A potential landslide identification method and system based on surface deformation cause analysis

Abstract

The application discloses a potential landslide identification method and system based on surface deformation cause analysis, and the method comprises the following steps: obtaining radar data and preprocessing to obtain surface deformation phase data; an IDANet-1 network is constructed and trained to predict a surface deformation abnormal area; the terrain DEM data, multi-temporal MODIS data and optical image data are processed respectively to obtain surface subsidence type, glacier snowmelt type and mine subsidence type surface deformation areas, and after set operation, the landslide type surface deformation target is obtained by difference set operation with the surface deformation abnormal area; the IDANet-2 network is trained by using the landslide type surface deformation identification data set; and wide-area comprehensive remote sensing data are input into the IDANet-2 network model for reasoning to obtain wide-area potential landslide hidden danger distribution information. Through the technical scheme, multi-level joint of multi-source multi-modal remote sensing data is realized, surface deformation information is accurately identified, and the identification efficiency of potential landslide hidden dangers in a large area is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing, 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

[0002] China has complex geological conditions and frequent geological structure activities. Landslides, as a major geological disaster, often cause significant casualties and economic losses. Due to the wide distribution, great harm and complex causes of landslides, it has been a focus and difficulty in preventing geological disasters. In recent years, with the continuous development of satellite remote sensing technology, the use of "space-air-ground" comprehensive remote sensing technology to carry out geological disaster investigation has achieved outstanding results. InSAR (Interferometric Synthetic Aperture Radar) technology has become one of the important means for landslide hazard identification and monitoring due to its high precision, wide range of monitoring capabilities and weather-independent characteristics.

[0003] However, this technology still faces two difficulties in the process of developing 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 the surface deformation information associated with landslides still needs further research.

[0004] Generally speaking, the cause types of surface deformation can be roughly summarized into four categories: mine subsidence type, surface subsidence type, collapse and landslide type, and glacier melting type. The collapse and landslide type of surface deformation has intersection with the other three types, so it is the most difficult to distinguish. In previous studies, the analysis usually focuses on a certain type of surface deformation, and the classification research of the four types of surface deformation is less. And the distinction of different types of surface deformation often involves the application of different remote sensing data, how to fuse multi-source remote sensing data to extract key features for surface deformation cause classification is also a difficulty. SUMMARY

[0005] To address the aforementioned issues, this invention provides a method and system for identifying potential landslides based on surface deformation causal analysis. Through a novel remote sensing intelligent identification application model employing multi-model hierarchical processing, it achieves multi-level synergy of multi-source, multi-modal remote sensing data, overcoming the limitations of traditional methods that are restricted to single-modal data. By utilizing two deep learning networks—one based on a large visual model and the other considering the complexity of multimodal remote sensing data—accurate identification of surface deformation information is achieved. This effectively avoids the problems of insufficient data information mining and inadequate feature extraction capabilities caused by differences in data computational complexity inherent in single encoder networks. Applicable to intelligent classification of surface deformation causal factors in wide-area scenarios, this invention effectively improves the efficiency of identifying potential landslide hazards in large areas, and is of great significance for enhancing the intelligent level of landslide disaster monitoring and early warning.

[0006] To achieve the above objectives, the present invention provides a method for identifying potential landslides based on surface deformation causal analysis, comprising:

[0007] The radar data of the preset demonstration area is acquired, preprocessed, and the surface deformation phase data is calculated.

[0008] An IDANet-1 network was constructed and trained, and the surface deformation phase data was input into the IDANet-1 network to predict and obtain the surface deformation anomaly area.

[0009] For the topographic DEM data, multi-temporal MODIS data and optical image data of the surface deformation anomaly area, slope and relief calculation, snow and ice coverage calculation and optical image interpretation analysis are performed respectively. The preset requirement area is intersected with the surface deformation anomaly area to obtain surface subsidence type surface deformation area, glacier melting type surface deformation area and mine collapse type surface deformation area.

[0010] After performing a union operation on the surface subsidence type surface deformation area, the glacier melting 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 the landslide type surface deformation target.

[0011] The data of the surface deformation anomaly area are resampled and channel-dimensionally superimposed to obtain comprehensive remote sensing data, which is then superimposed with the landslide-type surface deformation target to obtain a landslide-type surface deformation identification dataset.

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

[0013] In the technical scheme, preferably, the radar data of the preset demonstration area is acquired, preprocessed, and ground surface deformation phase data is calculated, and the specific process comprises:

[0014] Sentinel-1 radar data of the demonstration area is acquired, and sub-band data is spliced and registered.

[0015] A suitable image pair is selected by using a Stacking-InSAR multi-temporal analysis technology to perform differential interference calculation, phase unwrapping, phase stacking processing, and trend stripe removal, so as to obtain ground surface deformation phase data of the demonstration area.

[0016] In the technical scheme, preferably, the IDANet-1 network is constructed and trained, and the ground surface deformation phase data is input into the IDANet-1 network to obtain a ground surface deformation anomaly area, and the specific process comprises:

[0017] The IDANet-1 network is constructed by using an encoder and a decoder, wherein the encoder comprises a preset number of cross-arranged Trans modules and a dynamic adjustment layer, the dynamic adjustment layer comprises three multilayer perceptrons and two GeLU activation layers, and is used to migrate the model to a deformation phase identification scene, and the decoder comprises a Trans module and two multilayer perceptrons, and is used to restore the features to label data with a size consistent with the input image;

[0018] The SAM network parameters trained on the sa-1b data set are used to initialize all layers in the IDANet-1 network except the dynamic adjustment layer;

[0019] The ground surface deformation phase data and the true value label are used to train the IDANet-1 network, and during the training process, all layers except the dynamic adjustment layer are frozen, and only the dynamic adjustment layer is trained.

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

[0021] In the technical scheme, preferably, the terrain DEM data, the multi-temporal MODIS data, and the optical image data of the ground surface deformation anomaly area are respectively subjected to slope and relief degree calculation, ice and snow coverage rate calculation, and optical image interpretation analysis, and preset required areas are respectively intersected with the ground surface deformation anomaly area to obtain a ground surface subsidence type ground surface deformation area, a glacier snowmelt type ground surface deformation area, and a mine subsidence type ground surface deformation area, and the specific process comprises:

[0022] The slope and relief of the topographic DEM data of the surface deformation anomaly area are calculated, the area with a topographic relief lower than a preset value is intersected with the surface deformation anomaly area, and a surface subsidence type surface deformation area is obtained;

[0023] The ice and snow coverage rate in a preset time period is calculated for the multi-temporal MODIS data of the surface deformation anomaly area, the perennial ice and snow coverage area meeting a preset requirement is intersected with the surface deformation anomaly area, and a glacier snowmelt type surface deformation area is obtained;

[0024] Optical image interpretation analysis is performed on the optical image data of the surface deformation anomaly area, and a mine activity range is circled, the mine activity range is intersected with the surface deformation anomaly area, and a mine subsidence type surface deformation area is obtained.

[0025] In the above technical solution, preferably, the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine subsidence type surface deformation area are subjected to set operation, and then difference set operation is performed with the surface deformation anomaly area to obtain a landslide type surface deformation target, and the specific process includes:

[0026] The surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine subsidence type surface deformation area are subjected to set operation, and a non-landslide deformation target set of the demonstration area is obtained;

[0027] Difference set 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 a preset data are removed, and a landslide type surface deformation target of the demonstration area is obtained.

[0028] In the above technical solution, preferably, the data of the surface deformation anomaly area are subjected to resampling operation and channel dimension stacking to obtain comprehensive remote sensing data, and channel dimension stacking is performed with the landslide type surface deformation target to obtain a landslide type surface deformation recognition data set, and the specific process includes:

[0029] The topographic DEM data, multi-temporal MODIS data and optical image data of the demonstration area are subjected to resampling operation, and channel dimension stacking is performed to form comprehensive remote sensing data of the demonstration area with 9 channels;

[0030] The landslide type surface deformation target is subjected to mask processing as a reference true value, and channel dimension stacking is performed with the comprehensive remote sensing data of the demonstration area to obtain a data collection body with 10 channels;

[0031] Normalization and data cutting are performed on the data collection body to obtain the landslide type surface deformation recognition data set.

[0032] In the technical solution, preferably, the landslide-type ground deformation identification dataset is used to train the IDANet-2 network, and wide-area comprehensive remote sensing data are input into the trained IDANet-2 network model for inference to obtain wide-area potential landslide hazard distribution information, and the specific process includes:

[0033] The IDANet-2 network is constructed by 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 six DCB convolution modules, the DCB convolution module includes one 3×3 deep convolution layer, two 1×1 ordinary convolution layers and an activation layer, the decoder uses skip-layer connection to superimpose features from the encoder during the upsampling process from bottom to top, and the attention fusion module includes a multi-layer perception, a global pooling layer and different types of activation functions to respectively extract features for ground deformation, optical remote sensing, slope and relief and snow cover data.

[0034] The IDANet-2 network is trained by using the landslide-type ground deformation identification dataset.

[0035] The wide-area comprehensive remote sensing data are constructed by using data collected in an area outside the demonstration area.

[0036] The wide-area comprehensive remote sensing data are input into the trained IDANet-2 network for inference to obtain a wide-area landslide ground deformation area.

[0037] The wide-area landslide ground deformation area is comprehensively analyzed to obtain wide-area potential landslide hazard distribution information.

[0038] The application further provides a potential landslide identification system based on ground deformation cause analysis, characterized in that the potential landslide identification method based on ground deformation cause analysis is applied, and the method includes:

[0039] A deformation data acquisition module is configured to acquire radar data of a preset demonstration area for preprocessing and to calculate ground deformation phase data.

[0040] An abnormal data prediction module is configured to construct and train an IDANet-1 network, and to input the ground deformation phase data into the IDANet-1 network for prediction to obtain a ground deformation abnormal area.

[0041] a non-landslide type analysis module, configured to perform slope and relief degree calculation, ice and snow coverage rate calculation, and optical image interpretation analysis on terrain DEM data, multi-temporal MODIS data, and optical image data of the ground surface deformation anomaly area respectively, and perform intersection operation of a preset required area and the ground surface deformation anomaly area respectively to obtain a ground surface subsidence type ground surface deformation area, a glacier snowmelt type ground surface deformation area, and a mine collapse type ground surface deformation area;

[0042] a landslide deformation identification module, configured to perform set operation on the ground surface subsidence type ground surface deformation area, the glacier snowmelt type ground surface deformation area, and the mine collapse type ground surface deformation area, and perform difference set operation on the ground surface deformation anomaly area to obtain a landslide type ground surface deformation target;

[0043] a landslide data synthesis module, configured to perform resampling operation and channel dimension stacking on various data of the ground surface deformation anomaly area to obtain comprehensive remote sensing data, and perform channel dimension stacking on the comprehensive remote sensing data and the landslide type ground surface deformation target to obtain landslide type ground surface deformation identification data set;

[0044] a landslide hidden danger reasoning module, configured to train an IDANet-2 network by using the landslide type ground surface deformation identification data set, and input wide-area comprehensive remote sensing data into the trained IDANet-2 network model to perform reasoning, to obtain wide-area potential landslide hidden danger distribution information.

[0045] In the above technical solution, preferably, the non-landslide type analysis module is specifically configured to:

[0046] perform slope and relief degree calculation on terrain DEM data of the ground surface deformation anomaly area, and perform intersection operation on a region with terrain relief degree lower than a preset value and the ground surface deformation anomaly area to obtain a ground surface subsidence type ground surface deformation area;

[0047] perform ice and snow coverage rate calculation on multi-temporal MODIS data of the ground surface deformation anomaly area, and perform intersection operation on a perennial ice and snow coverage region meeting a preset requirement and the ground surface deformation anomaly area to obtain a glacier snowmelt type ground surface deformation area;

[0048] perform optical image interpretation analysis on optical image data of the ground surface deformation anomaly area, and circle a mine area activity range, and perform intersection operation on the mine area activity range and the ground surface deformation anomaly area to obtain a mine collapse type ground surface deformation area.

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

[0050] The ground subsidence type ground deformation area, the glacier snowmelt type ground deformation area and the mine subsidence type ground deformation area are subjected to a union operation to obtain a non-landslide deformation target set of the demonstration area;

[0051] The non-landslide deformation target set is subjected to a difference set operation with the ground deformation anomaly area of the demonstration area, and targets with a slope data lower than a preset data are removed from the result to obtain a landslide type ground deformation target of the demonstration area.

[0052] Compared with the prior art, the beneficial effects of the present application are: a new mode of remote sensing intelligent recognition is applied through multi-model hierarchical processing, multi-level joint of multi-source multi-modal remote sensing data is realized, and the problem that traditional methods are limited by single modal data is overcome; through two kinds of deep learning networks based on visual large model and considering the complexity of multi-modal remote sensing data, accurate recognition of ground deformation information is realized, and the problems of insufficient data information mining and insufficient feature extraction ability due to differences in data calculation complexity existing in single encoder network are effectively avoided, which is suitable for intelligent classification of ground deformation causes in wide-area scenarios, effectively improves the efficiency of potential landslide hazard identification in large areas, and has important significance for improving the intelligent level of landslide disaster monitoring and early warning. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 A flowchart of a potential landslide identification method based on ground deformation cause analysis according to an embodiment of the present application is disclosed;

[0054] Figure 2 A main framework and dynamic adjustment layer structure diagram of an IDANet-1 network according to an embodiment of the present application is disclosed;

[0055] Figure 3 A main framework diagram of an IDANet-1 network according to an embodiment of the present application is disclosed;

[0056] Figure 4 A module diagram of a potential landslide identification system based on ground deformation cause analysis according to an embodiment of the present application is disclosed.

[0057] In the figure, the correspondence between each component and the reference numeral is:

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

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] The present invention will now be described in further detail with reference to the accompanying drawings:

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

[0062] The radar data of the preset demonstration area is acquired, preprocessed, and the surface deformation phase data is calculated.

[0063] The IDANet-1 network was constructed and trained, and the surface deformation phase data was input into the IDANet-1 network to predict the surface deformation anomaly area.

[0064] For the topographic DEM data, multi-temporal MODIS data and optical image data of the surface deformation anomaly area, slope and relief calculation, snow and ice coverage calculation and optical image interpretation analysis are performed respectively. The preset requirement area is intersected with the surface deformation anomaly area to obtain the surface subsidence type surface deformation area, glacier melting type surface deformation area and mine collapse type surface deformation area.

[0065] After performing a union operation on the surface subsidence type surface deformation area, the glacier melting and 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 the landslide type surface deformation target.

[0066] By resampling and channel-dimensional overlaying of various data in the surface deformation anomaly area, comprehensive remote sensing data is obtained. This data is then overlaid with the landslide-type surface deformation target to obtain the landslide-type surface deformation identification dataset.

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

[0068] In the embodiment, a new mode is applied to remote sensing intelligent recognition through multi-model hierarchical processing, multi-level joint of multi-source multi-modal remote sensing data is realized, and the problem that traditional methods are limited by single modal data is overcome; through two kinds of deep learning networks based on visual large model and considering the complexity of multi-modal remote sensing data, accurate identification of ground surface deformation information is realized, and the problems of insufficient data information mining and insufficient feature extraction ability of single encoder network due to differences in data calculation complexity are effectively avoided, which is suitable for intelligent classification of ground surface deformation causes in wide-area scenarios, effectively improves the identification efficiency of potential landslide hazards in large areas, and has important significance for improving the intelligent level of landslide disaster monitoring and early warning.

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

[0070] Sentinel-1 radar data of the demonstration area is acquired, and sub-band data is spliced and registered;

[0071] The Stacking-InSAR multi-temporal analysis technology is used to select appropriate image pairs for differential interference calculation, phase unwrapping, phase stacking processing and trend stripe removal, and ground surface deformation phase data of the demonstration area is obtained.

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

[0073] The IDANet-1 network is constructed through an encoder and a decoder, wherein the encoder includes a plurality of 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, and is used to migrate the model to the deformation phase recognition scene, the decoder includes a Trans module and two multi-layer perceptrons, and is used to restore the features to label data with the same size as the input image;

[0074] Before training, the parameters of the Segment Anything (SAM) network trained on the sa-1b data set are used to initialize all layers in the IDANet-1 network except the dynamic adjustment layer;

[0075] The IDANet-1 network is trained by using surface deformation phase data and true value labels. During the training process, all layers except the dynamic adjustment layer are frozen, and the dynamic adjustment layer is trained only, so that the capability migration of the IDANet-1 network to the target scene is realized. The data required for model training can be surface deformation phase data and true value labels generated by artificial interpretation. The more extensive the data source used for training is, the richer the surface deformation type is, and the better the model performance is.

[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 are 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 embodiment, preferably, the terrain DEM data of the surface deformation anomaly area, the multi-temporal MODIS data and the optical image data are respectively subjected to slope and relief degree calculation, ice and snow coverage rate calculation and optical image interpretation analysis, and the preset required area is respectively intersected with the surface deformation anomaly area to obtain the surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine subsidence type surface deformation area. The specific process includes:

[0078] The terrain DEM data of the surface deformation anomaly area are subjected to slope and relief degree calculation. The area with a terrain relief degree lower than a preset value is intersected with the surface deformation anomaly area to obtain the surface subsidence type surface deformation area.

[0079] Specifically, ground subsidence induced surface deformation is a very common phenomenon. Ground subsidence usually occurs in plain areas due to underground water exploitation, geological climate change, engineering building load, etc., but only relying on elevation cannot accurately exclude this type of deformation. Since this type of deformation mainly exists in relatively flat areas, compared with other types, it has the characteristics of large spatial range and smaller spatial differentiation, so the terrain relief degree is used as a key distinguishing parameter to extract the surface subsidence type surface deformation area. The terrain relief degree is calculated by using DEM data. Let z(x, y) be the elevation value of the point (x, y) in the DEM data, N be the number of grid points in the window, and z mean The average value of the elevations of all points in the window is z, and the calculation formula of the terrain relief degree TRI is as follows:

[0080]

[0081] For the multi-temporal MODIS data of the surface deformation anomaly area, the ice and snow coverage rate in a preset time period (such as the past five years) is calculated. The perennial ice and snow coverage area meeting the preset requirements is intersected with the surface deformation anomaly area to obtain the glacier snowmelt type surface deformation area.

[0082] Specifically, the perennial snow and ice coverage area is extracted, and first of all, MODIS data of the demonstration area in the past five years needs to be collected, and the snow and ice index is calculated monthly, and the calculation formula is as follows:

[0083]

[0084] Wherein, NDSI ij represents the snow and ice index of each pixel. NIR and SWIR represent the near-infrared and short-wave infrared data of the MODIS data. Set k as the threshold value of the snow and ice index, when NDSI ij ≥k, it is considered that the pixel is covered by snow and ice, and the value is 1, otherwise the value is 0. K can be set to different values in different regions according to the actual situation, and the default value is 0.5.

[0085]

[0086] Assuming that T represents the total number of observations in five years (12x5), then the snow and ice coverage rate SIR ij is:

[0087]

[0088] Determine the area with a snow and ice coverage rate greater than 0.5, which is the perennial snow and ice coverage area.

[0089] The optical image data of the ground surface deformation anomaly area is analyzed by optical image interpretation, and the mining area activity range is circled. The intersection operation of the mining area activity range and the ground surface deformation anomaly area is performed to obtain the mine subsidence type ground surface deformation area.

[0090] Specifically, the Sentinel-2 true color optical image data of the demonstration area is obtained, and the stope, transfer field, waste and mine building areas related to mining in the demonstration area are circled by visual interpretation. The intersection operation of the above areas and the ground surface deformation anomaly area in the demonstration area is performed to obtain the mine subsidence type ground surface deformation area.

[0091] In the above embodiment, preferably, the ground surface subsidence type ground surface deformation area, the glacier snowmelt type ground surface deformation area and the mine subsidence type ground surface deformation area are set union operation, and the landslide type ground surface deformation target is obtained by difference set operation with the ground surface deformation anomaly area, and the specific process includes:

[0092] The ground surface subsidence type ground surface deformation area, the glacier snowmelt type ground surface deformation area and the mine subsidence type ground surface deformation area are set union operation to obtain the non-landslide deformation target set of the demonstration area;

[0093] The non-landslide deformation target set and the ground surface deformation anomaly area of the demonstration area are set difference set operation, and the targets with slope data lower than the preset data in the result are removed to obtain the landslide type ground surface deformation target of the demonstration area.

[0094] In this embodiment, the calculation method of the non-landslide deformation area is the union of the surface subsidence type, mine collapse type and glacial melting type surface deformation targets, and the difference set of the surface deformation anomaly area is the preliminary 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] Wherein, IDA a represents the set of surface subsidence type surface deformation targets, IDA b represents the set of mine collapse type surface deformation targets, IDA c represents the set of glacial melting type surface deformation targets, unIDA d represents the set of non-landslide deformation targets, nIDA d is the set of preliminary determined landslide deformation targets. ∪ represents the set operation, and – represents the difference set operation.

[0097] Finally, the slope data S of the demonstration area is used to eliminate the targets with low slope in the nIDA d set, and the landslide type surface deformation area IDA d of the demonstration area is obtained. It can be represented by the following formula:

[0098] IDA d = {f i ∈ nIDA d ∣s i ≥ θ}

[0099] Wherein, f i represents an element in the nIDA d set, S i represents the corresponding slope value, and θ is the slope threshold value, the default value is 5.

[0100] In the above embodiment, preferably, the data of the surface deformation anomaly area is resampled and channel dimension is superimposed to obtain comprehensive remote sensing data, and the landslide type surface deformation target is channel dimension superimposed to obtain landslide type surface deformation identification data set, and the specific process includes:

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

[0102] The landslide type ground surface deformation target is masked and treated as a reference true value, and is overlaid with channel dimension on the comprehensive remote sensing data of the demonstration area to obtain a data aggregate with 10 channels;

[0103] The data aggregate is normalized and data cropped to obtain a landslide type ground surface deformation recognition data set.

[0104] Specifically, the landslide type ground surface deformation area is masked, and the specific method is to assign the pixels in the landslide type ground surface deformation area as 1 and the rest of the area as 0, thereby forming a mask file.

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

[0106] X' = (x-x 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 cropping the data with 10 channels into data slices with a fixed size (256x256 or 512x512). The slice size is usually between 256 and 1024, and the specific setting needs to consider the size of the computer video memory and the size of the model parameters. There are two kinds of slice methods, sequential and random, and the latter is recommended.

[0109] In the above embodiment, preferably, the IDANet-2 network is trained using the landslide type ground surface deformation recognition data set, and the wide-area comprehensive remote sensing data is input into the trained IDANet-2 network model for inference to obtain wide-area potential landslide hazard distribution information, and the specific process includes:

[0110] The IDANet-2 network is constructed by using an encoder, a decoder and an attention fusion module, wherein the encoder comprises four groups of DCBB convolution modules, each group of DCBB convolution module comprises six DCB convolution modules, the DCB convolution module comprises one 3*3 deep convolution layer, two 1*1 ordinary convolution layers and an activation layer, the decoder is designed with reference to UNet, and in the upsampling process from bottom to top, the features from the encoder are stacked by using the jump layer connection, the features after the output of each branch are connected, and the CBR module (convolution layer, normalization layer and activation layer combination) is connected. The attention fusion (CATT) module uses the attention mechanism to perform weighted processing on the effective features, including a multilayer perceptron, a global pooling layer and different types of activation functions, so as to respectively extract features for the ground surface deformation, optical remote sensing, slope and relief degree and ice and snow coverage rate data, and effectively avoid the problems of insufficient data information mining and insufficient feature extraction capacity caused by the difference in data calculation complexity of a single encoder network.

[0111] The IDANet-2 network is trained by using a landslide type ground surface deformation identification data set;

[0112] Based on the same formation mode as the comprehensive remote sensing data of the demonstration area, the data collected in the area outside the demonstration area are used to construct wide-area comprehensive remote sensing data;

[0113] The wide-area comprehensive remote sensing data are input into the trained IDANet-2 network for inference, and a wide-area landslide ground surface deformation area is obtained;

[0114] The wide-area landslide ground surface deformation area is comprehensively analyzed to obtain wide-area potential landslide hazard distribution information.

[0115] As shown in Figure 4 The present application further provides a potential landslide identification system based on ground surface deformation cause analysis, characterized in that the potential landslide identification method based on ground surface deformation cause analysis is applied, and the method comprises:

[0116] A deformation data acquisition module 1 is used for acquiring radar data of a preset demonstration area for preprocessing and calculating ground surface deformation phase data;

[0117] An abnormal data prediction module 2 is used for constructing and training an IDANet-1 network, and inputting the ground surface deformation phase data into the IDANet-1 network for prediction to obtain a ground surface deformation abnormal area;

[0118] The non-landslide type analysis module 3 is configured to perform slope and relief degree calculation, ice and snow coverage rate calculation and optical image interpretation analysis on the terrain DEM data, multi-temporal MODIS data and optical image data of the ground surface deformation anomaly area respectively, and perform intersection operation of the preset required area and the ground surface deformation anomaly area respectively, to obtain a ground surface subsidence type ground surface deformation area, a glacier snowmelt type ground surface deformation area and a mine subsidence type ground surface deformation area.

[0119] The landslide deformation identification module 4 is configured to perform set operation on the ground surface subsidence type ground surface deformation area, the glacier snowmelt type ground surface deformation area and the mine subsidence type ground surface deformation area, and perform difference set operation on the ground surface deformation anomaly area to obtain a landslide type ground surface deformation target.

[0120] The landslide data synthesis module 5 is configured to perform resampling operation and channel dimension stacking on each data of the ground surface deformation anomaly area to obtain comprehensive remote sensing data, and perform channel dimension stacking on the landslide type ground surface deformation target to obtain landslide type ground surface deformation identification data set.

[0121] The landslide hidden danger reasoning module 6 is configured to train the IDANet-2 network by using the landslide type ground surface deformation identification data set, and input the wide-area comprehensive remote sensing data into the trained IDANet-2 network model to perform reasoning, to obtain wide-area potential landslide hidden danger distribution information.

[0122] In the above embodiment, preferably, the non-landslide type analysis module 3 is specifically configured to:

[0123] perform slope and relief degree calculation on the terrain DEM data of the ground surface deformation anomaly area, and perform intersection operation on the area with terrain relief degree lower than a preset value and the ground surface deformation anomaly area, to obtain a ground surface subsidence type ground surface deformation area;

[0124] calculate the ice and snow coverage rate in a preset time period for the multi-temporal MODIS data of the ground surface deformation anomaly area, and perform intersection operation on the perennial ice and snow coverage area meeting a preset requirement and the ground surface deformation anomaly area, to obtain a glacier snowmelt type ground surface deformation area;

[0125] perform optical image interpretation analysis on the optical image data of the ground surface deformation anomaly area, to delineate a mine area activity range, and perform intersection operation on the mine area activity range and the ground surface deformation anomaly area, to obtain a mine subsidence type ground surface deformation area.

[0126] In the above embodiment, preferably, the landslide deformation identification module 4 is specifically configured to:

[0127] perform set operation on the ground surface subsidence type ground surface deformation area, the glacier snowmelt type ground surface deformation area and the mine subsidence type ground surface deformation area, to obtain a non-landslide deformation target set of the demonstration area;

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

[0129] According to the potential landslide identification system based on surface deformation cause analysis disclosed in the above embodiments, the functions to be realized by each module correspond to the steps of the potential landslide identification method based on surface deformation cause analysis disclosed in the above embodiments respectively, and in the implementation process, the above embodiments are referred to for operation, and thus will not be described here again.

[0130] The above only is the preferred embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A potential landslide identification method based on surface deformation cause analysis, characterized by, The method comprises the following steps: acquiring radar data of a preset demonstration area for preprocessing and calculating ground surface deformation phase data; constructing and training an IDANet-1 network, and inputting the ground surface deformation phase data into the IDANet-1 network for prediction to obtain a ground surface deformation anomaly area; calculating slope and relief for topographic DEM data, multi-temporal MODIS data and optical image data of the ground surface deformation anomaly area, calculating ice and snow coverage, and performing optical image interpretation analysis, and performing intersection operation of preset requirement areas and the ground surface deformation anomaly area to obtain a ground surface deformation area of ground subsidence type, a ground surface deformation area of glacier snowmelt type and a ground surface deformation area of mine subsidence type; performing set operation of the ground surface deformation area of ground subsidence type, the ground surface deformation area of glacier snowmelt type and the ground surface deformation area of mine subsidence type, and performing difference set operation of the ground surface deformation anomaly area to obtain a landslide type ground surface deformation target; performing resampling operation and channel dimension stacking for each data of the ground surface deformation anomaly area to obtain comprehensive remote sensing data, and performing channel dimension stacking of the landslide type ground surface deformation target to obtain a landslide type ground surface deformation identification data set; training an IDANet-2 network by using the landslide type ground surface deformation identification data set, and inputting wide-area comprehensive remote sensing data into the trained IDANet-2 network model for inference to obtain wide-area potential landslide hazard distribution information; the process of constructing and training the IDANet-1 network and inputting the ground surface deformation phase data into the IDANet-1 network for prediction to obtain the ground surface deformation anomaly area comprises: constructing the IDANet-1 network through an encoder and a decoder, wherein the encoder comprises a preset number of cross-arranged Trans modules and a dynamic adjustment layer, the dynamic adjustment layer comprises three multilayer perceptrons and two GeLU activation layers, and is used for migrating the model to a deformation phase identification scene, and the decoder comprises a Trans module and two multilayer perceptrons, and is used for restoring features to label data with the same size as input images; initializing all layers in the IDANet-1 network except the dynamic adjustment layer by using SAM network parameters trained on a sa-1b data set; training the IDANet-1 network by using ground surface deformation phase data and true value labels, and in the training process, freezing all layers except the dynamic adjustment layer, and only training the dynamic adjustment layer; inputting the ground surface deformation phase data of the demonstration area into the trained IDANet-1 network for prediction to obtain the ground surface deformation anomaly area of the demonstration area.

2. The potential landslide identification method based on land surface deformation cause analysis according to claim 1, characterized in that, the process of acquiring radar data of a preset demonstration area for preprocessing and calculating ground surface deformation phase data comprises: acquiring Sentinel-1 radar data of the demonstration area, and splicing and registering substrip data; The surface deformation phase data of the demonstration area is obtained by using a stacking-InSAR multi-temporal analysis technology, selecting appropriate image pairs for differential interference calculation, phase unwrapping, phase stacking processing and trend stripe removal. 3.The potential landslide identification method based on land surface deformation cause analysis according to claim 1, characterized in that, The slope and relief of the DEM data of the surface deformation anomaly area are calculated, the ice and snow coverage rate is calculated, and the optical image interpretation analysis is performed on the optical image data of the surface deformation anomaly area, and the preset required area is intersected with the surface deformation anomaly 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, and the specific process includes: The slope and relief of the DEM data of the surface deformation anomaly area are calculated, the area with a terrain relief lower than a preset value is intersected with the surface deformation anomaly area to obtain the surface subsidence type surface deformation area; The ice and snow coverage rate in a preset time period is calculated based on the multi-temporal MODIS data of the surface deformation anomaly area, and the perennial ice and snow coverage area meeting the preset requirements is intersected with the surface deformation anomaly area to obtain the glacier snowmelt type surface deformation area; The optical image interpretation analysis is performed on the optical image data of the surface deformation anomaly area, and the mine activity range is circled, and the mine activity range is intersected with the surface deformation anomaly area to obtain the mine collapse type surface deformation area.

4. The potential landslide identification method based on land surface deformation cause analysis according to claim 3, characterized in that, The surface subsidence type surface deformation area, the glacier snowmelt type surface deformation area and the mine collapse type surface deformation area are unioned to obtain a non-landslide deformation target set of the demonstration area, and the non-landslide deformation target set is difference-set operated with the surface deformation anomaly area of the demonstration area to obtain a landslide type surface deformation target of the demonstration area. The slope data of the result lower than a preset data is removed to obtain the landslide type surface deformation target of the demonstration area. The data of the surface deformation anomaly area is resampled and channel dimension stacked to obtain comprehensive remote sensing data, and the landslide type surface deformation target is channel dimension stacked with the landslide type surface deformation target to obtain a landslide type surface deformation recognition data set, and the specific process includes: 5.The potential landslide identification method based on land surface deformation cause analysis according to claim 4, characterized in that, The DEM data, multi-temporal MODIS data and optical image data of the surface deformation anomaly area are resampled and channel dimension stacked to obtain comprehensive remote sensing data of the demonstration area with nine channels; The landslide type surface deformation target is used as a reference true value for mask processing, and the landslide type surface deformation target is channel dimension stacked with the comprehensive remote sensing data of the demonstration area to obtain a data collection body with ten channels; The data collection body is normalized and data cropped to obtain the landslide type surface deformation recognition data set. ​ 6.The potential landslide identification method based on land surface deformation cause analysis according to claim 5, wherein, The IDANet-2 network is trained by using the landslide-type ground surface deformation identification data set, and wide-area comprehensive remote sensing data are input into the trained IDANet-2 network model for inference to obtain wide-area potential landslide hazard distribution information, and the specific process comprises the following steps: The IDANet-2 network is constructed by using an encoder, a decoder and an attention fusion module, wherein the encoder comprises four groups of DCBB convolution modules, each group of the DCBB convolution modules comprises six DCB convolution modules, the DCB convolution module comprises one 3*3 deep convolution layer, two 1*1 ordinary convolution layers and an activation layer, the decoder adopts skip-layer connection to stack features from the encoder during the upsampling process from bottom to top, and the attention fusion module comprises a multi-layer perception, a global pooling layer and different types of activation functions to extract features for ground surface deformation, optical remote sensing, slope and relief and snow cover rate data respectively; The IDANet-2 network is trained by using the landslide-type ground surface deformation identification data set; The wide-area comprehensive remote sensing data are obtained by using the data collected in the area outside the demonstration area; The wide-area landslide ground surface deformation area is obtained by inputting the wide-area comprehensive remote sensing data into the trained IDANet-2 network for inference; The wide-area potential landslide hazard distribution information is obtained by comprehensively analyzing the wide-area landslide ground surface deformation area. 7.A potential landslide identification system based on surface deformation cause analysis, characterized by, The potential landslide identification method based on ground surface deformation cause analysis is applied, and the method comprises the following steps: A deformation data acquisition module is configured to acquire radar data of a preset demonstration area for preprocessing and to calculate ground surface deformation phase data; An abnormal data prediction module is configured to construct and train an IDANet-1 network, and to input the ground surface deformation phase data into the IDANet-1 network for prediction to obtain a ground surface deformation abnormal area; A non-landslide analysis module is configured to perform slope and relief calculation, snow cover rate calculation and optical image interpretation analysis on terrain DEM data, multi-temporal MODIS data and optical image data of the ground surface deformation abnormal area respectively, and to perform intersection operation on a preset required area and the ground surface deformation abnormal area respectively to obtain a ground surface subsidence-type ground surface deformation area, a glacier snowmelt-type ground surface deformation area and a mine subsidence-type ground surface deformation area; A landslide deformation identification module is configured to perform set operation on the ground surface subsidence-type ground surface deformation area, the glacier snowmelt-type ground surface deformation area and the mine subsidence-type ground surface deformation area, and to perform difference set operation on the ground surface deformation abnormal area to obtain a landslide-type ground surface deformation target; A landslide data comprehensive module is configured to perform resampling operation and channel dimension stacking on various data of the ground surface deformation abnormal area to obtain comprehensive remote sensing data, and to perform channel dimension stacking on the landslide-type ground surface deformation target to obtain a landslide-type ground surface deformation identification data set. The landslide hazard reasoning module is configured to train an IDANet-2 network by using the landslide-type ground surface deformation identification dataset, and input wide-area comprehensive remote sensing data into the trained IDANet-2 network model to perform reasoning, so as to obtain wide-area potential landslide hazard distribution information.

8. The potential landslide identification system based on land surface deformation cause analysis according to claim 7, characterized in that, The non-landslide-type analysis module is specifically configured to: perform slope and relief degree calculation on terrain DEM data of the ground surface deformation anomaly area, perform intersection operation on a region with a terrain relief degree lower than a preset value and the ground surface deformation anomaly area, and obtain a ground surface subsidence-type ground surface deformation region; perform ice and snow coverage calculation on multi-temporal MODIS data of the ground surface deformation anomaly area in a preset time period, perform intersection operation on a perennial ice and snow coverage region meeting a preset requirement and the ground surface deformation anomaly area, and obtain a glacier snowmelt-type ground surface deformation region; perform optical image interpretation analysis on optical image data of the ground surface deformation anomaly area, delineate a mining area activity range, perform intersection operation on the mining area activity range and the ground surface deformation anomaly area, and obtain a mine collapse-type ground surface deformation region.

9. The potential landslide identification system based on land surface deformation cause analysis according to claim 8, characterized in that, The landslide deformation identification module is specifically configured to: perform set union operation on the ground surface subsidence-type ground surface deformation region, the glacier snowmelt-type ground surface deformation region and the mine collapse-type ground surface deformation region, and obtain a non-landslide deformation target set of the demonstration area; perform set difference operation on the non-landslide deformation target set and the ground surface deformation anomaly area of the demonstration area, and remove a target with a slope data lower than a preset data in the result, and obtain a landslide-type ground surface deformation target of the demonstration area.

Images