A ground subsidence monitoring method and a training method of a feature fusion prediction network

Abstract

This invention discloses a method for monitoring land subsidence and a training method for a feature fusion prediction network, relating to the field of geological disaster monitoring and early warning. The land subsidence monitoring method includes: acquiring an InSAR subsidence dataset and a multi-source static environmental factor dataset of the target monitoring area; performing spatial registration and normalization processing on the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain land subsidence prediction data; and obtaining predicted land subsidence data based on the land subsidence prediction data and the feature fusion prediction network. The technical solution of this invention can more accurately predict land subsidence based on the multi-source data features of temporal subsidence data and static environmental factors, effectively improving the generalization of the prediction method.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and in particular to a method for monitoring land subsidence and a training method for a feature fusion prediction network. Background Technology

[0002] Land subsidence is a slowly changing geological hazard that poses a serious threat to infrastructure safety and people's lives and property. Synthetic InSAR (Interferometric Synthetic Aperture Radar) technology, especially SBAS-InSAR (Small Baseline Subset Interferometric Synthetic Aperture Radar), has become an important means of large-scale land subsidence monitoring due to its advantages such as wide coverage, high monitoring accuracy, and immunity to weather conditions.

[0003] Currently, InSAR-based settlement prediction research mainly falls into two categories: The first category employs purely data-driven deep learning models, such as Long Short-Term Memory networks, to directly fit and predict time-series settlement data obtained through InSAR. However, when mining conditions and hydrological boundaries change, purely data-driven models are prone to failure due to shifts in the distribution of training data, resulting in insufficient generalization ability. The second category integrates InSAR deformation results as dynamic evaluation factors into landslide susceptibility assessment models. Using the surface deformation rate obtained by SBAS-InSAR as a dynamic factor, combined with static environmental factors such as elevation, slope, and lithology, models such as Random Forest and XGBoost (eXtreme Gradient Boosting) are employed for landslide susceptibility assessment, significantly improving evaluation accuracy. However, this method is only used for susceptibility level classification and cannot predict future settlement amounts. Summary of the Invention

[0004] This invention provides a method for monitoring land subsidence and a training method for a feature fusion prediction network to solve the problem that existing subsidence prediction methods cannot deeply integrate time-series subsidence data with static environmental factors, resulting in weak accuracy and generalization ability in land subsidence prediction.

[0005] According to one aspect of the present invention, a method for monitoring land subsidence is provided, comprising: Acquire InSAR settlement datasets and multi-source static environmental factor datasets for the target monitoring area; Spatial registration and normalization were performed on the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain the surface subsidence prediction data. Based on the surface subsidence prediction data and the feature fusion prediction network, the surface subsidence prediction data is obtained.

[0006] According to another aspect of the present invention, a method for training a feature fusion prediction network is provided, comprising: Acquire InSAR historical settlement dataset and multi-source static environmental factor historical dataset; Spatial registration and normalization were performed on the InSAR historical subsidence dataset and the multi-source static environmental factor historical dataset to obtain the surface subsidence training sample data. The training sample data of land subsidence is input into a preset deep learning network to perform spatiotemporal feature fusion and land subsidence prediction, resulting in a feature fusion prediction network. The feature fusion prediction network is applied in the land subsidence monitoring method of any embodiment of the present invention.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to execute the land subsidence monitoring method according to any embodiment of the present invention, or to execute the feature fusion prediction network training method according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute, implement the land subsidence monitoring method according to any embodiment of the present invention, or execute the training method of the feature fusion prediction network according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the land subsidence monitoring method according to any embodiment of the present invention, or implements the training method of the feature fusion prediction network according to any embodiment of the present invention.

[0010] The technical solution of this invention acquires InSAR subsidence datasets and multi-source static environmental factor datasets of the target monitoring area. It then performs spatial registration and normalization on the InSAR subsidence dataset and the multi-source static environmental factor datasets to obtain surface subsidence prediction data. Based on this surface subsidence prediction data and a feature fusion prediction network, it obtains predicted surface subsidence data. This solution deeply aligns static environmental factors with InSAR subsidence data possessing temporal attributes. Through a feature fusion prediction network, it accurately predicts surface subsidence based on the aligned and normalized surface subsidence prediction data according to the fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply fuse temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. This solution enables more accurate prediction of surface subsidence based on the multi-source data features of temporal subsidence data and static environmental factors, effectively improving the generalization of the prediction method.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a surface subsidence monitoring method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a surface subsidence monitoring method provided in Embodiment 2 of the present invention; Figure 3 The flowchart illustrates the training of a feature fusion prediction network according to Embodiment 3 of the present invention. Figure 4 This is a schematic diagram of the structure of a surface subsidence monitoring device provided in Embodiment 4 of the present invention; Figure 5 This is a schematic diagram of the structure of a training device for a feature fusion prediction network provided in Embodiment 4 of the present invention; Figure 6 A schematic diagram of an electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "target," "current," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] Example 1 Figure 1 This is a flowchart of a land subsidence monitoring method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations requiring accurate prediction of land subsidence. The method can be executed by a land subsidence monitoring device, which can be implemented in hardware and / or software and can be configured in an electronic device. For example... Figure 1 As shown, the method includes: Step 110: Obtain the InSAR settlement dataset and multi-source static environmental factor dataset of the target monitoring area.

[0017] The target monitoring area can be the geographical area where land subsidence monitoring is conducted. The InSAR subsidence dataset can be a dataset acquired in real time using InSAR technology. The data in the InSAR subsidence dataset is time-series data. The multi-source static environmental factor dataset can be a dataset describing geographical environmental conditions that is acquired in real time and does not change significantly over time.

[0018] In this embodiment of the invention, after determining the target monitoring area for surface subsidence monitoring, relevant data of the target monitoring area can be collected in real time to obtain InSAR subsidence dataset and multi-source static environmental factor dataset.

[0019] Step 120: Spatial registration and normalization are performed on the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain the surface subsidence prediction data.

[0020] Spatial registration can be an operation that spatially aligns and matches the InSAR subsidence dataset with the multi-source static environmental factor dataset. The surface subsidence data to be predicted can be the result of spatial registration and normalization processing of the InSAR subsidence dataset and the multi-source static environmental factor dataset.

[0021] In this embodiment of the invention, in order to ensure that the InSAR subsidence dataset and the multi-source static environmental factor dataset can be deeply and effectively fused, the InSAR subsidence dataset and the multi-source static environmental factor dataset can be aligned according to sampling time and sampling space (i.e., spatial registration), and the aligned dataset can be normalized to obtain surface subsidence prediction data with uniform dimensions.

[0022] Step 130: Obtain surface subsidence prediction data based on the surface subsidence prediction data and the feature fusion prediction network.

[0023] The feature fusion prediction network can be a pre-trained model capable of predicting land subsidence. It fuses temporal and static environmental features in the land subsidence prediction data and predicts land subsidence based on the fused features. The land subsidence prediction data can be the predicted land subsidence output by the feature fusion prediction network.

[0024] In this embodiment of the invention, the surface subsidence prediction data can be input into a pre-trained feature fusion prediction network to generate surface subsidence prediction data, thereby issuing an early warning immediately when the surface subsidence prediction data reaches the warning level.

[0025] The technical solution of this invention acquires InSAR subsidence datasets and multi-source static environmental factor datasets of the target monitoring area. It then performs spatial registration and normalization on the InSAR subsidence dataset and the multi-source static environmental factor datasets to obtain surface subsidence prediction data. Based on this surface subsidence prediction data and a feature fusion prediction network, it obtains predicted surface subsidence data. This solution deeply aligns static environmental factors with InSAR subsidence data possessing temporal attributes. Through a feature fusion prediction network, it accurately predicts surface subsidence based on the aligned and normalized surface subsidence prediction data according to the fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply fuse temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. This solution enables more accurate prediction of surface subsidence based on the multi-source data features of temporal subsidence data and static environmental factors, effectively improving the generalization of the prediction method.

[0026] Example 2 Figure 2 This is a flowchart of a land subsidence monitoring method provided in Embodiment 2 of the present invention. This embodiment is a specific implementation based on the above embodiment, and provides a specific optional implementation method for spatially registering and normalizing InSAR subsidence datasets and multi-source static environmental factor datasets to obtain land subsidence prediction data. Figure 2 As shown, the method includes: Step 210: Obtain the InSAR settlement dataset and multi-source static environmental factor dataset of the target monitoring area.

[0027] In an optional embodiment of the present invention, obtaining the InSAR subsidence dataset and the multi-source static environmental factor dataset of the target monitoring area may include: obtaining the surface subsidence matrix and the multi-source static environmental factors of the target monitoring area according to a preset data acquisition frequency; constructing the InSAR subsidence dataset based on the surface subsidence matrix of the target monitoring area under the current time window, and constructing the multi-source static environmental factor dataset based on the multi-source static environmental factors of the target monitoring area under the current time window.

[0028] The preset data acquisition frequency can be a pre-set data acquisition frequency. The surface subsidence matrix can be a matrix composed of surface subsidence values ​​collected from different physical locations in a single data sampling operation. The current time window can be the current data acquisition period for the InSAR subsidence dataset and the multi-source static environmental factor dataset. The multi-source static environmental factors can be elements describing geographical environmental conditions that do not change significantly over time. For example, multi-source static environmental factors may include at least one of the following: elevation, slope, engineering geological rock group, distance from water system, distance from fault, land use type, and normalized difference vegetation index.

[0029] In this embodiment of the invention, the surface subsidence matrix of the target monitoring area can be obtained according to a preset data acquisition frequency, and multi-source static environmental factors can be obtained according to a preset data acquisition frequency. That is, the surface subsidence matrix and multi-source static environmental factors of the target monitoring area collected under one sampling operation are then used to construct an InSAR subsidence dataset based on the surface subsidence matrix of the target monitoring area under the current time window, and the multi-source static environmental factors of the target monitoring area under the current time window are used to construct a multi-source static environmental factor dataset.

[0030] Step 220: Spatial registration of the InSAR settlement dataset and the multi-source static environmental factor dataset according to their physical locations, to obtain the spatially registered InSAR settlement dataset and the spatially registered multi-source static environmental factor dataset.

[0031] Specifically, the spatially registered InSAR settlement dataset can be an InSAR settlement dataset spatially registered with a multi-source static environmental factor dataset. The spatially registered multi-source static environmental factor dataset can also be a multi-source static environmental factor dataset spatially registered with an InSAR settlement dataset.

[0032] In this embodiment of the invention, the InSAR settlement dataset and the multi-source static environmental factor dataset can be spatially registered to the same spatial resolution to obtain a spatially registered InSAR settlement dataset and a spatially registered multi-source static environmental factor dataset.

[0033] Step 230: Normalize the spatially registered InSAR subsidence dataset and the spatially registered multi-source static environmental factor dataset to obtain the surface subsidence prediction data.

[0034] In this embodiment of the invention, the spatially registered InSAR subsidence dataset and the spatially registered multi-source static environmental factor dataset can be normalized according to a preset unified dimension to obtain the surface subsidence prediction data.

[0035] Step 240: Obtain surface subsidence prediction data based on the surface subsidence prediction data and the feature fusion prediction network.

[0036] The technical solution of this invention involves acquiring an InSAR subsidence dataset and a multi-source static environmental factor dataset of the target monitoring area, then spatially registering the InSAR subsidence dataset and the multi-source static environmental factor dataset according to their physical locations to obtain a spatially registered InSAR subsidence dataset and a spatially registered multi-source static environmental factor dataset. These datasets are then normalized to obtain surface subsidence prediction data. Finally, based on the surface subsidence prediction data and a feature fusion prediction network, surface subsidence prediction data is obtained. This scheme deeply aligns static environmental factors with InSAR subsidence data possessing temporal attributes. Through a feature fusion prediction network, the aligned and normalized surface subsidence data is used to accurately predict surface subsidence based on fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply integrate temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. This approach enables more accurate prediction of surface subsidence based on multi-source data features of temporal subsidence data and static environmental factors, effectively improving the generalization of the prediction method.

[0037] Example 3 Figure 3 This is a flowchart of the training of a feature fusion prediction network provided in Embodiment 3 of the present invention, as follows: Figure 3 As shown, the method includes: Step 310: Obtain the InSAR settlement history dataset and the multi-source static environmental factor history dataset.

[0038] The InSAR historical settlement dataset can be a historical dataset collected using InSAR technology. The multi-source static environmental factor historical dataset can be a historical dataset describing the geological environmental conditions that does not change significantly over time. The InSAR historical settlement dataset and the multi-source static environmental factor historical dataset were collected over the same time period and in the same monitoring area.

[0039] In this embodiment of the invention, it may involve acquiring historical InSAR settlement datasets and historical datasets of multi-source static environmental factors from the same time period and the same monitoring area.

[0040] Step 320: Spatial registration and normalization are performed on the InSAR historical subsidence dataset and the multi-source static environmental factor historical dataset to obtain the surface subsidence training sample data.

[0041] Among them, the surface subsidence training sample data can be the result of spatial registration and normalization processing of InSAR subsidence historical dataset and multi-source static environmental factor historical dataset.

[0042] In this embodiment of the invention, the InSAR subsidence history dataset and the multi-source static environmental factor history dataset can be spatially aligned and normalized according to their physical locations to obtain surface subsidence training sample data.

[0043] Step 330: Input the surface subsidence training sample data into the preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain the feature fusion prediction network.

[0044] The feature fusion prediction network is applied in any embodiment of the land subsidence monitoring method. Spatiotemporal feature fusion can be an operation that fuses the temporal features and static spatial features in the land subsidence training sample data.

[0045] In this embodiment of the invention, the temporal features and static spatial features in the training sample data of land subsidence can be aligned and fused based on a preset deep learning network, and land subsidence prediction training can be performed based on the aligned and fused features. Then, the deep learning network that meets the accuracy requirements after training can be used as the feature fusion prediction network.

[0046] Optionally, the surface subsidence training sample data can be divided into training sample data and test sample data according to a preset ratio. The training sample data can be used to train a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction, and the test sample data can be used to test the feature fusion prediction network.

[0047] In an optional embodiment of the present invention, the preset deep learning network may include a static factor spatiotemporal coding branch, a sedimentation temporal coding branch, a feature fusion layer, and a temporal prediction layer.

[0048] The static factor spatiotemporal coding branch can be used to extract static spatial features from multi-source static factors. The settlement temporal coding branch can be used to extract temporal features from InSAR settlement data. The feature fusion layer can be a network layer that performs feature fusion. The temporal prediction layer can be a network layer that predicts land subsidence. The temporal prediction layer can be a gate control unit or a Transformer decoder structure.

[0049] In an optional embodiment of the present invention, after inputting the surface subsidence training sample data into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain a feature fusion prediction network, the method may further include: obtaining a total loss function composed of a temporal fitting loss function and a spatial consistency loss function; performing loss minimization training iterations on the feature fusion prediction network according to the total loss function to obtain network convergence parameters; and updating the feature fusion prediction network according to the network convergence parameters.

[0050] The temporal fitting loss function measures the difference between predicted and actual surface subsidence. The spatial consistency loss function measures the spatial continuity of predicted surface subsidence at adjacent physical locations. The total loss function is a weighted combination of the temporal fitting loss function and the spatial consistency loss function. The network convergence parameters are the network parameters obtained when the feature fusion prediction network reaches convergence after optimization iterations.

[0051] In this embodiment of the invention, the temporal fitting loss function and the spatial consistency loss function can be weighted to obtain the total loss function. Then, based on the end-to-end training method, the feature fusion prediction network is trained iteratively with the goal of minimizing the total loss function until the feature fusion prediction network converges, and the network convergence parameters are obtained. Then, the network parameters in the feature fusion prediction network are updated using the network convergence parameters.

[0052] In an optional embodiment of the present invention, after updating the feature fusion prediction network according to the network convergence parameters, the method may further include: inputting the surface subsidence test sample data into the updated feature fusion prediction network to obtain the network prediction test results; and evaluating the accuracy of the network prediction test results according to the network prediction accuracy measurement index; the network prediction accuracy measurement index may include at least one of root mean square error, mean absolute error, and coefficient of determination.

[0053] The surface subsidence test sample data can be test sample data divided into test sample data according to a preset ratio from the surface subsidence training sample data. The network prediction test result can be the prediction result of the feature fusion prediction network on the surface subsidence test sample data.

[0054] In this embodiment of the invention, surface subsidence test sample data, segmented from the surface subsidence training sample data, can be input into the updated feature fusion prediction network. That is, the feature fusion prediction network with updated parameters is tested, and the feature fusion prediction network outputs the network prediction test results. Then, at least one of the root mean square error, mean absolute error, and coefficient of determination of the actual surface subsidence corresponding to the surface subsidence test sample data is used as the network prediction accuracy measurement index. Further, the accuracy of the network prediction test results is evaluated based on the network prediction accuracy measurement index, that is, it is determined whether the network prediction test results and the network prediction accuracy measurement index meet the preset error range. The feature fusion prediction network that meets the error range is used as the prediction network for final use.

[0055] The technical solution of this invention involves acquiring historical InSAR subsidence datasets and historical datasets of multi-source static environmental factors, then performing spatial registration and normalization on the historical InSAR subsidence datasets and historical datasets of multi-source static environmental factors to obtain surface subsidence training sample data. This surface subsidence training sample data is then input into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain a feature fusion prediction network. This scheme deeply aligns historical static environmental factors with historical InSAR subsidence data possessing temporal attributes. The aligned and normalized surface subsidence training sample data is then input into a pre-defined deep learning network for spatiotemporal feature fusion and surface subsidence prediction training. This enables the feature fusion prediction network to predict surface subsidence through spatiotemporal feature fusion, achieving accurate prediction of surface subsidence based on the fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply integrate temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. Based on the multi-source data features of temporal subsidence data and static environmental factors, it can more accurately predict surface subsidence, effectively improving the generalization of the prediction method.

[0056] In a specific example, the method for predicting land subsidence includes the following steps: Step S1: Multi-source data acquisition and preprocessing Obtain the InSAR settlement dataset and the multi-source static environmental factor dataset of the target monitoring area. .in, Let H represent the surface subsidence matrix at time t. H and W represent the length and width of the area occupied by the target monitoring region, respectively. Spatial registration and normalization are performed on the InSAR subsidence dataset and the multi-source static environmental factor dataset.

[0057] Step S2: Construct a feature fusion prediction network A two-branch deep learning network is constructed, consisting of a static factor spatiotemporal coding branch, a sedimentation time-series coding branch, and a feature fusion layer.

[0058] The static factor spatiotemporal coding branch employs a convolutional neural network to extract spatial features from a historical dataset of multi-source static environmental factors. Through multi-layer convolution and pooling operations, it learns the nonlinear combination relationships between static environmental factors and outputs a static spatial feature map. ,in, and These represent the height and width of the feature map, respectively. This represents the static feature channel count, i.e., the feature dimension extracted by the convolutional neural network. This branch can capture the influence of geological environmental conditions on the spatial distribution of subsidence-sensitive areas.

[0059] The settlement temporal coding branch employs a bidirectional long short-term memory network to extract temporal features from each monitoring point in the InSAR settlement history dataset. For each spatial location... Extract the time dimension of the surface subsidence matrix sequence. Input a bidirectional long short-term memory network and output temporal hidden state features. , This is the temporal feature dimension, specifically the number of units in the hidden layers of a bidirectional long short-term memory network.

[0060] The feature fusion layer spatially aligns and fuses the static spatial feature map with the temporal hidden state features. Specifically, for each spatial location, its temporal hidden state features (i.e., the aforementioned temporal features) are concatenated with the corresponding static spatial features, and then the features are interacted through a fully connected layer to obtain the fused feature vector. The fusion process can be represented as: Where MLP stands for Multilayer Perceptron and Concat is a vector concatenation operation. and The specific value is determined by the network structure.

[0061] Step S3: Construct a multi-step prediction module based on fused features fuse feature vectors The input is fed into a time-series prediction layer, which outputs the predicted settlement values ​​for the next k time phases. The calculation process is as follows: ;in, For spatial location Predicted value of surface subsidence at time phase T+K.

[0062] Step S4: Construct a composite loss function for model training A composite loss function, comprising a temporal fitting loss function and a spatial consistency loss function, is constructed and used for end-to-end training of the feature fusion prediction network. The temporal fitting loss function uses mean squared error to measure the difference between the predicted and actual values ​​of land subsidence. Where N is the number of monitoring points and K is the prediction step size. Let be the predicted surface subsidence value of the nth monitoring point at the (T+K)th time phase. This corresponds to the actual value.

[0063] The spatial consistency loss function introduces a spatial smoothing constraint to ensure that the predicted surface subsidence at adjacent locations conforms to spatial continuity. Its role is to guide the model to learn the physical laws of surface subsidence, meaning that the subsidence field is typically continuous in space, avoiding unreasonable abrupt jumps between adjacent pixels. The spatial consistency loss function can be expressed as: ; in, To monitor the set of all spatial locations within the area, Indicates position The set of neighboring pixels (usually four or eight neighbors). for neighborhood location The predicted values ​​are from the same time phase. This loss, by penalizing the difference in predicted values ​​of adjacent pixels, prompts the model to output a spatially smooth settling field.

[0064] The total loss function is: ;in, The balancing coefficient is used to control the strength of spatial consistency constraints, and its optimal value is determined through cross-validation.

[0065] An end-to-end training approach is adopted, aiming to minimize the total loss L. The gradient of the network parameters is calculated using the backpropagation algorithm, and the parameters are iteratively updated using the Adam optimizer until the feature fusion prediction network converges. During training, To ensure the accuracy of the predicted values, To ensure the spatial rationality of the prediction results, the two are optimized in tandem, so that the model can simultaneously possess high-precision prediction capabilities and consistency with physical laws.

[0066] Step S5: Settlement Prediction and Accuracy Assessment The surface subsidence test sample data of the target monitoring area is input into a trained feature fusion prediction network, which outputs surface subsidence prediction results for multiple future time phases. The root mean square error, mean absolute error, and coefficient of determination are used to evaluate the accuracy of the prediction results, and a subsidence prediction distribution map is generated.

[0067] This scheme, for the first time, integrates multi-source static environmental factors with InSAR time-series subsidence data for prediction using deep learning. This enables the prediction network to learn both the temporal evolution patterns and the control effect of regional geological background on subsidence, significantly improving prediction accuracy and generalization ability. Through a spatiotemporal encoding branch of static factors, a convolutional neural network is used to extract the spatial distribution features of static factors, which are then fused with point-by-point time-series features. This achieves explicit modeling of the spatial continuity of subsidence, overcoming the deficiency of traditional single-point time-series prediction models that ignore spatial correlation. A spatial consistency loss is introduced to constrain the prediction results of adjacent regions to conform to the spatial smoothness characteristics of the subsidence field, making the prediction results more consistent with actual physical laws and avoiding local anomalous jumps. Static feature extraction, time-series feature extraction, and multi-step prediction are unified in an end-to-end network framework, achieving collaborative optimization of multi-source information and avoiding error accumulation caused by staged processing.

[0068] Example 4 Figure 4 This is a schematic diagram of the structure of a land subsidence monitoring device provided in Embodiment 4 of the present invention. Figure 4 As shown, the device includes: The first data acquisition module is used to acquire the InSAR settlement dataset and multi-source static environmental factor dataset of the target monitoring area. The first data preprocessing module is used to perform spatial registration and normalization of the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain the surface subsidence prediction data. The surface subsidence prediction module is used to obtain surface subsidence prediction data based on the surface subsidence data to be predicted and the feature fusion prediction network.

[0069] The technical solution of this invention acquires InSAR subsidence datasets and multi-source static environmental factor datasets of the target monitoring area. It then performs spatial registration and normalization on the InSAR subsidence dataset and the multi-source static environmental factor datasets to obtain surface subsidence prediction data. Based on this surface subsidence prediction data and a feature fusion prediction network, it obtains predicted surface subsidence data. This solution deeply aligns static environmental factors with InSAR subsidence data possessing temporal attributes. Through a feature fusion prediction network, it accurately predicts surface subsidence based on the aligned and normalized surface subsidence prediction data according to the fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply fuse temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. This solution enables more accurate prediction of surface subsidence based on the multi-source data features of temporal subsidence data and static environmental factors, effectively improving the generalization of the prediction method.

[0070] Optionally, the first data acquisition module is used to acquire the surface subsidence matrix and multi-source static environmental factors of the target monitoring area according to a preset data acquisition frequency; based on the surface subsidence matrix of the target monitoring area under the current time window, the InSAR subsidence dataset is constructed, and based on the multi-source static environmental factors of the target monitoring area under the current time window, the multi-source static environmental factor dataset is constructed.

[0071] Optionally, the first data preprocessing module is used to spatially register the InSAR subsidence dataset and the multi-source static environmental factor dataset according to their physical locations to obtain a spatially registered InSAR subsidence dataset and a spatially registered multi-source static environmental factor dataset; and to normalize the spatially registered InSAR subsidence dataset and the spatially registered multi-source static environmental factor dataset to obtain surface subsidence prediction data.

[0072] The surface subsidence monitoring device provided in the embodiments of the present invention can execute the surface subsidence monitoring method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0073] Figure 5 This is a schematic diagram of the structure of a training device for a feature fusion prediction network provided in Embodiment 4 of the present invention. Figure 5 As shown, the device includes: The second data acquisition module is used to acquire historical InSAR settlement datasets and historical datasets of multi-source static environmental factors.

[0074] The second data preprocessing module is used to perform spatial registration and normalization processing on the InSAR historical subsidence dataset and the multi-source static environmental factor historical dataset to obtain surface subsidence training sample data.

[0075] The model training module is used to input the surface subsidence training sample data into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training, thereby obtaining the feature fusion prediction network. The feature fusion prediction network can be applied to the surface subsidence monitoring method in any embodiment.

[0076] The technical solution of this invention involves acquiring historical InSAR subsidence datasets and historical datasets of multi-source static environmental factors, then performing spatial registration and normalization on the historical InSAR subsidence datasets and historical datasets of multi-source static environmental factors to obtain surface subsidence training sample data. This surface subsidence training sample data is then input into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain a feature fusion prediction network. This scheme deeply aligns historical static environmental factors with historical InSAR subsidence data possessing temporal attributes. The aligned and normalized surface subsidence training sample data is then input into a pre-defined deep learning network for spatiotemporal feature fusion and surface subsidence prediction training. This enables the feature fusion prediction network to predict surface subsidence through spatiotemporal feature fusion, achieving accurate prediction of surface subsidence based on the fused features. This avoids the failure of networks trained solely on InSAR subsidence data due to data distribution offsets, significantly improving the generalization ability of the feature fusion prediction network. It solves the problem that existing subsidence prediction methods cannot deeply integrate temporal subsidence data with static environmental factors, resulting in weak accuracy and generalization ability. Based on the multi-source data features of temporal subsidence data and static environmental factors, it can more accurately predict surface subsidence, effectively improving the generalization of the prediction method.

[0077] Optionally, the preset deep learning network includes a static factor spatiotemporal coding branch, a sedimentation temporal coding branch, a feature fusion layer, and a temporal prediction layer.

[0078] Optionally, the training device for the feature fusion prediction network further includes a network update module, used to obtain a total loss function composed of a temporal fitting loss function and a spatial consistency loss function; perform loss minimization training iterations on the feature fusion prediction network according to the total loss function to obtain network convergence parameters; and update the feature fusion prediction network according to the network convergence parameters.

[0079] Optionally, the training device for the feature fusion prediction network further includes a network accuracy evaluation module, which is used to input the surface subsidence test sample data into the updated feature fusion prediction network to obtain the network prediction test results; and to evaluate the accuracy of the network prediction test results according to the network prediction accuracy measurement index; the network prediction accuracy measurement index includes at least one of root mean square error, mean absolute error and coefficient of determination.

[0080] Example 5 Figure 6 A schematic diagram of an electronic device that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0081] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as ROM 12, RAM 13, etc., communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An I / O interface 15 is also connected to the bus 14. The ROM 12 is a read-only memory, the RAM 13 is a random access memory, and the I / O interface 15 is an input / output interface.

[0082] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0083] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as land subsidence monitoring methods or training methods for feature fusion prediction networks.

[0084] In some embodiments, the land subsidence monitoring method or the feature fusion prediction network training method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the land subsidence monitoring method or the feature fusion prediction network training method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the land subsidence monitoring method or the feature fusion prediction network training method by any other suitable means (e.g., by means of firmware).

[0085] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0086] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0087] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0088] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0089] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0090] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS servers, such as high management difficulty and weak business scalability.

[0091] This application also discloses a computer program product, which includes a computer program that, when executed by a processor, implements the land subsidence monitoring method or the feature fusion prediction network training method provided in any embodiment of this application. This program product shares the same inventive concept as the land subsidence monitoring method or the feature fusion prediction network training method disclosed in the embodiments of this application, and therefore will not be described in detail here.

[0092] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0093] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for monitoring land subsidence, characterized in that, include: Acquire synthetic aperture radar interferometry (InSAR) settlement dataset and multi-source static environmental factor dataset for the target monitoring area; Spatial registration and normalization are performed on the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain the surface subsidence prediction data. Based on the surface subsidence prediction data and the feature fusion prediction network, surface subsidence prediction data is obtained.

2. The method for monitoring land subsidence according to claim 1, characterized in that, Obtain the InSAR settlement dataset and multi-source static environmental factor dataset for the target monitoring area, including: According to the preset data acquisition frequency, obtain the surface subsidence matrix and multi-source static environmental factors of the target monitoring area; The InSAR subsidence dataset is constructed based on the surface subsidence matrix of the target monitoring area under the current time window, and the multi-source static environmental factor dataset is constructed based on the multi-source static environmental factors of the target monitoring area under the current time window.

3. The method for monitoring land subsidence according to any one of claims 1-2, characterized in that, Spatial registration and normalization are performed on the InSAR subsidence dataset and the multi-source static environmental factor dataset to obtain surface subsidence prediction data, including: The InSAR settlement dataset and the multi-source static environmental factor dataset are spatially registered according to their physical locations to obtain a spatially registered InSAR settlement dataset and a spatially registered multi-source static environmental factor dataset. The spatially registered InSAR subsidence dataset and the spatially registered multi-source static environmental factor dataset are normalized to obtain the surface subsidence prediction data.

4. A training method for a feature fusion prediction network, characterized in that, include: Acquire InSAR historical settlement dataset and multi-source static environmental factor historical dataset; Spatial registration and normalization are performed on the InSAR historical subsidence dataset and the multi-source static environmental factor historical dataset to obtain surface subsidence training sample data. The surface subsidence training sample data is input into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain the feature fusion prediction network. The feature fusion prediction network is applied in the land subsidence monitoring method according to any one of claims 1-3.

5. The training method for the feature fusion prediction network according to claim 4, characterized in that, The preset deep learning network includes a static factor spatiotemporal coding branch, a sedimentation temporal coding branch, a feature fusion layer, and a temporal prediction layer.

6. The training method for the feature fusion prediction network according to claim 4, characterized in that, After inputting the surface subsidence training sample data into a preset deep learning network for spatiotemporal feature fusion and surface subsidence prediction training to obtain the feature fusion prediction network, the method further includes: Obtain the total loss function, which is composed of the temporal fitting loss function and the spatial consistency loss function; Based on the total loss function, the feature fusion prediction network is trained iteratively to minimize the loss, and the network convergence parameters are obtained. The feature fusion prediction network is updated based on the network convergence parameters.

7. The training method for the feature fusion prediction network according to claim 6, characterized in that, After updating the feature fusion prediction network based on the network convergence parameters, the method further includes: The surface subsidence test sample data is input into the updated feature fusion prediction network to obtain the network prediction test results. The accuracy of the network prediction test results is evaluated based on network prediction accuracy metrics; the network prediction accuracy metrics include at least one of root mean square error, mean absolute error, and coefficient of determination.

8. An electronic device, characterized in that, The electronic device includes: At least one processor, and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the land subsidence monitoring method according to any one of claims 1-3, or to perform the training method of the feature fusion prediction network according to any one of claims 4-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the land subsidence monitoring method of any one of claims 1-3, or the training method of the feature fusion prediction network of any one of claims 4-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the land subsidence monitoring method according to any one of claims 1-3, or implements the training method for the feature fusion prediction network according to any one of claims 4-7.

Images