Urban land subsidence monitoring system based on multi-source remote sensing data fusion
By using a multi-source remote sensing data fusion system and super-resolution reconstruction technology and deep learning methods, the shortcomings of remote sensing data in spatial resolution are solved, enabling high-precision monitoring of urban land subsidence and supporting early warning and prevention.
Patent Information
- Application Number
- CN202410963519.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing remote sensing data fusion methods are unable to capture subtle changes in urban land subsidence in terms of spatial resolution, making early warning and prevention difficult.
A multi-source remote sensing data fusion system is adopted, including modules for data acquisition, preprocessing, fusion, super-resolution reconstruction, and settlement monitoring and analysis. The super-resolution reconstruction method is used to improve the spatial resolution of the data, and combined with deep learning and generative adversarial network technology, high-resolution images are generated and settlement analysis is performed.
It significantly improves the accuracy and sensitivity of urban ground subsidence monitoring, enabling it to keenly capture subtle subsidence changes and provide crucial data support for safety assessment and early warning of urban infrastructure.
Smart Images

Figure CN118918425A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of land subsidence monitoring, and in particular to an urban land subsidence monitoring system that integrates multi-source remote sensing data. Background Art
[0002] Urban land subsidence, also known as ground subsidence or land subsidence, is a global environmental challenge, especially in areas with rapid urbanization. Urban land subsidence refers to the phenomenon of the ground in urban areas dropping in height due to various reasons. It usually manifests itself in two forms: regional subsidence and local subsidence, which can cause buildings to tilt, destroy the stability of the foundation, and even cause serious impacts such as seawater backflow to coastal cities. The main purpose is to understand the development process of land subsidence, analyze its causes, predict its trends, and formulate corresponding prevention and control strategies. It is necessary to monitor land subsidence. Through monitoring, land subsidence problems can be discovered in time, and their impact on buildings, infrastructure and human activities can be evaluated, so that necessary measures can be taken to reduce or eliminate its hazards.
[0003] Traditional ground subsidence monitoring methods, such as leveling and GPS measurement, have high accuracy, but have defects such as limited monitoring range, low work efficiency and high cost. With the continuous development of remote sensing technology, it has been widely used in ground subsidence monitoring. However, a single remote sensing data source is often difficult to fully and accurately reflect the real situation of ground subsidence. Although multi-source remote sensing data fusion technology can make up for this deficiency to a certain extent, the existing fusion method still needs to be improved in spatial resolution and it is difficult to capture subtle changes in subsidence, which poses a challenge to early warning and precise prevention and control of ground subsidence. Therefore, it is necessary to propose an urban ground subsidence monitoring system based on multi-source remote sensing data fusion to solve the problems in the existing technology. Summary of the invention
[0004] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide an urban land subsidence monitoring system that integrates multi-source remote sensing data. It can use a unique super-resolution reconstruction method to fuse multi-source remote sensing data, thereby significantly improving the spatial resolution of urban land subsidence monitoring and being able to keenly capture more subtle settlement changes.
[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an urban land subsidence monitoring system integrating multi-source remote sensing data, the system comprising:
[0006] Data acquisition module, used to collect multiple remote sensing data sources from different sensors and platforms;
[0007] The data preprocessing module is used to perform preprocessing operations on the collected multi-source remote sensing data to eliminate and reduce errors and inconsistencies in the data;
[0008] Data fusion module, used to organically fuse pre-processed multi-source remote sensing data;
[0009] The super-resolution reconstruction module uses the super-resolution reconstruction method to perform super-resolution reconstruction on the fused data to improve the spatial resolution of the data. The specific execution steps of the super-resolution reconstruction method include: feature extraction stage, deep learning model construction stage, model training stage and image reconstruction stage;
[0010] The subsidence monitoring and analysis module is used to analyze the spatiotemporal characteristics and development trends of ground subsidence based on super-resolution reconstructed data;
[0011] The result output module is used to output the monitoring and analysis results in a visual form.
[0012] Furthermore, the specific execution steps of the data acquisition module include:
[0013] Covers a wide range of remote sensing data sources, including but not limited to high-resolution optical remote sensing satellites, synthetic aperture radar remote sensing satellites, and airborne LiDAR;
[0014] According to monitoring requirements and data source characteristics, set the collection time interval, spatial range and band selection parameters;
[0015] Collect data through sensors and platforms according to the set parameters;
[0016] Conduct a preliminary screening of the collected data to remove data with obvious quality issues and data that does not meet the preset requirements;
[0017] The collected valid data will be stored securely and transmitted to the data processing center.
[0018] Furthermore, the specific execution steps of the data preprocessing module include:
[0019] First, the original digital count DN is converted into radiance value and reflectance using the calibration parameters provided by the sensor: L = DN × G + O, where L is the radiance, DN is the original digital count value, G represents the gain, and O represents the offset.
[0020] Next, the atmospheric radiation transfer model and ground observation data are used to estimate and remove the influence of the atmosphere on radiation. Where ρ is the surface reflectivity, L is the radiation brightness received by the sensor, and L a is the downward radiance of the atmosphere, E0 is the solar irradiance, θ s is the solar zenith angle, τ is the atmospheric transmittance;
[0021] Select multiple corresponding control points on the remote sensing image and the reference map;
[0022] The polynomial model is used to perform least squares fitting on the control points to establish the conversion relationship between image coordinates and geographic coordinates: x′=a0+a1x+a2y+a3xy, y′=b0+b1x+b2y+b3xy, where (x′, y′) is the geographic coordinate, (x, y) is the image coordinate, and a i and b i are the polynomial coefficients;
[0023] According to the established geometric model, the original image is resampled to generate a corrected image;
[0024] The median filtering method is used to remove noise by taking the median value of its neighboring pixels as the new value for each pixel in the image. Assume that the neighboring pixel values are p1, p2, ..., p n , then the pixel value p after median filtering filtered For: p filtered =median(p1,p2,…,p n ).
[0025] Furthermore, the specific execution steps of the data fusion module include:
[0026] The preprocessed multi-source remote sensing data is used as input, and the multi-source remote sensing data includes optical remote sensing data and SAR remote sensing data;
[0027] For optical remote sensing data and SAR remote sensing data, their features are extracted respectively, including the spectral feature F of optical remote sensing data opt and the texture feature F of SAR remote sensing data sar , assuming that the optical remote sensing data is I opt , through some feature extraction function f opt Extract spectral features: F opt =f opt (I opt ), for SAR remote sensing data, F sar =f sar (I sar );
[0028] Calculate the similarity and correlation between the extracted features, using cosine similarity to measure. For two feature vectors F1 and F2, the calculation formula of cosine similarity S is:
[0029] According to the similarity and correlation, the features are weighted and fused to obtain the fused feature F. fusion : F fusion=w1F1+w2F2, where w1 and w2 are weights calculated based on similarity;
[0030] For the same type of remote sensing data in different phases, the fusion is performed based on statistical analysis and decision rules, and the mean statistics of multiple phase data are calculated. Suppose the phase data D1, D2, ..., D n , the result of mean fusion D mean for:
[0031] Furthermore, the specific execution steps of the feature extraction stage in the super-resolution reconstruction module include:
[0032] Using a multi-layer convolutional neural network CNN to input a low-resolution image I LR Perform preliminary feature extraction to obtain shallow basic features, including edges and colors. Assume that the output feature map of the first convolution operation is F l , the calculation formula is: F l =σ(W l *F l-1 +b l ), where W l is the convolution kernel weight of layer l, b l is the bias term, σ is the activation function, and * represents the convolution operation;
[0033] The residual network structure is introduced to deeply mine the preliminary features and extract more abstract and representative deep features. These features can capture the subtle structure and texture information in the image. The output R of the residual block l It can be expressed as: R l =F l +H(F l ), where H(F l ) is the feature after being processed by some convolutional layers;
[0034] The attention mechanism is used to automatically focus on important areas and features in the image, enhance the ability to extract key information, and improve the effectiveness and pertinence of features. The attention weight A is calculated by the following formula: A = σ(W a *F+b a ), where W a and b a are the parameters of the attention mechanism.
[0035] Furthermore, the specific execution steps of the deep learning model stage in the super-resolution reconstruction module include:
[0036] Design a super-resolution reconstruction model based on the generative adversarial network GAN, where the generator G is responsible for generating high-resolution images I HR, the discriminator D is used to determine whether the generated image is real. The output of the generator can be expressed as: I HR =G(I LR ), the output of the discriminator is a probability value, indicating the possibility that the input image is a true high-resolution image, which can be expressed as: P = D (I HR );
[0037] In the generator, densely connected convolutional layers and deconvolutional layers are used to strengthen feature propagation and reuse, thereby improving the learning efficiency and reconstruction performance of the model.
[0038] The discriminator adopts a multi-layer convolution structure, combined with batch normalization and Leaky ReLU activation function, to enhance the ability to distinguish true and false high-resolution images, thereby guiding the generator to generate more realistic high-resolution images. The Leaky ReLU function can be expressed as: Here, α is a constant less than 1.
[0039] Furthermore, the specific execution steps of the model training phase in the super-resolution reconstruction module include:
[0040] Prepare large-scale, high-resolution remote sensing images and its corresponding downsampled low-resolution image I LR As a training data set, and perform data enhancement operations;
[0041] Define the loss function, including content loss and adversarial loss, and the mean square error can be expressed as: The calculation of the structural similarity index is relatively complex, usually involving the mean, variance, and covariance statistics of the image. The adversarial loss is usually calculated based on the output probability of the discriminator.
[0042] The adaptive moment estimation Adam optimization algorithm is used to train the model. By dynamically adjusting the learning rate, the convergence speed of the model is accelerated while avoiding falling into the local optimal solution.
[0043] Furthermore, the specific execution steps of the image reconstruction stage in the super-resolution reconstruction module include:
[0044] The low-resolution remote sensing data after preprocessing and fusion is LR Input the trained super-resolution reconstruction model;
[0045] The model generates high-resolution predicted images through the generator And through the feedback adjustment of the discriminator, the output results are continuously optimized, and finally a high-quality, high-resolution reconstructed image is obtained.
[0046] Furthermore, the specific execution steps of the settlement monitoring and analysis module include:
[0047] The high-resolution image data output by the super-resolution reconstruction module is used as input;
[0048] The small baseline set InSAR technology is used to calculate the interferometric phase. Assuming that the complex values of the two SAR images are S1 and S2 respectively, the interferometric phase can be expressed as: Where * indicates complex conjugation;
[0049] The real deformation information d is obtained through the phase unwrapping algorithm of the minimum cost flow;
[0050] Using the ARIMA model, assuming that the time series data is y t , the model can be expressed as: Where c is a constant, φ i and θ j are model parameters, t is white noise;
[0051] Considering the urban geological structure, groundwater level changes and engineering construction factors, a comprehensive land subsidence prediction model is established. Based on multiple linear regression, it is assumed that the subsidence amount y is related to the factors x1, x2,…, x n The relationship is: Where β0 is the intercept, β i is the regression coefficient;
[0052] Analyze the impact of subsidence on urban infrastructure and buildings.
[0053] Furthermore, the specific execution steps of the result output module include:
[0054] Arrange and classify the result data obtained by the settlement monitoring and analysis module;
[0055] Draw a ground subsidence rate graph, assuming that the subsidence rate data for a certain area is v i , discrete data points are connected into continuous curves and surfaces through interpolation algorithms;
[0056] Draw a settlement contour map. For the settlement data d i , using the contour generation algorithm, according to the set contour interval, contour lines are generated to represent areas with the same amount of settlement;
[0057] The results of the summary analysis, including statistical data, are described in text form.
[0058] Compared with the existing technology, the urban land subsidence monitoring system fused with multi-source remote sensing data has the following advantages:
[0059] Beneficial effects:
[0060] The present invention adopts super-resolution reconstruction technology to greatly improve the spatial resolution of data without increasing the cost of data acquisition, thereby providing the possibility for refined urban ground subsidence monitoring and analysis, significantly improving the accuracy and sensitivity of urban ground subsidence monitoring, and being able to keenly capture more subtle settlement changes, providing key data support for safety assessment and early warning of urban infrastructure, effectively integrating multi-source remote sensing data, giving full play to the advantages of different remote sensing technologies, overcoming the limitations and uncertainties of a single data source, and thus improving the reliability and comprehensiveness of monitoring results.
[0061] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] 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 prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0063] Figure 1 The schematic diagram is a structural diagram of an urban land subsidence monitoring system based on multi-source remote sensing data fusion. DETAILED DESCRIPTION
[0064] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. 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.
[0065] Embodiment 1
[0066] An urban land subsidence monitoring system based on multi-source remote sensing data fusion, the system comprises:
[0067] Data acquisition module, used to collect multiple remote sensing data sources from different sensors and platforms;
[0068] Widely collect a variety of remote sensing data sources from different sensors and platforms, including but not limited to high-resolution optical remote sensing satellites (such as WorldView series, Gaofen series, etc.), synthetic aperture radar (SAR) remote sensing satellites (such as Sentinel-1 series, TerraSAR-X, etc.), airborne LiDAR, etc. At the same time, ensure that the time series of data collection has a certain degree of continuity and periodicity to meet the dynamic needs of ground subsidence monitoring.
[0069] The specific execution steps of the data acquisition module include: widely covering various types of remote sensing data sources, including but not limited to high-resolution optical remote sensing satellites (such as WorldView series, Gaofen series, etc.), synthetic aperture radar (SAR) remote sensing satellites (such as Sentinel-1 series, TerraSAR-X, etc.) and airborne LiDAR; setting the collection time interval, spatial range and band selection parameters according to monitoring requirements and data source characteristics to ensure the comprehensiveness and effectiveness of the data;
[0070] According to the set parameters, data is collected through sensors and platforms to ensure that the time series of data collection has a certain degree of continuity and periodicity to meet the dynamic needs of ground subsidence monitoring; the collected data is preliminarily screened to remove data with obvious quality problems and data that does not meet the preset requirements; the collected valid data is securely stored and transmitted to the data processing center through reliable communication methods to prepare for subsequent preprocessing and analysis.
[0071] The data preprocessing module is used to perform preprocessing operations on the collected multi-source remote sensing data to eliminate and reduce errors and inconsistencies in the data;
[0072] The specific execution steps of the data preprocessing module include: first, using the calibration parameters (such as gain and offset) provided by the sensor to convert the original digital count DN into radiation brightness value and reflectivity, L = DN × G + O, where L is the radiation brightness, DN is the original digital count value, G represents gain, and O represents offset. Then, the atmospheric radiation transfer model (such as MODTRAN, 6S, etc.) and ground observation data (such as aerosol optical thickness, water vapor content, etc.) are used to estimate and remove the impact of the atmosphere on radiation. Where ρ is the surface reflectivity, L is the radiation brightness received by the sensor, and L a is the downward radiance of the atmosphere, E0 is the solar irradiance, θ s is the solar zenith angle, τ is the atmospheric transmittance;
[0073] Select multiple corresponding control points on the remote sensing image and reference map (such as topographic map or higher precision satellite image); use polynomial model (such as affine transformation, quadratic or higher order polynomial) to perform least squares fitting on the control points, and establish the transformation relationship between image coordinates and geographic coordinates, x′=a0+a1x+a2y+a3xy, y′=b0+b1x+b2y+b3xy, where (x′, y′) is the geographic coordinate, (x, y) is the image coordinate, a i and b i are polynomial coefficients; according to the established geometric model, the original image is resampled to generate a corrected image;
[0074] The median filtering method is used to remove noise by taking the median value of its neighboring pixels as the new value for each pixel in the image. Assume that the neighboring pixel values are p1, p2, ..., p n , then the pixel value p after median filtering filtered For: p filtered =median(p1,p2,…,p n ).
[0075] Embodiment 2
[0076] Data fusion module, used to organically fuse pre-processed multi-source remote sensing data;
[0077] The pre-processed multi-source remote sensing data are organically fused by using fusion algorithms based on pixel level, feature level and decision level. For example, for optical remote sensing data and SAR remote sensing data, their edge, texture and other features can be extracted first, and then feature-level fusion can be performed; for similar remote sensing data in different phases, decision-level fusion can be performed based on statistical analysis and decision rules. Through fusion, the advantages of different data sources are fully integrated, such as the rich spectral information of optical remote sensing and the high-precision deformation information of SAR remote sensing, so as to obtain more comprehensive and accurate surface characteristics and change information.
[0078] The specific execution steps of the data fusion module include: taking the preprocessed multi-source remote sensing data as input, the multi-source remote sensing data includes optical remote sensing data and SAR remote sensing data; for the optical remote sensing data and SAR remote sensing data, extracting their features respectively, including the spectral feature F of the optical remote sensing data opt and the texture feature F of SAR remote sensing data sar , assuming that the optical remote sensing data is I opt , through some feature extraction function f opt Extract spectral features: F opt =f opt (I opt ), for SAR remote sensing data, F sar =f sar (Isar );
[0079] Calculate the similarity and correlation between the extracted features, using cosine similarity to measure. For two feature vectors F1 and F2, the calculation formula of cosine similarity S is: According to the similarity and correlation, the features are weighted and fused to obtain the fused feature F. fusion : F fusion =w1F1+w2F2, where w1 and w2 are weights calculated based on similarity;
[0080] For the same type of remote sensing data in different phases, the fusion is performed based on statistical analysis and decision rules, and the mean statistics of multiple phase data are calculated. Suppose the phase data D1, D2, ..., D n , the result of mean fusion D mean for:
[0081] Embodiment 3
[0082] The super-resolution reconstruction module uses the super-resolution reconstruction method to perform super-resolution reconstruction on the fused data to improve the spatial resolution of the data. The specific execution steps of the super-resolution reconstruction method include: feature extraction stage, deep learning model construction stage, model training stage and image reconstruction stage;
[0083] The specific steps of the feature extraction stage include: using a multi-layer convolutional neural network CNN to input a low-resolution image I LR Perform preliminary feature extraction to obtain shallow basic features, including edges and colors. Assume that the output feature map of the first convolution operation is F l , the calculation formula is: F l =σ(W l *F l-1 +b l ), where W l is the convolution kernel weight of layer l, b l is the bias term, σ is the activation function (such as ReLU function), and * represents the convolution operation; the residual network structure is introduced to deeply mine the preliminary features and extract more abstract and representative deep features. These features can capture the subtle structure and texture information in the image. The output R of the residual block l It can be expressed as: R l =F l +H(F l ), where H(F l) is the feature after some convolutional layer processing; the attention mechanism is used to automatically focus on the important areas and features in the image, enhance the ability to extract key information, and improve the effectiveness and pertinence of the features. The attention weight A is calculated by the following formula: A = σ(W a *F+b a ), where W a and b a are the parameters of the attention mechanism.
[0084] The specific execution steps of the deep learning model stage include: designing a super-resolution reconstruction model based on the generative adversarial network GAN, in which the generator G is responsible for generating high-resolution images I HR , the discriminator D is used to determine whether the generated image is real. The output of the generator can be expressed as: I HR =G(I LR ), the output of the discriminator is a probability value, indicating the possibility that the input image is a true high-resolution image, which can be expressed as: P = D (I HR ); In the generator, densely connected convolutional layers and deconvolutional layers are used to strengthen the propagation and reuse of features, improve the learning efficiency and reconstruction performance of the model; the discriminator adopts a multi-layer convolution structure, combined with batch normalization (Batch Normalization) and Leaky ReLU activation function, to enhance the ability to distinguish between true and false high-resolution images, thereby guiding the generator to generate more realistic high-resolution images. The Leaky ReLU function can be expressed as: Here, α is a constant less than 1.
[0085] The specific steps of the model training phase include: preparing large-scale high-resolution remote sensing images and its corresponding downsampled low-resolution image I LR As a training data set, data enhancement operations are performed, including rotation, flipping, and scaling, to increase the diversity and robustness of the data; the loss function is defined, including content loss (such as mean square error, structural similarity index, etc.) and adversarial loss. The mean square error can be expressed as: The calculation of the structural similarity index is relatively complex, usually involving the mean, variance and covariance statistics of the image. The adversarial loss is usually calculated based on the output probability of the discriminator. The adaptive moment estimation Adam optimization algorithm is used to train the model, and the learning rate is dynamically adjusted to accelerate the convergence speed of the model while avoiding falling into the local optimal solution.
[0086] The specific steps of the image reconstruction phase include: LR Input the trained super-resolution reconstruction model; the model generates a high-resolution predicted image through the generator And through the feedback adjustment of the discriminator, the output results are continuously optimized, and finally a high-quality, high-resolution reconstructed image is obtained.
[0087] Embodiment 4
[0088] The subsidence monitoring and analysis module is used to analyze the spatiotemporal characteristics and development trends of ground subsidence based on super-resolution reconstructed data;
[0089] Based on the super-resolution reconstructed data, a variety of advanced monitoring algorithms and models are used to analyze ground subsidence. For example, the small baseline set (SBAS) InSAR technology and the permanent scatterer (PS) InSAR technology are used to accurately measure the tiny deformation of the ground; combined with time series analysis methods such as the ARIMA model and Kalman filtering, the temporal and spatial evolution laws and development trends of ground subsidence are revealed. At the same time, considering the influence of various factors such as urban geological structure, groundwater level changes, and engineering construction, a comprehensive ground subsidence prediction model is established to provide a scientific basis for urban planning and disaster prevention.
[0090] The specific execution steps of the subsidence monitoring and analysis module include: taking the high-resolution image data output by the super-resolution reconstruction module as input; using the small baseline set (SBAS) InSAR technology to calculate the interference phase. Assuming that the complex values of the two SAR images are S1 and S2 respectively, the interference phase can be expressed as: Where * represents complex conjugate; the real deformation information d is obtained by the phase unwrapping algorithm of the minimum cost flow;
[0091] Using the ARIMA model (autoregressive moving average model), assuming that the time series data is y t , the model can be expressed as: Where c is a constant, φ i and θ j are model parameters, t is white noise; considering the urban geological structure, groundwater level changes and engineering construction factors, a comprehensive land subsidence prediction model is established. Based on multiple linear regression, it is assumed that the settlement amount y is related to the factors x1, x2,…, x n The relationship is: Where β0 is the intercept, β i is the regression coefficient; analyzing the impact of settlement on urban infrastructure and buildings may involve mechanical models and risk assessment algorithms.
[0092] The result output module is used to output the monitoring and analysis results in a visual form.
[0093] The results of subsidence monitoring and analysis are output in a diversified and visual form. The output content includes but is not limited to ground subsidence rate maps, subsidence contour maps, subsidence trend prediction maps, subsidence impact assessment reports, etc. The output forms include two-dimensional and three-dimensional charts, maps, and dynamic demonstration multimedia files. By intuitively and clearly displaying the distribution and changes of ground subsidence, it provides decision-making support for relevant personnel such as urban managers, planning departments, and geological experts, so that effective prevention and control measures can be taken in a timely manner to reduce the risks and losses caused by ground subsidence.
[0094] The specific execution steps of the result output module include: arranging and classifying the result data obtained by the settlement monitoring and analysis module, for example, storing the settlement rate, settlement amount and other data of different areas separately;
[0095] Draw a ground subsidence rate graph, assuming that the subsidence rate data for a certain area is v i (indicates different positions), and connects discrete data points into continuous curves and surfaces through interpolation algorithms (such as linear interpolation or spline interpolation) to intuitively display the spatial distribution of sedimentation rate; draws sedimentation contour maps, and for sedimentation data d i , using an isoline generation algorithm (such as the MarchingSquares algorithm), based on the set isoline interval, generate isolines to represent areas with the same amount of settlement;
[0096] Summarize the analysis results, including statistical data (such as average sedimentation amount, maximum sedimentation rate, etc.), and describe them in text form. Natural language processing algorithms may be used to summarize and describe the results, such as calculating text similarity, keyword extraction, etc. If it is necessary to generate multimedia files for dynamic presentation, animation production algorithms may be used to display the sedimentation results at different times in a dynamic form.
[0097] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the same elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. An urban land subsidence monitoring system based on multi-source remote sensing data fusion, characterized in that: The system includes: Data acquisition module, used to collect multiple remote sensing data sources from different sensors and platforms; The data preprocessing module is used to perform preprocessing operations on the collected multi-source remote sensing data to eliminate and reduce errors and inconsistencies in the data; Data fusion module, used to organically fuse pre-processed multi-source remote sensing data; The super-resolution reconstruction module uses the super-resolution reconstruction method to perform super-resolution reconstruction on the fused data to improve the spatial resolution of the data. The specific execution steps of the super-resolution reconstruction method include: feature extraction stage, deep learning model construction stage, model training stage and image reconstruction stage; The subsidence monitoring and analysis module is used to analyze the spatiotemporal characteristics and development trends of ground subsidence based on super-resolution reconstructed data; The result output module is used to output the monitoring and analysis results in a visual form.
2. The urban land subsidence monitoring system based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the data acquisition module include: Covers a wide range of remote sensing data sources, including but not limited to high-resolution optical remote sensing satellites, synthetic aperture radar remote sensing satellites, and airborne LiDAR; According to monitoring requirements and data source characteristics, set the collection time interval, spatial range and band selection parameters; Collect data through sensors and platforms according to the set parameters; Conduct a preliminary screening of the collected data to remove data with obvious quality issues and data that does not meet the preset requirements; The collected valid data will be stored securely and transmitted to the data processing center.
3. The urban land subsidence monitoring system based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the data preprocessing module include: First, the original digital count DN is converted into radiance value and reflectance using the calibration parameters provided by the sensor: L = DN × G + O, where L is the radiance, DN is the original digital count value, G represents the gain, and O represents the offset. Next, the atmospheric radiation transfer model and ground observation data are used to estimate and remove the influence of the atmosphere on radiation. Where ρ is the surface reflectivity, L is the radiation brightness received by the sensor, and L a is the downward radiance of the atmosphere, E0 is the solar irradiance, θ s is the solar zenith angle, τ is the atmospheric transmittance; Select multiple corresponding control points on the remote sensing image and the reference map; The polynomial model is used to perform least squares fitting on the control points to establish the conversion relationship between image coordinates and geographic coordinates: x′=a0+a1x+a2y+a3xy, y′=b0+b1x+b2y+b3xy, where (x′, y′) is the geographic coordinate, (x, y) is the image coordinate, and a i and b i are the polynomial coefficients; According to the established geometric model, the original image is resampled to generate a corrected image; The median filtering method is used to remove noise by taking the median value of its neighboring pixels as the new value for each pixel in the image. Assume that the neighboring pixel values are p1, p2, ..., p n , then the pixel value p after median filtering filtered For: p filtered =median(p1,p2,…,p n ).
4. The urban land subsidence monitoring system based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the data fusion module include: The preprocessed multi-source remote sensing data is used as input, and the multi-source remote sensing data includes optical remote sensing data and SAR remote sensing data; For optical remote sensing data and SAR remote sensing data, their features are extracted respectively, including the spectral feature F of optical remote sensing data opt and the texture feature F of SAR remote sensing data sar , assuming that the optical remote sensing data is I opt , through some feature extraction function f opt Extract spectral features: F opt =f opt (I opt ), for SAR remote sensing data, F sar =f sar (I sar ); Calculate the similarity and correlation between the extracted features, using cosine similarity to measure. For two feature vectors F1 and F2, the calculation formula of cosine similarity S is: According to the similarity and correlation, the features are weighted and fused to obtain the fused feature F. fusion : F fusion =w1F1+w2F2, where w1 and w2 are weights calculated based on similarity; For the same type of remote sensing data in different phases, the fusion is performed based on statistical analysis and decision rules, and the mean statistics of multiple phase data are calculated. Suppose the phase data D1, D2, ..., D n , the result of mean fusion D mean for:
5. The urban land subsidence monitoring system based on multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the feature extraction stage in the super-resolution reconstruction module include: Using a multi-layer convolutional neural network CNN to input a low-resolution image I LR Perform preliminary feature extraction to obtain shallow basic features, including edges and colors. Assume that the output feature map of the first convolution operation is F l , the calculation formula is: F l =σ(W l *F l-1 +b l ), where W l is the convolution kernel weight of layer l, b l is the bias term, σ is the activation function, and * represents the convolution operation; The residual network structure is introduced to deeply mine the preliminary features and extract more abstract and representative deep features. These features can capture the subtle structure and texture information in the image. The output R of the residual block l It can be expressed as: R l =F l +H(F l ), where H(F l ) is the feature after being processed by some convolutional layers; The attention mechanism is used to automatically focus on important areas and features in the image, enhance the ability to extract key information, and improve the effectiveness and pertinence of features. The attention weight A is calculated by the following formula: A = σ(W a *F+b a ), where W a and b a are the parameters of the attention mechanism.
6. The urban land subsidence monitoring system of multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the deep learning model stage in the super-resolution reconstruction module include: Design a super-resolution reconstruction model based on the generative adversarial network GAN, where the generator G is responsible for generating high-resolution images I HR , the discriminator D is used to determine whether the generated image is real. The output of the generator can be expressed as: I HR =G(I LR ), the output of the discriminator is a probability value, indicating the possibility that the input image is a true high-resolution image, which can be expressed as: P = D (I HR ); In the generator, densely connected convolutional layers and deconvolutional layers are used to strengthen feature propagation and reuse, thereby improving the learning efficiency and reconstruction performance of the model. The discriminator adopts a multi-layer convolution structure, combined with batch normalization and Leaky ReLU activation function, to enhance the ability to distinguish true and false high-resolution images, thereby guiding the generator to generate more realistic high-resolution images. The Leaky ReLU function can be expressed as: Here, α is a constant less than 1.
7. The urban land subsidence monitoring system of multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the model training phase in the super-resolution reconstruction module include: Prepare large-scale, high-resolution remote sensing images and its corresponding downsampled low-resolution image I LR As a training data set, and perform data enhancement operations; Define the loss function, including content loss and adversarial loss, and the mean square error can be expressed as: The calculation of the structural similarity index is relatively complex, usually involving the mean, variance, and covariance statistics of the image. The adversarial loss is usually calculated based on the output probability of the discriminator. The adaptive moment estimation Adam optimization algorithm is used to train the model. By dynamically adjusting the learning rate, the convergence speed of the model is accelerated while avoiding falling into the local optimal solution.
8. The urban land subsidence monitoring system of multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the image reconstruction stage in the super-resolution reconstruction module include: The low-resolution remote sensing data after preprocessing and fusion is LR Input the trained super-resolution reconstruction model; The model generates high-resolution predicted images through the generator And through the feedback adjustment of the discriminator, the output results are continuously optimized, and finally a high-quality, high-resolution reconstructed image is obtained.
9. The urban land subsidence monitoring system of multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the settlement monitoring and analysis module include: The high-resolution image data output by the super-resolution reconstruction module is used as input; The small baseline set InSAR technology is used to calculate the interferometric phase. Assuming that the complex values of the two SAR images are S1 and S2 respectively, the interferometric phase can be expressed as: Where * indicates complex conjugation; The real deformation information d is obtained through the phase unwrapping algorithm of the minimum cost flow; Using the ARIMA model, assuming that the time series data is y t , the model can be expressed as: Where c is a constant, φ i and θ j is the model parameter, t is white noise; Considering the urban geological structure, groundwater level changes and engineering construction factors, a comprehensive land subsidence prediction model is established. Based on multiple linear regression, it is assumed that the subsidence amount y is related to the factors x1, x2,…, x n The relationship is: Where β0 is the intercept, β i is the regression coefficient; Analyze the impact of subsidence on urban infrastructure and buildings.
10. The urban land subsidence monitoring system of multi-source remote sensing data fusion according to claim 1 is characterized in that: The specific execution steps of the result output module include: Arrange and classify the result data obtained by the settlement monitoring and analysis module; Draw a ground subsidence rate graph, assuming that the subsidence rate data for a certain area is v i , discrete data points are connected into continuous curves and surfaces through interpolation algorithms; Draw a settlement contour map. For the settlement data d i , using the contour generation algorithm, according to the set contour interval, contour lines are generated to represent areas with the same amount of settlement; The results of the summary analysis, including statistical data, are described in text form.
Citation Information
Cited By
Geological park geological relic information acquisition method and system based on multi-source data
CN120107635A
Multi-source data fusion-based urban land subsidence analysis method and system
CN120338292A
Full-focusing super-resolution imaging method based on deconvolution
CN120355809A
Coal mine subsidence monitoring method based on multi-source data fusion
CN120558173A
Farmland monitoring method and system based on remote sensing technology
CN120564062A
