A method, device, system and storage medium for predicting susceptibility to land subsidence

Through the dual-branch parallel fusion neural network combined with CNN and LSTM modules to extract and fusion spatial and sequence features, the problem that the existing LSSA method fails to effectively consider the sequence characteristics of environmental factors, and improves the reliability and accuracy of ground settlement susceptibility prediction sex.

CN118913211BActive Publication Date: 2024-12-03LANZHOU JIAOTONG UNIV
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202411415101.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-11
Publication Date
2024-12-03
Estimated Expiration
2044-10-11

AI Technical Summary

Technical Problem

The existing ground settlement sensitivity assessment (LSSA) machine learning method fails to effectively consider sequence characteristics of environmental factors during feature learning, resulting in insufficient and imbalanced ground settlement samples, reducing the reliability of predictions.

Method used

Two-branch parallel fusion neural network is adopted to extract two-dimensional spatial neighborhood features through the CNN module, and one-dimensional pixel sequence features are extracted, and combined in the fusion module. The surface settlement sensitivity probability is output using the Concatenate structure and the fully connected layer. value to improve the accuracy of prediction.

Benefits of technology

It improves the reliability of predicting ground settlement susceptibility, reduces the error in predicting high-prone areas, and enhances the ability to identify ground settlement risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118913211B_ABST
    Figure CN118913211B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device, a system and a storage medium for predicting the susceptibility of ground subsidence, belonging to the technical field of geological disaster prediction, and comprising: step S1, obtaining historical InSAR observation data; step S2, preprocessing the historical InSAR observation data; step S3, training a double-branch parallel fusion neural network according to the preprocessed historical InSAR observation data; step S4, inputting the InSAR observation data to be processed into the trained double-branch parallel fusion neural network for predicting the susceptibility of ground subsidence. The technical solution of the present invention is adopted to improve the reliability of LSSM.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster prediction, and in particular relates to a method and device, a system and a storage medium for predicting the susceptibility of ground subsidence. Background Art

[0002] Land subsidence is a geological disaster caused by natural and human factors. LS has regional, long-term and different characteristics. Land subsidence is generally not obvious in the early stage and mostly occurs in large and medium-sized cities, affecting local urban safety and sustainable economic and social development, and seriously affecting the structural health of urban infrastructure. Land subsidence sensitivity assessment (LSSA) can be used to indicate the probability of land subsidence and is one of the important factors in identifying land subsidence risks. LSSA can provide a basis for monitoring network planning, construction and differentiated prevention and control measures. Conducting LSSA is of great significance for early warning and preventive measures for damage to urban infrastructure.

[0003] At present, LSSA machine learning methods include: random forest (RF), support vector machine (SVM), artificial neural network (ANN) and convolutional neural network. These machine learning methods can accurately identify the implicit relationship between environmental factors and land subsidence in small scene tasks and achieve good results. However, in the process of feature learning, these machine learning methods only consider the spatial characteristics of environmental factors, but not the sequence characteristics after the superposition of environmental factors, and the obtained land subsidence samples are insufficient and unbalanced, resulting in low reliability of LSSM. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an InSAR ground subsidence susceptibility prediction method and device, system and storage medium to improve the reliability of LSSM.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for predicting the susceptibility of land subsidence, comprising:

[0007] Step S1, obtaining historical InSAR observation data;

[0008] Step S2, preprocessing the historical InSAR observation data;

[0009] Step S3, training a dual-branch parallel fusion neural network according to the preprocessed historical InSAR observation data;

[0010] Step S4, inputting the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence;

[0011] Step S2 includes:

[0012] Based on the historical InSAR observation data, the land subsidence adjustment factor (LSCF) is obtained; the land subsidence adjustment factor (LSCF) includes: altitude, slope, slope path, curvature, profile curvature, lithology, faults, precipitation, river density, groundwater, land use and land cover and building density;

[0013] According to the land subsidence adjustment factor (LSCF), a spatial feature data set and a pixel sequence feature data set are obtained; wherein the spatial feature data set and the pixel sequence feature data set are used to train a dual-branch parallel fusion neural network;

[0014] The dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features according to a spatial feature data set; the LSTM module is used to extract one-dimensional pixel sequence features according to a pixel sequence feature data set; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

[0015] As a preference, a Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module, and the two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, which are

[0016] The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into a one-dimensional vector with a dimension of 64 respectively; after feature fusion using the following formula, the fused feature The dimension size is 128; after integrating the spatial features and sequence features, a two-layer full connection is used, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity.

[0017]

[0018] in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal, used to operate on each feature;

[0019] The preprocessed historical InSAR observation data are input into a dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

[0020] The present invention also provides a device for predicting the susceptibility of land subsidence, comprising:

[0021] Acquisition module, used to obtain historical InSAR observation data;

[0022] Preprocessing module, used to preprocess historical InSAR observation data;

[0023] A training module, used to train a dual-branch parallel fusion neural network based on preprocessed historical InSAR observation data;

[0024] The prediction module is used to input the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence;

[0025] Preprocessing modules include:

[0026] The first processing unit is used to obtain a land subsidence adjustment factor (LSCF) according to historical InSAR observation data; wherein the land subsidence adjustment factor (LSCF) includes: altitude, slope, slope path, curvature, profile curvature, lithology, fault, precipitation, river density, groundwater, land use and land cover and building density;

[0027] The second processing unit is used to obtain a spatial feature data set and a pixel sequence feature data set according to a land subsidence adjustment factor (LSCF); wherein the spatial feature data set and the pixel sequence feature data set are used to train a dual-branch parallel fusion neural network;

[0028] The dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features according to a spatial feature data set; the LSTM module is used to extract one-dimensional pixel sequence features according to a pixel sequence feature data set; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

[0029] As a preference, a Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module, and the two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, which are

[0030] The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into a one-dimensional vector with a dimension of 64 respectively; after feature fusion using the following formula, the fused feature The dimension size is 128; after integrating the spatial features and sequence features, a two-layer full connection is used, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity.

[0031]

[0032] in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal, used to operate on each feature;

[0033] The preprocessed historical InSAR observation data are input into a dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

[0034] The present invention also provides a ground subsidence susceptibility prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a ground subsidence susceptibility prediction method when executed by the processor.

[0035] The present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the ground subsidence susceptibility prediction method when running.

[0036] The present invention trains a dual-branch parallel fusion neural network according to historical InSAR observation data; inputs the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence; adopts the technical scheme of the present invention, the error of predicting the area with high susceptibility is small, and the reliability of LSSM can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for predicting the susceptibility of land subsidence according to an embodiment of the present invention;

[0039] Figure 2 It is a schematic diagram of the structure of the fusion module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Embodiment 1:

[0043] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the susceptibility of land subsidence, comprising:

[0044] Step S1, obtaining historical InSAR observation data;

[0045] Step S2, preprocessing the historical InSAR observation data;

[0046] Step S3, training a dual-branch parallel fusion neural network according to the preprocessed historical InSAR observation data;

[0047] Step S4: input the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence.

[0048] As an implementation of the embodiment of the present invention, step S2 includes:

[0049] Based on the historical InSAR observation data, the land subsidence adjustment factor (LSCF) was obtained; among them, the land subsidence adjustment factor (LSCF) includes altitude, slope, slope path, curvature, profile curvature, lithology, faults, precipitation, river density, groundwater, land use and land cover and building density; further, the altitude was obtained from 30m resolution elevation data (DEM); the slope, slope path, curvature and profile curvature were calculated based on DEM data using QGIS open source software; lithology and faults came from public 1:2 million resolution vector geological data; precipitation was obtained through visual interpretation of Sentinel-2 satellite open source remote sensing images; river density was calculated using the line density tool of QGIS software; groundwater data, land use and land cover data were from the public environmental monitoring database; building density was calculated using the building index tool using Sentinel-2 satellite data.

[0050] According to the land subsidence adjustment factor (LSCF), a spatial feature dataset and a pixel sequence feature dataset are obtained; wherein, the spatial feature dataset and the pixel sequence feature dataset are used to train a dual-branch parallel fusion neural network.

[0051] Furthermore, the spatial feature dataset is obtained according to the land subsidence adjustment factor (LSCF). Specifically, the land subsidence adjustment factor is first superimposed according to the channel dimension to form a grating factor database with a dimension size of (2986, 1942, 12). Then, the sliding cutting method is used to cut the row and column directions using a 13×13 sliding window. Finally, the spatial features are extracted from the generated 228979 data with a dimension of (13, 13, 12).

[0052] Furthermore, the pixel sequence feature dataset was obtained according to the land subsidence adjustment factor (LSCF). Specifically, the LSCFs were superimposed according to their importance to form a sequence dataset of size (12, 2986, 1942). Secondly, the dimension of the sequence dataset was rearranged to meet the input requirements of model training. Finally, 228979 data of dimension size (12, 1) were generated, and the pixel sequence features were extracted according to a 1×1 window.

[0053] As an implementation method of an embodiment of the present invention, a dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features; the LSTM module is used to extract one-dimensional pixel sequence features; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

[0054] Furthermore, the CNN module includes: 3 convolutional layers, 3 maximum pooling layers, 3 dropout layers and 2 fully connected layers connected in sequence, and extracts 2D spatial neighborhood features according to the spatial feature data set. The input spatial feature data set is a 3D tensor, which represents the length, width and number of channels. The number of channels in the 3D tensor is 12. Nonlinear activation function is introduced in the convolution calculation of the convolution layer. , enabling CNN to learn and represent more complex functional relationships between factors that affect ground subsidence. The size of the convolutional layer is 3×3, and the sliding calculation step of the patch is 1 to ensure that features with more ground subsidence regulating factors are extracted from the convolutional layer. The maximum pooling layer reduces the spatial dimension of the input data by selecting the maximum value in the pooling window. The size of the pooling window is 2×2; this helps to reduce computational complexity, extract key features, reduce the model's sensitivity to location, and enhance the robustness of the model. Through the processing of three convolutional layers and pooling layers, the features of neighboring pixels can be obtained. The Dropout layer is used to reduce the activity of some neurons, so that the data set does not participate in the entire processing process and improves the applicability of the model prediction. The initial learning rate is 0.001 and the number of neural units is 128. The fully connected layer is used to integrate all the features of the previous layer of the network, and the sigmoid activation function is used in the output layer. The abstract features extracted by the convolutional layer are mapped to the class space and the two-dimensional spatial neighborhood features are output.

[0055] Furthermore, the LSTM module includes: three layers of superimposed LSTM units, which extract one-dimensional pixel sequence features based on the pixel sequence feature data set. Each LSTM unit includes: input gate, forget gate, and output gate; the input gate is used to determine the current pixel sequence feature data set. How much information should be stored in the memory cell, which uses the current and the hidden state of the previous moment To calculate; In addition, the input gate generates a value between 0 and 1 through the sigmoid function σ to determine how much current Should be stored in memory location Middle; Forget Gate Used to control the information in the memory cell at the previous moment The impact on the current moment, and with the help of the σ function, determine how much information is retained in the memory unit of the previous step; the output gate For operation through and the previously hidden state to determine how much information to extract from the memory cell; combine with the tanh function to generate a new memory cell value, and multiply the new memory cell value by the element to get the hidden form , hidden form The final output is obtained through the sigmoid function , according to the three-layer output The one-dimensional pixel sequence features are obtained by superimposing them on each other.

[0056] Furthermore, a Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module. Assuming that the two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into a one-dimensional vector with a dimension of 64 respectively; after feature fusion using the following formula, the fused feature The dimension size is 128; after fusing the spatial features and sequence features, a two-layer full connection is adopted, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity.

[0057]

[0058] in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal is used to operate on each feature.

[0059] Therefore, the preprocessed historical InSAR observation data are input into the dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

[0060] Embodiment 2:

[0061] The embodiment of the present invention further provides a device for predicting the susceptibility of land subsidence, comprising:

[0062] Acquisition module, used to obtain historical InSAR observation data;

[0063] Preprocessing module, used to preprocess historical InSAR observation data;

[0064] A training module, used to train a dual-branch parallel fusion neural network based on preprocessed historical InSAR observation data;

[0065] The prediction module is used to input the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence;

[0066] Preprocessing modules include:

[0067] The first processing unit is used to obtain a land subsidence adjustment factor (LSCF) according to historical InSAR observation data; wherein the land subsidence adjustment factor (LSCF) includes altitude, slope, slope path, curvature, profile curvature, lithology, fault, precipitation, river density, groundwater, land use and land cover and building density;

[0068] The second processing unit is used to obtain a spatial feature data set and a pixel sequence feature data set according to a land subsidence adjustment factor (LSCF); wherein the spatial feature data set and the pixel sequence feature data set are used to train a dual-branch parallel fusion neural network;

[0069] The dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features according to a spatial feature data set; the LSTM module is used to extract one-dimensional pixel sequence features according to a pixel sequence feature data set; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

[0070] As an implementation method of an embodiment of the present invention, a Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module, and the two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, which are respectively

[0071] The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into a one-dimensional vector with a dimension of 64 respectively; after feature fusion using the following formula, the fused feature The dimension size is 128; after integrating the spatial features and sequence features, a two-layer full connection is used, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity.

[0072]

[0073] in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal, used to operate on each feature;

[0074] The preprocessed historical InSAR observation data are input into a dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

[0075] Embodiment 3:

[0076] An embodiment of the present invention further provides a ground subsidence susceptibility prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a ground subsidence susceptibility prediction method when executed by the processor.

[0077] Embodiment 4:

[0078] An embodiment of the present invention further provides a storage medium having a computer program stored thereon, and the computer program executes a method for predicting ground subsidence susceptibility when running.

[0079] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the susceptibility of land subsidence, characterized in that: include: Step S1, obtaining historical InSAR observation data; Step S2, preprocessing the historical InSAR observation data; Step S3, training a dual-branch parallel fusion neural network according to the preprocessed historical InSAR observation data; Step S4, inputting the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence; Step S2 includes: Based on the historical InSAR observation data, the land subsidence adjustment factors are obtained; the land subsidence adjustment factors include altitude, slope, slope path, curvature, profile curvature, lithology, faults, precipitation, river density, groundwater, land use and land cover and building density; According to the land subsidence adjustment factor, a spatial feature data set and a pixel sequence feature data set are obtained; wherein the spatial feature data set and the pixel sequence feature data set are used to train a dual-branch parallel fusion neural network; The dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features according to a spatial feature data set; the LSTM module is used to extract one-dimensional pixel sequence features according to a pixel sequence feature data set; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

2. The method for predicting land subsidence susceptibility according to claim 1, characterized in that: A Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module. The two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, respectively. The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into one-dimensional vectors with a dimension of 64 respectively; After feature fusion using the following formula, the fusion feature The dimension size is 128; after integrating the spatial features and sequence features, a two-layer full connection is used, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity. in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal, used to operate on each feature; The preprocessed historical InSAR observation data are input into a dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

3. A device for predicting the susceptibility of land subsidence, characterized in that: include: Acquisition module, used to obtain historical InSAR observation data; Preprocessing module, used to preprocess historical InSAR observation data; A training module, used to train a dual-branch parallel fusion neural network based on preprocessed historical InSAR observation data; The prediction module is used to input the InSAR observation data to be processed into the trained dual-branch parallel fusion neural network to predict the susceptibility of ground subsidence; Preprocessing modules include: The first processing unit is used to obtain a land subsidence adjustment factor based on historical InSAR observation data; wherein the land subsidence adjustment factor includes altitude, slope, slope path, curvature, profile curvature, lithology, fault, precipitation, river density, groundwater, land use and land cover and building density; The second processing unit is used to obtain a spatial feature data set and a pixel sequence feature data set according to the ground subsidence adjustment factor; wherein the spatial feature data set and the pixel sequence feature data set are used to train a dual-branch parallel fusion neural network; The dual-branch parallel fusion neural network includes: a CNN module, an LSTM module and a fusion module; wherein the CNN module is used to extract two-dimensional spatial neighborhood features according to a spatial feature data set; the LSTM module is used to extract one-dimensional pixel sequence features according to a pixel sequence feature data set; and the fusion module is used to fuse two-dimensional spatial neighborhood features and one-dimensional pixel sequence features.

4. The device for predicting the susceptibility of land subsidence according to claim 3, characterized in that: A Concatenate structure is embedded in the process of merging the two-dimensional spatial neighborhood features and the one-dimensional pixel sequence features in the fusion module. The two feature channels extracted in parallel by the CNN and LSTM models have the same number of channels, respectively. The calculation process of the Concatenate structure is as follows: the spatial features extracted by the CNN model and the pixel sequence features extracted by the LSTM are mapped into one-dimensional vectors with a dimension of 64 respectively; After feature fusion using the following formula, the fusion feature The dimension size is 128; after integrating the spatial features and sequence features, a two-layer full connection is used, and the ReLU and Sigmoid activation functions are used to output the probability value of surface subsidence sensitivity. in, is the number of features in each channel; The i-th channel feature extracted by the CNN module; For Multiplication weights; The i-th channel feature extracted for the LSTM module; For Multiplication weights; is the index variable, from 1 to Traversal, used to operate on each feature; The preprocessed historical InSAR observation data are input into a dual-branch parallel fusion neural network, and feature extraction and fusion are performed through the CNN module, LSTM module and fusion module to obtain the surface subsidence sensitivity probability value based on the historical data to complete the training process of the dual-branch parallel fusion neural network.

5. A land subsidence susceptibility prediction system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the method for predicting the susceptibility of ground subsidence as described in any one of claims 1 to 2 is executed.

6. A storage medium, characterized in that: The storage medium stores a computer program, which, when running, executes the method for predicting the susceptibility of land subsidence as described in any one of claims 1 to 2.

Citation Information

Patent Citations

  • Multi-model fusion shield construction settlement prediction method and system

    CN113204824A

  • Pipe network geological landslide disaster early warning method based on multiple modes

    CN115563572A

  • Ground subsidence monitoring method

    CN115638767A

  • Ground surface settlement prediction method and system based on fusion of graph convolutional network and multi-point relationship

    CN116644654A

  • Monitoring and early warning method and system for mining subsidence deformation of mining area

    CN116659410A