.

[0021] In the above technical solution, preferably, the terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, calculate the ice and snow coverage rate and interpret the optical image, and the preset required area is intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area. The specific process includes:

[0022] Calculate the slope and relief of the terrain DEM data of the abnormal surface deformation area, intersect the area where the terrain relief is lower than a preset value with the abnormal surface deformation area to obtain a surface subsidence type surface deformation area;

[0023] The multi-phase MODIS data of the abnormal surface deformation area are used to calculate the ice and snow coverage rate within a preset time period, and the perennial ice and snow covered area that meets the preset requirements is intersected with the abnormal surface deformation area to obtain the glacier snowmelt type surface deformation area;

[0024] Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range, and the mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

[0025] In the above technical solution, preferably, after performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, a difference operation is performed with the surface deformation anomaly area to obtain a landslide type surface deformation target, and the specific process includes:

[0026] Performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area to obtain a non-landslide deformation target set of the demonstration area;

[0027] A difference operation is performed between the non-landslide deformation target set and the surface deformation anomaly area of ​​the demonstration area, and targets with slope data lower than preset data in the result are eliminated to obtain the landslide surface deformation targets of the demonstration area.

[0028] In the above technical solution, preferably, the data of the abnormal surface deformation area are resampled and channel-dimensionally superimposed to obtain comprehensive remote sensing data, and the data are channel-dimensionally superimposed with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set, and the specific process includes:

[0029] The terrain DEM data, multi-temporal MODIS data, optical image data, slope data, terrain relief data and ice and snow coverage data of the demonstration area are resampled and channel-dimensionally superimposed to form comprehensive remote sensing data of the demonstration area with 9 channels;

[0030] The landslide-type surface deformation target is used as a reference true value for mask processing, and is channel-dimensionally superimposed with the comprehensive remote sensing data of the demonstration area to obtain a data set with 10 channels;

[0031] The data set is normalized and data cut to obtain the landslide-type surface deformation identification data set.

[0032] In the above technical solution, preferably, the IDANet-2 network is trained using the landslide-type surface deformation identification dataset, and the wide-area integrated remote sensing data is input into the trained IDANet-2 network model for reasoning to obtain the wide-area potential landslide hazard distribution information, and the specific process includes:

[0033] The IDANet-2 network is constructed using an encoder, a decoder and an attention fusion module, wherein the encoder includes four groups of DCBB convolution modules, each group of the DCBB convolution modules includes 6 DCB convolution modules, and the DCB convolution module includes a 3×3 deep convolution layer, two 1×1 ordinary convolution layers and an activation layer. The decoder uses skip-layer connections to superimpose features from the encoder during bottom-up upsampling. The attention fusion module includes a multi-layer perceptron, a global pooling layer and different types of activation functions to extract features from surface deformation, optical remote sensing, slope and undulation, and ice and snow coverage data respectively;

[0034] Using the landslide type surface deformation recognition data set to train the IDANet-2 network;

[0035] Wide-area integrated remote sensing data were constructed using data collected outside the demonstration area;

[0036] Inputting the wide-area integrated remote sensing data into the trained IDANet-2 network for inference to obtain a wide-area landslide surface deformation area;

[0037] A comprehensive analysis is conducted on the wide-area landslide surface deformation area to obtain the wide-area potential landslide hazard distribution information.

[0038] The present invention also proposes a potential landslide identification system based on surface deformation cause analysis, characterized in that the potential landslide identification method based on surface deformation cause analysis as described in any one of claims 1 to 7 is applied, comprising:

[0039] The deformation data acquisition module is used to obtain radar data of a preset demonstration area for preprocessing and calculate the surface deformation phase data;

[0040] An abnormal data prediction module is used to construct and train an IDANet-1 network, and input the surface deformation phase data into the IDANet-1 network to predict the surface deformation abnormal area;

[0041] The non-landslide analysis module is used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation analysis for the terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area, and intersect the preset required area with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area;

[0042] A landslide deformation identification module is used to perform a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, and then perform a difference operation with the surface deformation anomaly area to obtain a landslide type surface deformation target;

[0043] A landslide data synthesis module is used to perform resampling operations and channel-dimensional superposition on various data of the surface deformation anomaly area to obtain comprehensive remote sensing data, and perform channel-dimensional superposition with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set;

[0044] The landslide hazard inference module is used to train the IDANet-2 network using the landslide type surface deformation identification data set, and input the wide-area comprehensive remote sensing data into the trained IDANet-2 network model for inference to obtain the wide-area potential landslide hazard distribution information.

[0045] In the above technical solution, preferably, the non-landslide analysis module is specifically used for:

[0046] Calculate the slope and relief of the terrain DEM data of the abnormal surface deformation area, intersect the area where the terrain relief is lower than a preset value with the abnormal surface deformation area to obtain a surface subsidence type surface deformation area;

[0047] The multi-phase MODIS data of the abnormal surface deformation area are used to calculate the ice and snow coverage rate within a preset time period, and the perennial ice and snow covered area that meets the preset requirements is intersected with the abnormal surface deformation area to obtain the glacier snowmelt type surface deformation area;

[0048] Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range, and the mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

[0049] In the above technical solution, preferably, the landslide deformation identification module is specifically used for:

[0050] Performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area to obtain a non-landslide deformation target set of the demonstration area;

[0051] A difference operation is performed between the non-landslide deformation target set and the surface deformation anomaly area of ​​the demonstration area, and targets with slope data lower than preset data in the result are eliminated to obtain the landslide surface deformation targets of the demonstration area.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows: through a new remote sensing intelligent recognition application mode of multi-model hierarchical processing, multi-level combination of multi-source and multi-modal remote sensing data is realized, overcoming the problem that traditional methods are limited to single-modal data; through two deep learning networks based on large visual models and taking into account the complexity of multi-modal remote sensing data, accurate recognition of surface deformation information is achieved, effectively avoiding the problems of insufficient data information mining and insufficient feature extraction capabilities caused by differences in data calculation complexity in a single encoder network, and being suitable for intelligent classification of surface deformation causes in wide-area scenarios, effectively improving the efficiency of identifying potential landslide hazards in large areas, and having important significance for improving the intelligent level of landslide disaster monitoring and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic flow chart of a potential landslide identification method based on surface deformation cause analysis disclosed in an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of the main framework and dynamic adjustment layer structure of the IDANet-1 network disclosed in an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of the main framework of the IDANet-1 network disclosed in an embodiment of the present invention;

[0056] Figure 4 The module schematic diagram of a potential landslide identification system based on surface deformation cause analysis disclosed in an embodiment of the present invention.

[0057] In the figure, the corresponding relationship between each component and the reference numeral is as follows:

[0058] 1. Deformation data acquisition module, 2. Abnormal data prediction module, 3. Non-landslide analysis module, 4. Landslide deformation identification module, 5. Landslide data synthesis module, 6. Landslide hazard reasoning module. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0061] like Figure 1 As shown, a potential landslide identification method based on surface deformation cause analysis provided by the present invention includes:

[0062] Obtain radar data of a preset demonstration area for preprocessing, and calculate the surface deformation phase data;

[0063] Construct and train the IDANet-1 network, and input the surface deformation phase data into the IDANet-1 network to predict the surface deformation anomaly area;

[0064] The terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation and analysis respectively. The preset required areas are intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area.

[0065] After performing a union operation on the surface deformation areas of surface subsidence, glacier snowmelt and mine collapse, a difference operation is performed with the surface deformation anomaly area to obtain the landslide surface deformation target.

[0066] The data of the abnormal surface deformation area are resampled and channel-dimensionally superimposed to obtain comprehensive remote sensing data, and then channel-dimensionally superimposed with the landslide-type surface deformation target to obtain the landslide-type surface deformation identification data set;

[0067] The IDANet-2 network was trained using the landslide-type surface deformation recognition dataset, and the wide-area comprehensive remote sensing data was input into the trained IDANet-2 network model for inference to obtain the wide-area potential landslide hazard distribution information.

[0068] In this implementation, a new mode of remote sensing intelligent recognition application based on multi-model hierarchical processing is used to achieve multi-level integration of multi-source and multi-modal remote sensing data, overcoming the problem that traditional methods are limited by single-modal data; through two deep learning networks based on large visual models and taking into account the complexity of multi-modal remote sensing data, accurate recognition of surface deformation information is achieved, effectively avoiding the problems of insufficient data information mining and insufficient feature extraction capabilities caused by differences in data calculation complexity in a single encoder network, and is suitable for intelligent classification of surface deformation causes in wide-area scenarios, effectively improving the efficiency of identifying potential landslide hazards in large areas, and is of great significance for improving the intelligent level of landslide disaster monitoring and early warning.

[0069] In the above implementation, preferably, radar data of a preset demonstration area is obtained for preprocessing, and surface deformation phase data is calculated, and the specific process includes:

[0070] Acquire Sentinel-1 radar data of the demonstration area and stitch and register the sub-strip data;

[0071] Using Stacking-InSAR multi-temporal analysis technology, appropriate image pairs were selected for differential interferometry calculation, phase unwrapping, phase stacking processing and trend fringes removal to obtain the surface deformation phase data of the demonstration area.

[0072] like Figure 2 and Figure 3 As shown, in the above implementation, preferably, an IDANet-1 network is constructed and trained, and the surface deformation phase data is input into the IDANet-1 network for prediction to obtain the surface deformation anomaly area. The specific process includes:

[0073] The IDANet-1 network is constructed through an encoder and a decoder. The encoder includes multiple cross-arranged Vision Transformer (Trans) modules and a dynamic adjustment layer. The dynamic adjustment layer includes three multi-layer perceptrons and two Gaussian Error Linear Unit (GeLU) activation layers, which are used to migrate the model to the deformation phase recognition scenario. The decoder includes a Trans module and two multi-layer perceptrons, which are used to restore the features to label data consistent with the size of the input image.

[0074] Before training, all layers except the dynamic adjustment layer in the IDANet-1 network are initialized using the parameters of the Segment Anything (SAM) network trained on the sa-1b dataset;

[0075] The IDANet-1 network is trained using surface deformation phase data and true value labels. During the training process, all layers except the dynamic adjustment layer are frozen. Only by training the dynamic adjustment layer, its ability to migrate to the target scene can be achieved. The data required for model training can be the surface deformation phase data and the true value labels generated by manual interpretation. The wider the data source used for training and the richer the surface deformation types, the better the model performance.

[0076] The trained model is applied to the surface deformation phase data of the demonstration area. The surface deformation phase data of the demonstration area is input into the trained IDANet-1 network for prediction, and the surface deformation anomaly area of ​​the demonstration area is obtained.

[0077] In the above implementation, preferably, the terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation analysis, and the preset required area is intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area. The specific process includes:

[0078] The slope and relief are calculated for the terrain DEM data of the abnormal surface deformation area, and the area with terrain relief lower than the preset value is intersected with the abnormal surface deformation area to obtain the surface deformation area of ​​surface subsidence type.

[0079] Specifically, surface deformation caused by ground subsidence is a very common phenomenon. Ground subsidence usually occurs in plain areas due to groundwater exploitation, geological and climate changes, engineering and construction loads, etc., but this type of deformation cannot be accurately eliminated by elevation alone. Since this type of deformation mainly exists in relatively flat areas, compared with other types, it is characterized by a large spatial range and smaller spatial heterogeneity. Therefore, the terrain undulation is used as the key distinguishing parameter to extract the surface deformation area of ​​surface subsidence. Use DEM data to calculate the terrain undulation. Let z(x,y) be the elevation value of the midpoint (x,y)(x,y) of the DEM data, N be the number of grid points in the window, and z mean is the average elevation of all points in the window, then the calculation formula of terrain relief TRI is as follows:

[0080]

[0081] Based on the multi-phase MODIS data of the surface deformation anomaly area, the ice and snow coverage rate within the preset time period (for example, the past five years) is calculated, and the perennial ice and snow covered areas that meet the preset requirements are intersected with the surface deformation anomaly area to obtain the glacier snowmelt type surface deformation area.

[0082] Specifically, to extract the perennial snow-covered area, it is necessary to collect MODIS data of the demonstration area in the past five years and calculate the snow and ice index on a monthly basis. The calculation formula is as follows:

[0083]

[0084] Among them, NDSI ij Represents the snow and ice index of each pixel. NIR and SWIR represent the near infrared and short wave infrared data of MODIS data respectively. Set k as the threshold of snow and ice index. When NDSI ij When ≥ k, the pixel is considered to be covered by ice and snow and is assigned a value of 1, otherwise it is assigned a value of 0. K can be set to different values ​​in different regions according to actual conditions, and the default value is 0.5.

[0085]

[0086] Assuming T represents the total number of observations in five years (12×5), the snow cover ratio SIR ij for:

[0087]

[0088] Areas with an ice and snow coverage rate greater than 0.5 are considered to be areas covered with ice and snow all year round.

[0089] Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range. The mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

[0090] Specifically, the Sentinel-2 true-color optical image data of the demonstration area were obtained, and the mining-related areas such as the stopes, transfer yards, waste and mine buildings in the demonstration area were delineated through visual interpretation. The above areas were intersected with the abnormal surface deformation areas in the demonstration area to obtain the mine subsidence-type surface deformation areas.

[0091] In the above implementation, preferably, after performing a union operation on the surface deformation area of ​​surface settlement type, the surface deformation area of ​​glacier snowmelt type and the surface deformation area of ​​mine collapse type, a difference operation is performed with the surface deformation anomaly area to obtain the landslide type surface deformation target, and the specific process includes:

[0092] The surface deformation areas of surface subsidence, glacier snowmelt and mine collapse are combined to obtain the non-landslide deformation target set of the demonstration area.

[0093] The non-landslide deformation target set is subtracted from the surface deformation anomaly area in the demonstration area, and the targets with slope data lower than the preset data are eliminated to obtain the landslide surface deformation targets in the demonstration area.

[0094] In this embodiment, the calculation method of the non-landslide deformation area is the combination of surface settlement type, mine collapse type and glacier snowmelt type surface deformation targets, and the difference between it and the surface deformation abnormal area is the initially determined landslide deformation target. The calculation process is as follows:

[0095] nIDA d =IDA e –unIDA d =IDA e –(IDA a ∪IDA b ∪IDA c )

[0096] Among them, IDA a Represents a set of surface deformation targets of surface subsidence type, IDA b Represents a collection of mine subsidence-type ground deformation targets, IDA c represents a set of glacier-melt surface deformation targets, unIDA d represents the set of non-landslide deformation targets, nIDA d is the set of landslide deformation targets initially determined. ∪ represents the union operation, and – represents the difference operation.

[0097] Finally, the slope data S of the demonstration area was used to analyze nIDA d The targets with low slope in the set are eliminated to obtain the landslide-type surface deformation area IDA of the demonstration area. d . It can be expressed as the following formula:

[0098] IDA d ={f i ∈nIDA d ∣s i ≥θ}

[0099] Among them, f i Represents nIDA d An element in the set, S i Represents the corresponding slope value, θ is the slope threshold, and the default value is 5.

[0100] In the above implementation, preferably, the data of the abnormal surface deformation area are resampled and channel-dimensionally superimposed to obtain comprehensive remote sensing data, and the data are channel-dimensionally superimposed with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set. The specific process includes:

[0101] The terrain DEM data, multi-temporal MODIS data, optical image data, slope data, terrain relief data and ice and snow coverage data of the demonstration area were resampled and superimposed in the channel dimension to form a comprehensive remote sensing data of the demonstration area with 9 channels;

[0102] The landslide-type surface deformation target is used as the reference truth value for mask processing, and the channel dimension is superimposed with the comprehensive remote sensing data of the demonstration area to obtain a data set with 10 channels.

[0103] The data set is normalized and trimmed to obtain the landslide-type surface deformation identification data set.

[0104] Specifically, the landslide type surface deformation area is masked by assigning pixels in the landslide type surface deformation area to 1 and assigning pixels in other areas to 0, thereby forming a mask file.

[0105] Data normalization refers to normalizing each channel of the 9-channel demonstration area comprehensive remote sensing data according to the following formula so that its value range is between 0 and 1.

[0106] X'=(xx min ) / (x max -x min )

[0107] Where X' represents the normalized data, x min and x max Represent the minimum and maximum values ​​of the data respectively.

[0108] Data cropping refers to cutting 10-channel data into data slices of fixed size (256×256 or 512×512). The slice size is usually between 256-1024. The specific setting needs to consider the size of the computer video memory and the number of model parameters. There are two slicing methods: sequential and random. The latter is recommended.

[0109] In the above implementation, preferably, the IDANet-2 network is trained using the landslide type surface deformation identification data set, and the wide-area integrated remote sensing data is input into the trained IDANet-2 network model for reasoning to obtain the wide-area potential landslide hazard distribution information. The specific process includes:

[0110] The IDANet-2 network is constructed using encoder, decoder and attention fusion module. The encoder includes four groups of DCBB convolution modules, each of which includes 6 DCB convolution modules. The DCB convolution module includes 1 3×3 deep convolution layer, 2 1×1 ordinary convolution layers and activation layer. The decoder refers to the design of UNet. In the process of bottom-up upsampling, skip-layer connection is used to superimpose the features from the encoder. The features output by each branch are connected to the CBR module (convolution layer, normalization layer and activation layer combination). The attention fusion (CATT) module uses the attention mechanism to weight the effective features, including multi-layer perceptron, global pooling layer and different types of activation functions, to extract features for surface deformation, optical remote sensing, slope and undulation, and ice and snow coverage data, effectively avoiding the problems of insufficient data information mining and insufficient feature extraction capability caused by differences in data calculation complexity in a single encoder network.

[0111] The IDANet-2 network is trained using the landslide-type surface deformation recognition dataset;

[0112] Based on the same formation method as the comprehensive remote sensing data of the demonstration area, the wide-area comprehensive remote sensing data are constructed using the data collected outside the demonstration area;

[0113] The wide-area integrated remote sensing data is input into the trained IDANet-2 network for inference to obtain the wide-area landslide surface deformation area;

[0114] A comprehensive analysis of the wide-area landslide surface deformation area is conducted to obtain the wide-area potential landslide hazard distribution information.

[0115] like Figure 4 As shown, the present invention also proposes a potential landslide identification system based on surface deformation cause analysis, characterized in that the potential landslide identification method based on surface deformation cause analysis according to any one of claims 1 to 7 is applied, comprising:

[0116] The deformation data acquisition module 1 is used to obtain radar data of a preset demonstration area for preprocessing and calculate the surface deformation phase data;

[0117] Abnormal data prediction module 2 is used to build and train the IDANet-1 network, and input the surface deformation phase data into the IDANet-1 network to predict the surface deformation abnormal area;

[0118] Non-landslide analysis module 3 is used to calculate the slope and undulation, ice and snow coverage and optical image interpretation and analysis for the terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area, and intersect the preset required area with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area;

[0119] The landslide deformation identification module 4 is used to perform a union operation on the surface deformation area of ​​the surface settlement type, the surface deformation area of ​​the glacier snowmelt type and the surface deformation area of ​​the mine collapse type, and then perform a difference operation with the surface deformation anomaly area to obtain the landslide type surface deformation target;

[0120] The landslide data synthesis module 5 is used to perform resampling operations and channel dimension superposition on various data of the surface deformation abnormal area to obtain comprehensive remote sensing data, and perform channel dimension superposition with the landslide type surface deformation target to obtain a landslide type surface deformation identification data set;

[0121] The landslide hazard inference module 6 is used to train the IDANet-2 network using the landslide type surface deformation identification data set, and input the wide-area comprehensive remote sensing data into the trained IDANet-2 network model for inference to obtain the wide-area potential landslide hazard distribution information.

[0122] In the above implementation, preferably, the non-landslide analysis module 3 is specifically used for:

[0123] The slope and relief are calculated for the terrain DEM data of the abnormal surface deformation area, and the area with terrain relief lower than the preset value is intersected with the abnormal surface deformation area to obtain the surface deformation area of ​​surface settlement type;

[0124] Based on the multi-temporal MODIS data of the surface deformation anomaly area, the ice and snow coverage rate in the preset time period is calculated, and the perennial ice and snow covered area that meets the preset requirements is intersected with the surface deformation anomaly area to obtain the glacier snowmelt type surface deformation area;

[0125] Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range. The mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

[0126] In the above implementation, preferably, the landslide deformation identification module 4 is specifically used for:

[0127] The surface deformation areas of surface subsidence, glacier snowmelt and mine collapse are combined to obtain the non-landslide deformation target set of the demonstration area.

[0128] The non-landslide deformation target set is subtracted from the surface deformation anomaly area in the demonstration area, and the targets with slope data lower than the preset data are eliminated to obtain the landslide surface deformation targets in the demonstration area.

[0129] According to the potential landslide identification system based on surface deformation causal analysis disclosed in the above-mentioned embodiment, the functions to be implemented by each module thereof correspond to the steps of the potential landslide identification method based on surface deformation causal analysis disclosed in the above-mentioned embodiment. During the implementation process, the operations are performed with reference to the above-mentioned embodiment, which will not be repeated here.

[0130] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A potential landslide identification method based on surface deformation cause analysis, characterized in that: include: The radar data of the preset demonstration area is obtained for preprocessing, and the surface deformation phase data is calculated; Constructing and training an IDANet-1 network, and inputting the surface deformation phase data into the IDANet-1 network for prediction to obtain a surface deformation anomaly area; The terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation and analysis respectively, and the preset required area is intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area; After performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, a difference operation is performed with the surface deformation anomaly area to obtain a landslide type surface deformation target; Resampling and channel-dimensionally superimposing the data of the abnormal surface deformation area to obtain comprehensive remote sensing data, and superimposing the data with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set; The landslide-type surface deformation identification dataset is used to train the IDANet-2 network, and the wide-area integrated remote sensing data is input into the trained IDANet-2 network model for reasoning to obtain the wide-area potential landslide hazard distribution information.

2. The potential landslide identification method based on surface deformation cause analysis according to claim 1 is characterized in that: The radar data of the preset demonstration area is obtained for preprocessing, and the surface deformation phase data is calculated. The specific process includes: Acquire Sentinel-1 radar data of the demonstration area and stitch and register sub-strip data; By using Stacking-InSAR multi-temporal analysis technology, appropriate image pairs were selected for differential interferometry calculation, phase unwrapping, phase stacking processing and trend fringes removal to obtain the surface deformation phase data of the demonstration area.

3. The potential landslide identification method based on surface deformation cause analysis according to claim 2 is characterized in that: The IDANet-1 network is constructed and trained, and the surface deformation phase data is input into the IDANet-1 network for prediction to obtain the surface deformation anomaly area. The specific process includes: An IDANet-1 network is constructed through an encoder and a decoder, wherein the encoder includes a preset number of cross-arranged Trans modules and a dynamic adjustment layer, the dynamic adjustment layer includes three multi-layer perceptrons and two GeLU activation layers, which are used to migrate the model to the deformation phase recognition scenario, and the decoder includes a Trans module and two multi-layer perceptrons, which are used to restore the features to label data consistent with the size of the input image; Initialize all layers except the dynamic adjustment layer in the IDANet-1 network using the SAM network parameters trained on the sa-1b data set; The IDANet-1 network is trained using surface deformation phase data and true value labels. During the training process, all layers except the dynamic adjustment layer are frozen, and only the dynamic adjustment layer is trained; The surface deformation phase data of the demonstration area is input into the trained IDANet-1 network for prediction to obtain the surface deformation anomaly area of ​​the demonstration area.

4. The potential landslide identification method based on surface deformation cause analysis according to claim 3 is characterized in that: The terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area are used to calculate the slope and undulation, calculate the ice and snow coverage rate and interpret the optical image, and the preset required area is intersected with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area. The specific process includes: Calculate the slope and relief of the terrain DEM data of the abnormal surface deformation area, intersect the area where the terrain relief is lower than a preset value with the abnormal surface deformation area to obtain a surface subsidence type surface deformation area; The multi-phase MODIS data of the abnormal surface deformation area are used to calculate the ice and snow coverage rate within a preset time period, and the perennial ice and snow covered area that meets the preset requirements is intersected with the abnormal surface deformation area to obtain the glacier snowmelt type surface deformation area; Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range, and the mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

5. The potential landslide identification method based on surface deformation cause analysis according to claim 4 is characterized in that: After performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, a difference operation is performed with the surface deformation anomaly area to obtain a landslide type surface deformation target. The specific process includes: Performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area to obtain a non-landslide deformation target set of the demonstration area; A difference operation is performed between the non-landslide deformation target set and the surface deformation anomaly area of ​​the demonstration area, and targets with slope data lower than preset data in the result are eliminated to obtain the landslide surface deformation targets of the demonstration area.

6. The potential landslide identification method based on surface deformation cause analysis according to claim 5 is characterized in that: The data of the abnormal surface deformation area are resampled and channel-dimensionally superimposed to obtain comprehensive remote sensing data, and the landslide-type surface deformation target is channel-dimensionally superimposed to obtain a landslide-type surface deformation identification data set. The specific process includes: The terrain DEM data, multi-temporal MODIS data, optical image data, slope data, terrain relief data and ice and snow coverage data of the demonstration area are resampled and channel-dimensionally superimposed to form comprehensive remote sensing data of the demonstration area with 9 channels; The landslide-type surface deformation target is used as a reference true value for masking, and is channel-dimensionally superimposed with the comprehensive remote sensing data of the demonstration area to obtain a data set with 10 channels; The data set is normalized and data cut to obtain the landslide type surface deformation identification data set.

7. The potential landslide identification method based on surface deformation cause analysis according to claim 6 is characterized in that: The IDANet-2 network is trained by using the landslide type surface deformation identification data set, and the wide-area integrated remote sensing data is input into the trained IDANet-2 network model for reasoning to obtain the wide-area potential landslide hazard distribution information. The specific process includes: The IDANet-2 network is constructed using an encoder, a decoder and an attention fusion module, wherein the encoder includes four groups of DCBB convolution modules, each group of the DCBB convolution modules includes 6 DCB convolution modules, and the DCB convolution module includes a 3×3 deep convolution layer, two 1×1 ordinary convolution layers and an activation layer. The decoder uses skip-layer connections to superimpose features from the encoder during bottom-up upsampling. The attention fusion module includes a multi-layer perceptron, a global pooling layer and different types of activation functions to extract features from surface deformation, optical remote sensing, slope and undulation, and ice and snow coverage data respectively; Using the landslide type surface deformation recognition data set to train the IDANet-2 network; Wide-area integrated remote sensing data were constructed using data collected outside the demonstration area; Inputting the wide-area integrated remote sensing data into the trained IDANet-2 network for inference to obtain a wide-area landslide surface deformation area; A comprehensive analysis is conducted on the wide-area landslide surface deformation area to obtain the wide-area potential landslide hazard distribution information.

8. A potential landslide identification system based on surface deformation cause analysis, characterized in that: The potential landslide identification method based on the cause analysis of surface deformation according to any one of claims 1 to 7 is applied, comprising: The deformation data acquisition module is used to obtain radar data of a preset demonstration area for preprocessing and calculate the surface deformation phase data; An abnormal data prediction module is used to construct and train an IDANet-1 network, and input the surface deformation phase data into the IDANet-1 network to predict the surface deformation abnormal area; The non-landslide analysis module is used to calculate the slope and undulation, the ice and snow coverage rate and the optical image interpretation analysis for the terrain DEM data, multi-temporal MODIS data and optical image data of the abnormal surface deformation area, and intersect the preset required area with the abnormal surface deformation area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area; A landslide deformation identification module is used to perform a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area, and then perform a difference operation with the surface deformation anomaly area to obtain a landslide type surface deformation target; A landslide data synthesis module is used to perform resampling operations and channel-dimensional superposition on various data of the surface deformation anomaly area to obtain comprehensive remote sensing data, and perform channel-dimensional superposition with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification data set; The landslide hazard inference module is used to train the IDANet-2 network using the landslide type surface deformation identification data set, and input the wide-area comprehensive remote sensing data into the trained IDANet-2 network model for inference to obtain the wide-area potential landslide hazard distribution information.

9. The potential landslide identification system based on surface deformation cause analysis according to claim 8 is characterized in that: The non-landslide analysis module is specifically used for: Calculate the slope and relief of the terrain DEM data of the abnormal surface deformation area, intersect the area where the terrain relief is lower than a preset value with the abnormal surface deformation area to obtain a surface subsidence type surface deformation area; The multi-phase MODIS data of the abnormal surface deformation area are used to calculate the ice and snow coverage rate within a preset time period, and the perennial ice and snow covered area that meets the preset requirements is intersected with the abnormal surface deformation area to obtain the glacier snowmelt type surface deformation area; Optical image interpretation and analysis are performed on the optical image data of the abnormal surface deformation area to delineate the mining area activity range, and the mining area activity range is intersected with the abnormal surface deformation area to obtain the mine subsidence type surface deformation area.

10. The potential landslide identification system based on surface deformation cause analysis according to claim 9 is characterized in that: The landslide deformation identification module is specifically used for: Performing a union operation on the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area to obtain a non-landslide deformation target set of the demonstration area; A difference operation is performed between the non-landslide deformation target set and the surface deformation anomaly area of ​​the demonstration area, and targets with slope data lower than preset data in the result are eliminated to obtain the landslide surface deformation targets of the demonstration area.

Images