Three gorges reservoir area ground surface settlement monitoring method, system and device based on multi-source remote sensing data

Abstract

The invention belongs to the field of remote sensing technology and environment monitoring, and particularly relates to a three gorges reservoir area ground surface settlement monitoring method, system and device based on multi-source remote sensing data, and the method comprises the steps: obtaining a plurality of single-view complex images and corresponding image surface coverage data in a period of time, and carrying out the preprocessing of the image data; carrying out SBAS-InSAR processing on the preprocessed image data, wherein the SBAS-InSAR processing comprises carrying out interference, inversion and geocoding processing on the image; fusing the coded data by adopting a multi-source remote sensing data fusion technology; inputting the fused image into a pre-trained machine learning model to obtain a three gorges reservoir area ground surface settlement monitoring result; by introducing multi-source remote sensing data fusion and an advanced machine learning model, the problems of single data source, difference in temporal-spatial resolution, poor model adaptability and the like in the prior art are solved, and the precision, the stability and the robustness of ground surface settlement monitoring and evaluation are remarkably improved.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology and environmental monitoring, and in particular relates to a method, system and device for monitoring surface subsidence in the Three Gorges reservoir area based on multi-source remote sensing data. Background Art

[0002] In the field of surface subsidence monitoring and assessment, especially for subsidence detection and prediction in complex geological environments, the application of multi-source data fusion technology and machine learning models is becoming more and more widespread. For subsidence monitoring in areas such as the Three Gorges Reservoir, recent research has mainly relied on remote sensing technology, especially the combination of synthetic aperture radar interferometry (SBAS-InSAR) and other remote sensing data sources, in order to more accurately monitor the dynamics of surface subsidence. However, the application of a single remote sensing data source in surface subsidence monitoring still faces problems such as insufficient accuracy, differences in temporal and spatial resolution, and data reliability.

[0003] In terms of multi-source data fusion technology, different types of remote sensing data, such as optical images and synthetic aperture radar images, are combined with environmental variables such as rainfall, surface water level, etc. to enhance the accuracy and reliability of monitoring results. In particular, through multi-source data fusion, the limitations of a single data source can be overcome and the perception of subsidence behavior in complex geological environments can be improved. For the specific area of ​​the Three Gorges Reservoir, studies have shown that combining SBAS-InSAR data with environmental data, such as water level changes and precipitation, can effectively reveal the impact of reservoir water storage on surface subsidence. However, due to differences in resolution, acquisition frequency and data quality of different remote sensing data sources, data fusion faces challenges. Fusion technology requires unified processing of heterogeneous data sources to solve the problem of inconsistent time scales and spatial resolutions, which has not yet been fully solved in existing technologies.

[0004] In terms of machine learning, a variety of machine learning methods are used to improve the prediction accuracy of surface subsidence, including traditional algorithms such as support vector regression (SVR), decision tree, random forest (RF), gradient boosting decision tree (GBDT), extreme random tree (ERT), and deep learning models such as long short-term memory network (LSTM). Especially in the Three Gorges Reservoir area where the prediction accuracy is high, the LSTM model and GBDT model show good prediction ability due to their good adaptability to time series data and the advantage of capturing patterns in long time series data. Studies have shown that when the LSTM model is applied, the temporal dynamic changes of surface subsidence around the reservoir can be captured more accurately, especially under the influence of environmental factors such as water level changes and precipitation. The LSTM model can reveal the complex evolution of surface subsidence better than traditional statistical methods and other machine learning models.

[0005] However, although machine learning technology has achieved remarkable results in surface subsidence prediction, there are still some problems and challenges. First, the performance of machine learning models has high requirements on the quality, quantity and processing methods of data. When the data noise is large or the amount of data is insufficient, the training effect and prediction accuracy of the model will be greatly reduced. Secondly, the existing machine learning models are usually trained for specific areas and data sets, lack sufficient universality, and are limited in their promotion and application in different geological environments. In addition, although deep learning models (such as LSTM) have strong time series prediction capabilities, their model training process is relatively complex and has high requirements for computing resources, which may lead to low efficiency when processing large-scale data.

[0006] In general, although the current surface subsidence monitoring and prediction technology has made some progress, there are still problems such as single data source, immature data fusion technology and insufficient generalization ability of machine learning models. These problems are particularly prominent in applications in special areas such as the Three Gorges Reservoir area. Therefore, a more efficient and accurate technical means is urgently needed to solve these limitations and improve the overall ability of subsidence monitoring and assessment. Summary of the invention

[0007] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a surface subsidence monitoring method for the Three Gorges Reservoir area based on multi-source remote sensing data, the method comprising: obtaining a plurality of single-view composite images and corresponding image surface coverage data within a period of time, and preprocessing the image data; performing SBAS-InSAR processing on the preprocessed image data, including interfering, inverting and geocoding the images; fusing the coded data using multi-source remote sensing data fusion technology; inputting the fused image into a pre-trained machine learning model to obtain the surface subsidence monitoring result of the Three Gorges Reservoir area.

[0008] A surface subsidence monitoring system for the Three Gorges Reservoir area based on multi-source remote sensing data, the system comprising: a data acquisition module, a preprocessing module, a coding module, a fusion module and a monitoring module;

[0009] The data acquisition module is used to acquire a plurality of single-view composite images and corresponding image surface coverage data within a period of time;

[0010] The preprocessing module is used to preprocess the image data;

[0011] The encoding module is used to perform interference, inversion and geocoding processing on the pre-processed image;

[0012] The fusion module is used to fuse the encoded data to obtain a fused image;

[0013] The monitoring module is used to monitor the surface subsidence of the fused image and obtain monitoring results.

[0014] To achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned methods for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data.

[0015] To achieve the above-mentioned purpose, the present invention also provides a Three Gorges Reservoir area surface subsidence monitoring device based on multi-source remote sensing data, comprising a processor and a memory; the memory is used to store computer programs; the processor is connected to the memory, and is used to execute the computer programs stored in the memory, so that the Three Gorges Reservoir area surface subsidence monitoring device based on multi-source remote sensing data executes any of the above-mentioned Three Gorges Reservoir area surface subsidence monitoring methods based on multi-source remote sensing data.

[0016] Beneficial effects of the present invention:

[0017] The present invention overcomes the problems of single data source, spatial and temporal resolution difference, poor model adaptability, etc. in the prior art by introducing multi-source remote sensing data fusion and advanced machine learning models, and significantly improves the accuracy, stability and robustness of surface subsidence monitoring and assessment. This technical solution can not only provide more accurate subsidence monitoring results, but also provide reliable technical support for subsidence prediction and assessment in other complex environments, and has broad application prospects and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a regional image of the present invention;

[0019] Figure 2 It is the overall flow chart of the present invention;

[0020] Figure 3 It is a deformation diagram of different positions of the region of the present invention;

[0021] Figure 4 is a cumulative deformation diagram at equal time intervals in the deformation region 1 of the present invention;

[0022] Figure 5 It is a ground change map of the deformation area 1 of the present invention;

[0023] Figure 6 is a periodic cumulative deformation diagram of the deformation region 2 of the present invention;

[0024] Figure 7 It is a ground change map of the deformation area 2 of the present invention;

[0025] Figure 8 is a periodic cumulative deformation diagram at equal time intervals in the deformation region 3 of the present invention;

[0026] Fig. 9 It is a ground change map of the deformation area 3 of the present invention;

[0027] Fig.10 is a performance evaluation diagram of each model of the present invention;

[0028] Fig.11 Surface deformation prediction results of three typical regions of the present invention;

[0029] Fig.12 It is a box plot comparison diagram of the observed and predicted deformation results of the three surface deformation zones of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The present invention aims to solve the problems of insufficient accuracy, immature data fusion technology and poor adaptability of machine learning models in complex geological environments in existing surface subsidence monitoring and assessment technologies. Although the existing technology has made certain progress through multi-source remote sensing data fusion and machine learning methods, the limitations of a single data source, the spatiotemporal inconsistency problems faced by data fusion, and the insufficient generalization ability of existing machine learning models in specific areas still restrict the accuracy and reliability of surface subsidence prediction. For complex areas such as the Three Gorges Reservoir area, the present invention improves the multi-source remote sensing data fusion method, integrates different types of remote sensing data (such as historical optical images and synthetic aperture radar images) and environmental variables (such as water level changes and precipitation), overcomes the challenges brought by data source heterogeneity, and improves data fusion accuracy. At the same time, combined with more advanced machine learning and deep learning models such as long short-term memory networks (LSTM) and gradient boosting decision trees (GBDT), the accuracy and universality of surface subsidence prediction are enhanced. The present invention aims to provide a more efficient and accurate technical solution to solve the limitations of the existing technology and improve the overall ability of surface subsidence monitoring and assessment.

[0032] A method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data, such as Figure 2As shown, the method includes: obtaining multiple single-view composite images and corresponding image surface coverage data within a period of time, and preprocessing the image data; performing SBAS-InSAR processing on the preprocessed image data, including interferometry, inversion and geocoding processing on the image; fusing the coded data using multi-source remote sensing data fusion technology; inputting the fused image into the pre-trained machine learning model to obtain the surface subsidence monitoring results of the Three Gorges Reservoir area.

[0033] The technical solution of the present invention first combines synthetic aperture radar (SAR) data with optical remote sensing images in terms of data acquisition. The main data include 36 single-view complex (SLC) images obtained from the ascending orbit of the Sentinel-1A satellite provided by NASA's Synthetic Aperture Radar Data Center. These images cover the period from January 1, 2020 to December 31, 2022, providing monthly surface coverage data of the Three Gorges Reservoir area and providing valuable time series information. In order to improve the accuracy of the analysis, the 30-meter resolution Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) is used. This data set plays a vital role in InSAR data processing and can perform differential InSAR processing and synthetic aperture radar offset tracking processing, thereby eliminating the influence of terrain phase. In order to fully understand the driving factors behind surface deformation, relevant environmental data were collected, including daily water level data of the Yangtze River from the Hubei Hydrology and Water Resources Center and daily rainfall data from the National Center for Environmental Information of the National Oceanic and Atmospheric Administration (NOAA). These environmental data sets can analyze the relationship between hydrological factors and surface deformation patterns.

[0034] In terms of data processing, the present invention performs strict preprocessing on the acquired data, including merging, clipping, extraction, denoising, and filtering, to improve the quality, consistency, and efficiency of the processing process of the data. By optimizing the original data, the interference of noise on model training is reduced, the stability and reliability of the data are improved, and more accurate input data is provided for the machine learning model. SAR images and accurate track files are converted into SLC format in SARscape. The data is processed using SBAS-InSAR technology, and the subsidence areas along the river are processed and identified through interference, inversion, geocoding and other processes in sequence. In order to make up for the temporal and spatial resolution differences and data reliability problems of a single remote sensing data source in the prior art, the present invention adopts multi-source remote sensing data fusion technology, and overcomes the limitations brought by a single data source by combining different types of remote sensing data (such as historical Google optical images and SAR images) with environmental variables (such as water level changes, precipitation, etc.). Especially in applications for special areas such as the Three Gorges Reservoir, combining SBAS-InSAR data with environmental data such as water level and precipitation can effectively reveal the impact of reservoir water storage and environmental changes on surface subsidence. This multi-source data fusion can fully make up for the shortcomings of a single data source in terms of accuracy, timeliness, etc., and provide more reliable basic data for subsequent settlement monitoring and evaluation.

[0035] In the selection of machine learning models, the present invention has verified the effects of various traditional machine learning algorithms through a large number of experiments, including support vector regression (SVR), extreme random trees (ERT), random forests (RF) and gradient boosting decision trees (GBDT), as well as deep learning models long short-term memory (LSTM). These models are applied to three areas with different geological characteristics: urban ground deformation along the river (Area1), road slope settlement along the river (Area2) and ancient landslide deformation in the reservoir area (Area3). For model training and testing, we divided the SBAS-InSAR-derived time series data (2020-2022) using a 7:3 ratio time segmentation strategy. However, although these traditional algorithms can capture the changing trend of surface subsidence to a certain extent, due to their relatively limited modeling capabilities for time series data, it is difficult to achieve the ideal level of prediction accuracy. Finally, six evaluation indicators are combined to comprehensively measure the model performance: root mean square error (RMSE), coefficient of determination (R 2 ), mean absolute error (MAE), Pearson correlation coefficient, mean absolute percentage error (MAPE) and symmetric mean absolute percentage error (SMAPE), as well as combining the characteristics of time series data and the geological environment characteristics of the Three Gorges Reservoir area, the present invention selects a combination of long short-term memory network (LSTM) and gradient boosting decision tree (GBDT) models.

[0036] In this embodiment, by adopting multi-source remote sensing data fusion technology, combining synthetic aperture radar (SAR) data with optical remote sensing images, the present invention can effectively overcome the shortcomings of a single remote sensing data source in terms of accuracy, spatiotemporal resolution and data reliability. Through SBAS-InSAR technology processing and statistical analysis, the average deformation rate in urban areas is -3.48±2.91mm / yr (mean±standard deviation); the average deformation rate of road slopes is -5.19±3.62mm / yr; and the average deformation rate in ancient landslide areas is -6.02±4.55mm / yr. In the complex geological environment of the Three Gorges Reservoir area, the fusion of SBAS-InSAR technology and environmental variables (such as precipitation, water level changes, etc.) makes the monitoring results of surface subsidence more accurate and reliable, and can truly reflect the comprehensive impact of reservoir water storage and environmental changes on surface subsidence.

[0037] In this embodiment, the machine learning models are random forest (RF), extreme random tree (ERT), gradient boosted decision tree (GBDT), support vector regression (SVR) and long short-term memory (LSTM) algorithm models, which are used to train and evaluate the data processed by SBAS-InSAR. Six evaluation indicators are used to comprehensively measure the model performance: root mean square error (RMSE), determination coefficient (R 2), mean absolute error (MAE), Pearson correlation coefficient, mean absolute percentage error (MAPE), and symmetric mean absolute percentage error (SMAPE). These indicators were chosen because they can reflect the prediction accuracy and fit of the model from different perspectives. The models were applied to three regions with different geological characteristics: urban ground deformation along the river, road slope settlement along the river, and ancient landslide deformation in the reservoir area. For each region, all five models were trained using the same input features and target variables. The input features of the prediction models include: (1) time series data consisting of historical deformation measurements every 24 days from Sentinel-1 InSAR observations; (2) normalized differential deformation rates calculated using continuous measurements to capture the rate of change; and (3) contextual spatial information is retained by maintaining pixel relationships in the InSAR data matrix. For model training, a temporary sliding window of 5 time steps is used to predict the deformation of the next time step. The dataset is split in a 7:3 ratio, where the first 70% of the chronological period (approximately January 2020 to March 2021) is used for training, and the remaining 30% (approximately April 2021 to December 2022) is used for testing. To handle data imbalance and ensure computational efficiency, systematic sampling is adopted when the training sample exceeds 20,050 points. The hyperparameters of each model are optimized using grid search in the following ranges: for SVR (gamma = [0.1, 0.5, 1], C = [1, 10, 100], epsilon = [0.01, 0.05, 0.1]); for tree-based models (n_estimators = [50, 100, 150], max_depth = [5, 10, 15]); for LSTM

[0038] (hidden_units = [32, 64, 128], dropout_rate = [0.1, 0.2, 0.3]). This comprehensive approach is able to capture the temporal dynamics and spatial characteristics of the deformation process while maintaining model generalization. After training, these models were applied to the test set and the six evaluation metrics mentioned above were calculated. In addition, long-term predictions were performed for the study area based on the time series InSAR technique, using the best prediction model for each region.

[0039] In this embodiment, taking the long short-term memory and gradient boosting decision tree as examples, training the machine learning model includes: acquiring original image data within a period of time, preprocessing the image data, performing SBAS-InSAR processing on the preprocessed image, and using the SBAS-InSAR processed data as a training set; fusing the corresponding data in the training set, and using the fused data; inputting the fused data into the long short-term memory network for deep feature extraction; inputting the extracted features into the gradient boosting decision tree for monitoring to obtain monitoring results; evaluating the monitoring results using the evaluation results, adjusting the model parameters according to the evaluation results, and completing the model training when the evaluation results are optimal.

[0040] The present invention uses advanced machine learning technology, especially the combined model of long short-term memory network (LSTM) and gradient boosting decision tree (GBDT). Compared with traditional machine learning algorithms such as support vector regression (SVR) and extreme random tree (ERT), LSTM and GBDT can more effectively capture the temporal dynamic changes and complex nonlinear relationships of surface subsidence. The coefficient of determination (R) of the three regions of the LSTM model is 2 ) is between 0.45 and 0.75, and the coefficient of determination (R 2 ) is between 0.5 and 0.67, and the other indicators such as root mean square error (RMSE), determination coefficient (R 2 ), mean absolute error (MAE), Pearson correlation coefficient, mean absolute percentage error (MAPE), symmetric mean absolute percentage error (SMAPE), etc. are generally lower than other models. It can be seen that the combination of the two generally performs better than other models. When processing time series data, the LSTM model can more accurately capture the temporal variation of sedimentation, especially under the influence of environmental factors such as water level changes and precipitation, and can provide higher prediction accuracy. GBDT has significant advantages in processing multidimensional features and improving model performance, which can further enhance the stability and accuracy of predictions.

[0041] The technical solution of the present embodiment combines multi-source remote sensing data and environmental data, and through a reasonable pre-processing process, accurate input data is obtained to monitor and evaluate surface subsidence. The data is processed using SBAS-InSAR technology, and processes such as interference, inversion, and geocoding are successively processed and identified along the river subsidence area. In order to make up for the spatiotemporal resolution difference and data reliability problem of a single remote sensing data source in the prior art, the present invention adopts multi-source remote sensing data fusion technology, and by combining different types of remote sensing data (such as historical Google optical images and SAR images) with environmental variables (such as water level changes, precipitation, etc.), the limitations caused by a single data source are overcome. Especially in applications for special areas such as the Three Gorges Reservoir, combining SBAS-InSAR data with environmental data such as water level and precipitation, the impact of reservoir water storage and environmental changes on surface subsidence can be effectively revealed. This multi-source data fusion can fully make up for the shortcomings of a single data source in terms of accuracy, timeliness, etc., and provide more reliable basic data for subsequent subsidence monitoring and evaluation.

[0042] In this embodiment, long short-term memory (LSTM) and gradient boosting decision tree (GBDT) are selected as the main machine learning models, and model training and prediction are carried out on the processed data of three typical areas (urban areas along the river, road slopes along the river, and ancient landslides) in the Three Gorges Reservoir area.

[0043] Data partitioning: A 7:3 time splitting strategy is adopted to divide the SBAS-InSAR derived data from 2020 to 2022 into training and test sets.

[0044] Comparison of traditional machine learning models: In order to evaluate the effect of the LSTM and GBDT combination model, this embodiment is also compared with traditional algorithms such as support vector regression (SVR), extreme random trees (ERT) and random forest (RF).

[0045] Through the SBAS-InSAR processing, multi-source remote sensing data fusion and LSTM-GBDT hybrid model of this embodiment, the present invention realizes the monitoring and prediction of surface subsidence in different areas of the Three Gorges Reservoir area, and has achieved remarkable results:

[0046] After SBAS-InSAR processing and multi-source remote sensing data fusion:

[0047] Urban areas: The average deformation rate is -3.48±2.91mm / yr. By comparing 2013 ( Figure 5 c) and 2020 ( Figure 5d) Historical satellite images show that the subsidence areas are mainly concentrated in the past river wetlands, which may be related to infrastructure construction and river landfill. Analysis of the relationship between cumulative deformation, rainfall and reservoir water level shows that the subsidence rate significantly accelerated during periods of rapid water level decline (such as February-June 2020 and January-June 2021), increasing from -7mm / yr to -16mm / yr. This phenomenon may be due to the drop in groundwater levels caused by the drop in water levels, triggering stratum compaction and subsidence. However, the direct effect of rainfall on the subsidence rate is not obvious, and an increase in cumulative settlement is only observed after the rainy season.

[0048] Road slope area: The average deformation rate is -5.19±3.62mm / yr. Historical images show that the road in this area had not been expanded in 2013, but it has been significantly widened in 2020 ( Figure 7 c, d), it is speculated that the main cause of subsidence is the change in foundation structure and groundwater flow caused by road construction. Rainfall analysis shows that the rainfall before the two major subsidence events (May) in 2020 and 2021 was low (<5mm), and there was no obvious correlation with short-term rainfall. On the contrary, the reservoir water level dropped from 175m to 145m during the two events, and the rapid drop in water level was the main cause. The drought in 2022 caused the water level to be at 145-160m for a long time, which also explains the high subsidence state that year. Once the water level rises, the subsidence condition improves, which is different from the previous view that rising water levels will aggravate subsidence.

[0049] Ancient landslide area: The average deformation rate is -6.02±4.55mm / yr, showing a trend of continuous accelerated settlement. The annual average deformation rate increased from -5mm / yr before 2020 to more than -15mm / yr in 2022 ( Fig. 9 d). The settlement in the middle of the landslide is most obvious ( Fig. 9 a, b), which is speculated to be related to the increase in the frequency of human activities ( Fig. 9 c). The seasonal variation of rainfall is consistent with that of deformation rate. During the peak rainfall periods (such as July 2020, 2021 and 2022), the rainfall reached 5 mm, 5 mm and 12.5 mm, respectively, and the corresponding settlement rates were -5 mm / yr to -10 mm / yr, -10 mm / yr to -20 mm / yr and -17.5 mm / yr to -22.5 mm / yr, respectively, which are the maximum settlement rates of the whole year, indicating that rainfall is one of the important factors affecting ground deformation. In addition, the change of reservoir water level is basically consistent with rainfall and deformation rate, which also has an important impact on ground deformation.

[0050] In summary, the impact of reservoir water level adjustment on riverside cities, riverside road slope areas and ancient landslide behavior was analyzed. There is a significant negative correlation between the two, revealing the key role of reservoir water level adjustment in triggering and accelerating ground subsidence - the lower the reservoir water level, the faster the ground subsidence rate. The complex interaction between natural and human factors drives the subsidence, showing significant seasonal changes and spatial heterogeneity. Subsidence is also closely related to engineering projects such as road expansion. The cumulative deformation of ancient landslides shows an obvious periodic pattern, and there is a significant correlation between landslide activity and reservoir water storage process.

[0051] Machine Learning Model Evaluation:

[0052] LSTM model: In terms of prediction accuracy, the LSTM model performs best in terms of RMSE and SMAPE, and can better capture the dynamic changes in time series, especially under the influence of environmental factors such as water level changes and precipitation, and can provide higher prediction accuracy. The root mean square error (RMSE) of the three regions is between 4.68 and 7.52, and the coefficient of determination (R 2 ) is between 0.45 and 0.75, the mean absolute error (MAE) is between 1.88 and 4.58, the Pearson correlation coefficient is between 0.67 and 0.87, and the symmetric mean absolute percentage error (SMAPE) is between 58.8% and 79.27%.

[0053] GBDT model: The GBDT model performs well in multidimensional feature processing, especially in MAE and MAPE, which indicates that GBDT may have advantages in dealing with outliers and nonlinear relationships. The use of this model can further improve the stability and accuracy of prediction. 2 ) is between 0.50 and 0.67.

[0054] Error band plot: These regions are predicted using the best performing GBDT and LSTM models respectively.

[0055] In the ground deformation of cities along the river, the observed deformation gradually decreases from 0 mm to about -10 mm, while the predicted deformation decreases to -15 mm, showing a slight overestimation trend in the predicted data. The standard deviation increases in the later period of the study, indicating greater uncertainty in the prediction.

[0056] In the deformation of the road slope along the river, the observed deformation during the study period dropped from 0mm to -15mm, while the predicted data dropped to nearly -20mm. The standard deviation of the prediction increased significantly in the later period, reaching about 15mm, indicating that the model has a large uncertainty in the prediction of this area.

[0057] In the ancient landslide deformation in the reservoir area, the observed data remained relatively stable at around -10 mm, while the predicted deformation dropped to below -30 mm, and the predicted standard deviation reached 25 mm in the later stage of the prediction, which was much higher than the observed data, reflecting the most significant trend of overestimation of settlement in the ancient landslide deformation model in the reservoir area.

[0058] The present invention systematically demonstrates the design architecture and implementation process of the surface subsidence monitoring and prediction system based on multi-source remote sensing data and machine learning models. In the system architecture, the complete process from data acquisition to result analysis is demonstrated, including core components such as data acquisition part, preprocessing part, analysis and modeling module and visualization display part. Among them, the data acquisition part integrates remote sensing data, optical remote sensing images and environmental data (precipitation and water level changes, etc.) obtained by SBAS-InSAR technology to provide comprehensive data support; the preprocessing part completes merging, cropping, extraction, denoising and filtering to improve the quality, consistency and efficiency of the data processing process; the analysis and modeling part introduces machine learning algorithms such as LSTM deep learning and GBDT to achieve accurate prediction of surface subsidence through feature extraction and model optimization; the visualization display module graphically displays the spatial distribution and dynamic change law of subsidence, which is convenient for result interpretation.

[0059] The present invention designs a systematic chart scheme. Figure 1 and Figure 3 The location distribution of the study area and the detailed characteristics of three key subsidence areas are shown, including the subsidence trends and spatial distribution of urban ground along the river, road slopes and ancient landslides. Figures 4 to 9 The dynamic changes of settlement in typical areas are described, and combined with rainfall and water level changes, the relationship between environmental factors and surface settlement is revealed. Fig.10 The performance of several machine learning models (Random Forest RF, Gradient Boosting Decision Tree GBDT, etc.) and deep learning models (Long Short-Term Memory Network Model LSTM) were comprehensively compared, using indicators (Root Mean Square Error RMSE, Coefficient of Determination R 2 , mean absolute error MAE, Pearson correlation coefficient, mean absolute percentage error MAPE, and symmetric mean absolute percentage error SMAPE) can intuitively display the advantages and disadvantages of the model. Fig.11 and Fig.12 The dynamic comparison between observed and predicted values, as well as the statistical distribution of box plots, demonstrate the accuracy and application value of model prediction. These charts not only reflect the implementation effect of the technical solution, but also provide important references for subsequent research and application.

[0060] A surface subsidence monitoring system for the Three Gorges Reservoir area based on multi-source remote sensing data, the system comprising: a data acquisition module, a preprocessing module, a coding module, a fusion module and a monitoring module;

[0061] The data acquisition module is used to acquire a plurality of single-view composite images and corresponding image surface coverage data within a period of time;

[0062] The preprocessing module is used to preprocess the image data;

[0063] The encoding module is used to perform interference, inversion and geocoding processing on the pre-processed image;

[0064] The fusion module is used to fuse the encoded data to obtain a fused image;

[0065] The monitoring module is used to monitor the surface subsidence of the fused image and obtain monitoring results.

[0066] The system implementation of the present invention is the same as the method implementation.

[0067] In one embodiment of the present invention, the present invention also includes a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, any of the above-mentioned methods for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data is implemented.

[0068] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0069] A Three Gorges Reservoir area surface subsidence monitoring device based on multi-source remote sensing data comprises a processor and a memory; the memory is used to store computer programs; the processor is connected to the memory and is used to execute the computer programs stored in the memory, so that the Three Gorges Reservoir area surface subsidence monitoring device based on multi-source remote sensing data executes any of the above-mentioned Three Gorges Reservoir area surface subsidence monitoring methods based on multi-source remote sensing data.

[0070] Specifically, the memory includes: ROM, RAM, magnetic disk, USB flash drive, memory card or optical disk, etc., which can store program codes.

[0071] Preferably, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0072] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data, characterized in that: include: Acquire multiple single-view complex images and corresponding image surface coverage data within a period of time, and pre-process the image data; Perform SBAS-InSAR processing on the pre-processed image data, including interferometry, inversion and geocoding of the image; Multi-source remote sensing data fusion technology is used to fuse the encoded data; the fused image is input into the pre-trained machine learning model to obtain the surface subsidence monitoring results of the Three Gorges Reservoir area.

2. The method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data according to claim 1 is characterized in that: Preprocessing of image data includes merging, cropping, extracting, denoising and filtering.

3. The method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data according to claim 1 is characterized in that: Interference processing of images includes: aligning different images in space; performing coherence operation on the aligned images to obtain interference patterns; filtering and phase unwrapping processing on the interference patterns; The image inversion process includes: performing radiation calibration on the image, that is, converting the digital quantization value of the image into physical radiation quantity; performing atmospheric correction on the radiation calibrated image; inputting the atmospherically corrected image into the inversion model to obtain the inverted image; The geocoding process of an image includes: constructing an original coordinate system and a target geographic coordinate system; selecting control points on the image with known accurate geographic coordinates; establishing a conversion relationship between the image coordinates and geographic coordinates of the control points through a mathematical model, and solving the parameters in the model using optimization algorithms such as the least squares method to achieve the best fit for the conversion relationship; calculating all pixel points in the image according to the established coordinate conversion model, and converting their original pixel coordinates into corresponding geographic coordinates.

4. The method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data according to claim 1 is characterized in that: The image fusion technology using multi-source remote sensing data fusion technology includes: using the spatial registration method to align the synthetic aperture radar data and the optical remote sensing image to generate a settlement point map; generating a related line graph based on the reservoir water level and precipitation, and optimizing the settlement point map based on the line graph to improve the accuracy and timeliness of settlement monitoring.

5. The method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data according to claim 1 is characterized in that: The machine learning model consists of long short-term memory networks and gradient boosted decision trees.

6. The method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data according to claim 1 is characterized in that: Training the machine learning model includes: obtaining raw image data within a period of time, preprocessing the image data, performing SBAS-InSAR processing on the preprocessed images, and using the SBAS-InSAR processed data as a training set; fusing the corresponding data in the training set, and inputting the fused data into a long short-term memory network for deep feature extraction; inputting the extracted features into a gradient boosting decision tree for monitoring to obtain monitoring results; evaluating the monitoring results using the evaluation results, adjusting the model parameters according to the evaluation results, and completing the model training when the evaluation results are optimal.

7. A Three Gorges Reservoir Area Surface Subsidence Monitoring System Based on Multi-Source Remote Sensing Data, the system is used to execute the Three Gorges Reservoir Area Surface Subsidence Monitoring Method Based on Multi-Source Remote Sensing Data as claimed in any one of claims 1 to 6, characterized in that: include: Data acquisition module, preprocessing module, encoding module, fusion module and monitoring module; The data acquisition module is used to acquire a plurality of single-view composite images and corresponding image surface coverage data within a period of time; The preprocessing module is used to preprocess the image data; The encoding module is used to perform interference, inversion and geocoding processing on the pre-processed image; The fusion module is used to fuse the encoded data to obtain a fused image; The monitoring module is used to monitor the surface subsidence of the fused image and obtain monitoring results.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the method for monitoring surface subsidence in the Three Gorges Reservoir area based on multi-source remote sensing data as described in any one of claims 1 to 6.

9. A surface subsidence monitoring device for the Three Gorges Reservoir area based on multi-source remote sensing data, characterized in that: It comprises a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the Three Gorges Reservoir area surface subsidence monitoring device based on multi-source remote sensing data executes the Three Gorges Reservoir area surface subsidence monitoring method based on multi-source remote sensing data as described in any one of claims 1 to 6.

Images