Satellite image water resource dynamic monitoring method and system based on deep learning

Abstract

The application discloses a kind of satellite image water resource dynamic monitoring method and system based on deep learning.The water resource dynamic monitoring system of the application includes satellite image processing module, deep learning feature extraction module, differential geometry water body boundary representation module, curvature flow dynamic monitoring module, multi-scale evolution prediction module, multi-source data fusion module, decision support module and water body capacity estimation module, system uses differential geometry theory to model water body boundary as continuous manifold, the dynamic change of boundary is represented by curvature flow, and the evolution trend is predicted by using Ricci flow to construct multi-scale representation.The system can accurately represent the characteristics of water body boundary, achieve high-precision change detection, and the boundary positioning accuracy reaches sub-pixel level.It can be applied to watershed water resource assessment, urban water resource management, reservoir monitoring and disaster warning and other scenarios.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, specifically relating to a method and system for dynamic monitoring of water resources based on deep learning using satellite images. Background Technology

[0002] Water resources, as a fundamental resource for human survival and social development, directly impact regional economic development and ecological security through their distribution and changes. Traditional water resource monitoring methods primarily rely on manual field surveys, which are not only time-consuming and labor-intensive but also difficult to implement for large areas in real time. With the development of remote sensing technology, using satellite imagery for water resource monitoring has become an efficient and economical method.

[0003] Existing satellite imagery-based water resource monitoring technologies mainly include water body extraction methods based on spectral features and water body identification methods based on image classification. Spectral feature-based methods typically utilize the differences in reflectance of water bodies in different wavelength bands to identify water bodies, such as the Normalized Difference Water Index (NDWI). Image classification-based methods, on the other hand, use traditional machine learning algorithms, such as Support Vector Machines (SVM) and Random Forests (RF), to classify images and identify water body regions.

[0004] However, these methods have some limitations: on the one hand, spectral feature-based methods are easily affected by factors such as atmospheric conditions and water turbidity, leading to a decrease in recognition accuracy; on the other hand, traditional machine learning methods rely on human experience for feature extraction and selection, making it difficult to adapt to complex and ever-changing water features. Furthermore, existing methods mainly focus on static water body identification, lacking the ability to effectively monitor and predict dynamic changes in water resources.

[0005] With the development of deep learning technology, algorithms such as Convolutional Neural Networks (CNN) and U-Net have demonstrated powerful performance in image recognition and segmentation. Applying deep learning technology to satellite imagery for water resource monitoring holds promise for overcoming the limitations of traditional methods and improving the accuracy and efficiency of water body identification. However, a systematic and high-precision solution is still lacking in dynamic monitoring and change prediction of water resources.

[0006] Therefore, there is an urgent need for a satellite imagery-based dynamic water resources monitoring system that can combine deep learning technology and differential geometry theory to achieve accurate identification, change detection, and trend prediction of water resources, providing a scientific basis for water resources management and decision-making. Summary of the Invention

[0007] The purpose of this invention is to provide a method for dynamic monitoring of water resources based on deep learning using satellite images.

[0008] Another objective of this invention is to provide a satellite image dynamic monitoring system for water resources based on deep learning.

[0009] This invention is achieved through the following technical solution:

[0010] The deep learning-based satellite imagery-based dynamic water resource monitoring method of this invention includes:

[0011] Acquire multi-channel satellite images of the detection area and perform image denoising, band adjustment, and geometric correction to enhance image quality;

[0012] The input image is converted to a grayscale image, and features are extracted.

[0013] The extracted water body boundary is regarded as a two-dimensional manifold, the boundary curvature characteristics are calculated, and the boundary manifold representation is constructed.

[0014] Construct the optimal transport mapping between water body boundaries in different time phases, calculate the curvature field difference, determine the adaptive threshold, identify the changing region, and calculate the changing area and rate of change;

[0015] We construct a multi-scale representation of water body boundaries using Ricci flow, analyze boundary evolution patterns, and predict future water body boundary morphology.

[0016] By integrating satellite imagery, publicly available internet data, and public basic data, spatial distribution characteristics of water resources are extracted.

[0017] Estimate the water volume and assess the impact of water changes on the surrounding environment;

[0018] Risk assessment and early warning information are generated based on water body change prediction results and fused data.

[0019] The present invention describes acquiring multi-channel satellite images of the detection area and performing image denoising, band adjustment, and geometric correction to enhance image quality, specifically including:

[0020] The received satellite images are denoised to remove noise and interference: an improved nonlocal mean filtering algorithm is used to denoise the images by searching for similar blocks in the images and using the similarity between blocks to suppress noise.

[0021] Adjusting multi-channel satellite images to achieve multi-channel image fusion: enhancing water body characteristics by weighting and fusing different bands;

[0022] Using satellite image data, DEM image data, and a reference base map as input, correction control points are generated through coordinate matching, and new coordinates are established using bilinear interpolation. The satellite image data is then registered onto the reference base map: first, feature points are extracted from the satellite image, and then matched with corresponding points in the reference base map to establish a conversion model.

[0023] The present invention describes converting an input image into a grayscale image and performing feature extraction, specifically including:

[0024] The input image is converted into a grayscale image by weighted averaging of the RGB channels of the color image.

[0025] A two-dimensional Fourier transform is performed on the grayscale image to extract low-frequency, mid-frequency, and high-frequency features. The low-frequency features represent the image's average brightness, medium intensity, roughness, contrast, and histogram; the mid-frequency features represent the line texture and edge information in different directions; and the high-frequency features represent the image's detail features and noise.

[0026] The transformed grayscale image is normalized, Euclidean distance is calculated, and cluster centers are generated to obtain the feature map set of the image.

[0027] The present invention treats the extracted water body boundary as a two-dimensional manifold, calculates the boundary curvature features, and constructs a boundary manifold representation, specifically including:

[0028] A contour tracking algorithm is used to extract the set of boundary points from the water body mask; the water body mask is a binary image generated by grayscale image conversion and feature extraction, wherein white areas represent water bodies and black areas represent non-water bodies.

[0029] Parameterize the discrete boundary point set into a continuous curve;

[0030] Calculate local curvature on the boundary manifold to generate curvature distribution characteristics;

[0031] By constructing the discrete exterior differential form of the boundary, the connectivity and directionality of the boundary are preserved;

[0032] Construct a boundary representation triple containing a vertex set, a connection set, and a curvature value set.

[0033] The present invention treats the extracted water body boundary as a two-dimensional manifold, calculates the boundary curvature features, and constructs a boundary manifold representation, specifically as follows:

[0034] A contour tracing algorithm is used to extract the set of boundary points from the water body mask; the water body mask is a binary image generated in step 2, where white areas represent water bodies and black areas represent non-water bodies.

[0035] Parameterize the discrete boundary point set into a continuous curve;

[0036] Calculate local curvature on the boundary manifold to generate curvature distribution characteristics;

[0037] By constructing the discrete exterior differential form of the boundary, the connectivity and directionality of the boundary are preserved;

[0038] Construct a boundary representation triple containing a vertex set, a connection set, and a curvature value set.

[0039] The present invention describes the construction of optimal transport mapping between water body boundaries at different time phases, calculation of curvature field differences, determination of adaptive thresholds, identification of changing regions, and calculation of changing area and rate of change, specifically as follows:

[0040] Extracting water body boundary representations at different time phases: By processing satellite images acquired at different times, extracting water body boundary representation triplets for each time phase;

[0041] Construct the optimal transmission mapping between different temporal boundaries and establish point-to-point correspondences;

[0042] Calculate the curvature difference between corresponding points and generate a difference curvature field;

[0043] By analyzing the statistical characteristics of the differential curvature field, the change detection threshold is dynamically determined;

[0044] Locate the boundary segments with significant changes, and calculate the area and rate of change of the changes.

[0045] The present invention describes the construction of a multi-scale representation of water body boundaries using Ricci flow, analysis of boundary evolution patterns, and prediction of future water body boundary morphology. Specifically:

[0046] By processing multiple satellite images acquired over a period of time, water body boundary characteristics are extracted for each period to form a temporal boundary characterization sequence.

[0047] By constructing a discrete Ricci flow model, the continuous Ricci flow equation is discretized, generating curvature evolution equations applicable to water body boundaries;

[0048] By solving the discrete Ricci flow equation, the boundary representation under different flow parameters is obtained, the scale space of the boundary is constructed, and multi-scale geometric features are generated.

[0049] Calculate the feature spectrum of the boundary manifold and extract invariant geometric features;

[0050] By analyzing the temporal boundary characterization sequence and multi-scale geometric features, we can identify the evolution patterns of the boundaries and learn the laws of evolution dynamics.

[0051] Based on the identified evolution patterns and learned dynamics, the water boundary morphology at future moments is predicted, and the prediction uncertainty is assessed.

[0052] The present invention describes the fusion of satellite imagery, publicly available internet data, and public basic data to extract spatial distribution characteristics of water resources, specifically as follows:

[0053] Using land survey data, we analyze the spatial location and characteristic parameters of various types of water bodies; using meteorological satellite data, we calculate multi-source meteorological data and analyze the water body capacity and changing trends through spatiotemporal correlation; combining the environmental information around the water body, we establish buffer analysis on the GIS platform to assess the impact of the environment on water resources.

[0054] The present invention describes the estimation of water volume and the assessment of the impact of water changes on the surrounding environment, specifically as follows:

[0055] Based on DEM data, calculate the difference in elevation between the water level and the riverbed to estimate the water depth.

[0056] Calculate the water volume by combining the water depth and water area;

[0057] Perform water volume statistics on various types of water bodies;

[0058] The capacity is classified and the results of water resource capacity testing are output.

[0059] The deep learning-based satellite imagery-based dynamic water resources monitoring system of this invention includes:

[0060] The satellite image processing module is used to receive multi-channel satellite images, perform image denoising, band adjustment, and geometric correction to enhance image quality.

[0061] A deep learning feature extraction module is connected to the satellite image processing module and is used to extract features from the images processed by the satellite image processing module to form a feature map set.

[0062] The differential geometry water body boundary representation module is connected to the deep learning feature extraction module. It is used to treat the extracted water body boundary as a two-dimensional manifold, calculate the boundary curvature features, and construct the boundary manifold representation.

[0063] The curvature flow dynamic monitoring module is connected to the differential geometry water body boundary characterization module. It is used to construct the optimal transport mapping between water body boundaries in different time phases, calculate the curvature field difference, determine the adaptive threshold, identify the changing region, and output the changing area and rate of change.

[0064] The multi-scale evolution prediction module, connected to the curvature flow dynamic monitoring module, is used to construct a multi-scale representation of the Ricci flow of the water body boundary, analyze the boundary evolution pattern, and predict the future water body boundary morphology.

[0065] A multi-source data fusion module, connected to the curvature flow dynamic monitoring module, is used to fuse satellite images, publicly available Internet data, and public basic data to extract spatial distribution characteristics of water resources.

[0066] The decision support module, connected to the multi-scale evolution prediction module and the multi-source data fusion module, is used to generate risk assessment and early warning information based on the water body change prediction results and fused data.

[0067] The water volume estimation module, connected to the differential geometry water boundary characterization module and the curvature flow dynamic monitoring module, is used to estimate the water volume and assess the impact of water changes on the surrounding environment.

[0068] The satellite image processing module of the present invention includes:

[0069] The image denoising unit is used to denoise the received satellite images, removing noise and interference;

[0070] A band adjustment unit, connected to the image denoising unit, is used to adjust multi-channel satellite images to achieve multi-channel image fusion;

[0071] The geometric correction unit, connected to the band adjustment unit, is used to take satellite image data, DEM image data and reference base map as input, generate correction control points through coordinate matching, establish new coordinates using bilinear interpolation, and register the satellite image data onto the reference base map.

[0072] The differential geometry water body boundary characterization module of the present invention includes:

[0073] The boundary extraction unit is used to extract the initial set of water body boundary points from the water body mask;

[0074] A boundary parameterization unit, connected to the boundary extraction unit, is used to parameterize a discrete set of boundary points into a continuous curve.

[0075] The curvature calculation unit, connected to the boundary parameterization unit, is used to calculate the local curvature on the boundary manifold and generate curvature distribution features;

[0076] A topology preservation unit, connected to the curvature calculation unit, is used to maintain the connectivity and directionality of the boundary and construct a discrete exterior differential form;

[0077] A boundary descriptor generation unit, connected to the topology maintenance unit, is used to construct a boundary representation triplet containing a vertex set, a connection relation set, and a curvature value set.

[0078] The curvature flow dynamic monitoring module of the present invention includes:

[0079] Multi-temporal boundary extraction unit is used to extract water body boundary characteristics at different temporal phases;

[0080] The boundary mapping unit, connected to the multi-temporal boundary extraction unit, is used to construct the optimal transmission mapping between different temporal boundaries and establish a point-to-point correspondence.

[0081] The curvature difference analysis unit, connected to the boundary mapping unit, is used to calculate the curvature difference between corresponding points and generate a difference curvature field.

[0082] An adaptive threshold unit, connected to the curvature difference analysis unit, is used to dynamically calculate the change detection threshold based on the statistical characteristics of curvature differences.

[0083] The change region identification unit, connected to the adaptive threshold unit, is used to locate significantly changing boundary segments and calculate the change area and change rate.

[0084] The multi-scale evolution prediction module of this invention includes:

[0085] The time-series data processing unit is used to collect satellite images from multiple periods and extract time-series boundary characterization sequences.

[0086] A discrete Ricci flow unit, connected to the time-series data processing unit, is used to construct a discrete Ricci flow model and generate boundary curvature evolution equations.

[0087] A multi-scale representation unit, connected to the discrete Ricci flow unit, is used to construct the scale space of the boundary and generate multi-scale geometric features;

[0088] The boundary spectrum analysis unit, connected to the multi-scale representation unit, is used to calculate the feature spectrum of the boundary manifold and extract invariant geometric features;

[0089] An evolution pattern recognition unit, connected to the boundary spectrum analysis unit, is used to identify boundary evolution patterns and learn evolution dynamics.

[0090] The boundary prediction unit, connected to the evolution pattern recognition unit, is used to predict the boundary shape at future moments and assess the prediction uncertainty.

[0091] The multi-source data fusion module of the present invention includes:

[0092] The land use module is used to analyze the spatial location and characteristic parameters of various types of water bodies such as rivers, reservoirs, lakes, wetlands, and farmland using land surveying data;

[0093] The meteorological data module, connected to the land use module, is used to calculate multi-source meteorological data using meteorological satellite data, and analyze water volume and change trends through spatiotemporal correlation.

[0094] The environmental parameter module, connected to the meteorological data module, is used to combine information about the surrounding environment of the water body to establish a buffer analysis on the GIS platform and assess the impact of the environment on water resources.

[0095] The application scenarios of the water resources dynamic monitoring system described in this invention include:

[0096] Watershed water resources assessment includes acquiring satellite time-series images of various water bodies within the monitoring area, acquiring internet mapping data of the monitoring area, analyzing water body images, identifying rivers, lakes, reservoirs, wetlands, and farmland, generating change detection maps, acquiring land use and environmental parameter information, calculating the capacity of various water bodies and changes in the surrounding environment, and generating watershed water resources assessment layers.

[0097] Urban water resource management includes acquiring satellite time-series images of various types of water bodies in the monitoring area, combining them with publicly available internet databases and real-scene mapping data, analyzing changes in water bodies, analyzing changes in water bodies, outputting water body capacity, generating urban water resource management layers, and making plans and decisions for urban construction.

[0098] Compared with the prior art, the present invention has the following beneficial effects:

[0099] 1. By combining deep learning and differential geometry, the water body boundary is modeled as a continuous manifold, achieving accurate representation of the water body boundary. Compared with traditional pixel-level methods, the boundary positioning accuracy is improved to the sub-pixel level, thus improving the accuracy of water body identification.

[0100] 2. By adopting curvature flow theory to model water boundary changes as a continuous evolution process, subtle boundary changes can be captured, improving detection accuracy by 30% to 50%. In particular, the detection effect of changes in complex boundary areas is significantly better than that of traditional methods.

[0101] 3. The introduction of the Ricci flow multiscale analysis framework can simultaneously capture macroscopic and microscopic water body change patterns. The accuracy rate reaches over 85% in short-term forecasts (1-3 months) and remains around 75% in medium-term forecasts (3-6 months), providing forward-looking guidance for water resource management.

[0102] 4. A complete dynamic water resources monitoring technology system has been constructed, forming a closed loop from data acquisition, preprocessing, feature extraction, change detection, and trend prediction, realizing full-process automation of water resources monitoring and greatly improving monitoring efficiency.

[0103] 5. The system has broad application adaptability and can be used in various scenarios such as watershed water resource assessment, urban water resource management, reservoir monitoring and disaster early warning, and has significant socio-economic value. Attached Figure Description

[0104] Figure 1 The overall architecture diagram of the satellite image dynamic monitoring system for water resources based on deep learning in this invention;

[0105] Figure 2A schematic diagram of the satellite image processing module of this invention;

[0106] Figure 3 The processing flowchart of the deep learning feature extraction module of this invention;

[0107] Figure 4 A schematic diagram of the differential geometry water body boundary characterization module of this invention;

[0108] Figure 5 A schematic diagram of the curvature flow dynamic monitoring module of the present invention;

[0109] Figure 6 A schematic diagram of the structure of the multi-scale evolution prediction module of this invention;

[0110] Figure 7 A schematic diagram of the structure of the multi-source data fusion module of this invention;

[0111] Figure 8 The processing flowchart of the water volume estimation module of this invention. Detailed Implementation

[0112] The technical solution of the present invention will be further described in detail below through specific embodiments.

[0113] Example 1

[0114] Please refer to Figure 1-8 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0115] Reference Figure 1 The satellite image dynamic monitoring system for water resources based on deep learning provided by this invention includes a satellite image processing module 1, a deep learning feature extraction module 2, a differential geometric water body boundary characterization module 3, a curvature flow dynamic monitoring module 4, a multi-scale evolution prediction module 5, a multi-source data fusion module 6, a decision support module 7, and a water body capacity estimation module 8.

[0116] Satellite image processing module 1 receives multi-channel satellite images, performs image denoising, band adjustment, and geometric correction to enhance image quality and provide a high-quality data source for subsequent processing. Deep learning feature extraction module 2, connected to satellite image processing module 1, receives the processed images, extracts deep features, and forms a feature set. Differential geometry water body boundary representation module 3, connected to deep learning feature extraction module 2, treats the extracted water body boundary as a two-dimensional manifold, calculates boundary curvature features, and constructs a boundary manifold representation. Curvature flow dynamic monitoring module 4, connected to differential geometry water body boundary representation module 3, constructs the optimal transport mapping between water body boundaries at different time phases, calculates curvature field differences, determines adaptive thresholds, identifies changing regions, and outputs the changing area and rate of change. Multi-scale evolution prediction module 5, connected to curvature flow dynamic monitoring module 4, constructs a Ricci flow multi-scale representation of the water body boundary, analyzes boundary evolution patterns, and predicts future water body boundary morphology. Multi-source data fusion module 6, connected to curvature flow dynamic monitoring module 4, fuses satellite images, publicly available internet data, and public basic data to extract spatial distribution characteristics of water resources. The decision support module 7 is connected to the multi-scale evolution prediction module 5 and the multi-source data fusion module 6, and generates risk assessment and early warning information based on the water body change prediction results and fused data. The water body capacity estimation module 8 is connected to the differential geometry water body boundary characterization module 3 and the curvature flow dynamic monitoring module 4, and is used to estimate the water body capacity and assess the impact of water body changes on the surrounding environment.

[0117] Reference Figure 2 The satellite image processing module 1 includes an image denoising unit 11, a band adjustment unit 12, and a geometric correction unit 13.

[0118] The image denoising unit 11 is used to denoise the received satellite images, removing noise and interference. In practical applications, satellite images are often affected by factors such as atmospheric scattering and sensor noise, leading to a decrease in image quality. The image denoising unit 11 uses an improved nonlocal mean filtering algorithm to denoise the images. This algorithm suppresses noise by searching for similar blocks in the image and utilizing the similarity between blocks. Preferably, for water images, the search window size is set to 21×21 pixels, the similar block size is set to 7×7 pixels, and the filter intensity parameter h is set to 10. These parameter settings can effectively suppress noise while preserving the details of the water body boundaries.

[0119] The band adjustment unit 12 is connected to the image denoising unit 11 and is used to adjust multi-channel satellite images to achieve multi-channel image fusion. The band adjustment unit 12 enhances water features by weighting and fusing different bands. Preferably, for Sentinel-2 satellite images, the weight of the near-infrared band (B8) can be set to 0.4, the weight of the short-wave infrared band (B11) can be set to 0.3, and the weights of the visible light bands (B2, B3, B4) can be set to 0.1 respectively. This weighting configuration can effectively highlight water features.

[0120] The geometric correction unit 13 is connected to the band adjustment unit 12 and is used to take satellite image data, DEM image data, and a reference base map as input. It generates correction control points through coordinate matching, establishes new coordinates using bilinear interpolation, and registers the satellite image data onto the reference base map. In this invention, the geometric correction unit 13 first extracts feature points from the satellite image and then matches them with corresponding points in the reference base map to establish a conversion model. Preferably, a feature point matching method based on the RANSAC algorithm is used, which can effectively remove mismatched points and improve the geometric correction accuracy. When selecting control points, they should be evenly distributed across the image area, generally 30 to 50 control points are selected, and the control point registration accuracy is better than 0.5 pixels.

[0121] Reference Figure 3 The deep learning feature extraction module 2's processing includes three main steps:

[0122] First, the input image is converted to a grayscale image. This step is achieved by weighted averaging of the RGB channels of the color image. The conversion formula is as follows:

[0123] ;

[0124] in: The image is the converted grayscale image. R is the red channel value of the original image, G is the green channel value of the original image, and B is the blue channel value of the original image.

[0125] Secondly, a two-dimensional Fourier transform is performed on the grayscale image to extract its low-frequency, mid-frequency, and high-frequency features. The formula for calculating the two-dimensional Fourier transform is:

[0126] ,

[0127] in: For frequency domain images, For the spatial domain image, M is the image width, N is the image height, and u is the horizontal frequency variable, with a value range of [value missing]. v is the vertical frequency variable, and its value range is... , Here, is the complex exponential transform kernel, and j is the imaginary unit. By performing frequency separation on the transform result, low-frequency, mid-frequency, and high-frequency features of the image can be obtained. Low-frequency features represent overall information such as average brightness, medium density, roughness, contrast, and histogram of the image; mid-frequency features represent line texture and edge information in different directions of the image; and high-frequency features represent detailed features and noise of the image.

[0128] Finally, the transformed grayscale image is normalized, Euclidean distance is calculated, and cluster centers are generated to obtain the image's feature map set. Preferably, the Z-score normalization method is used, and the calculation formula is as follows:

[0129] ,

[0130] in: The normalized image in coordinates The value at that location, For the original image in coordinates The value at that location, The mean of the image is calculated using the following formula: , The standard deviation of the image is calculated using the following formula: M and N represent the width and height of the image, respectively. After normalization, cluster centers are generated using the K-means clustering algorithm by calculating the Euclidean distance between image patches, forming a feature map set. In practical applications, the number of clusters K can be set to 5-8, a range that achieves a good balance between representational power and computational efficiency.

[0131] Reference Figure 4 The differential geometry water body boundary characterization module 3 includes a boundary extraction unit 31, a boundary parameterization unit 32, a curvature calculation unit 33, a topology preservation unit 34, and a boundary descriptor generation unit 35.

[0132] Boundary extraction unit 31 is used to extract an initial set of water body boundary points from the water body mask. In this invention, the water body mask is generated by the deep learning feature extraction module 2 and is a binary image, where white areas represent water bodies and black areas represent non-water bodies. Boundary extraction unit 31 uses a contour tracking algorithm to extract the set of boundary points from the water body mask. Preferably, an improved Moore neighborhood tracking algorithm can be used, which starts from any boundary point and searches for adjacent boundary points according to fixed rules until it returns to the starting point, forming a closed boundary. To reduce the influence of noise, the extracted boundaries can be smoothed, such as by using Gaussian filtering with a kernel size of 5×5 and a standard deviation σ of 1.5. These parameter settings can remove the influence of noise while preserving the main features of the boundary.

[0133] Boundary parameterization unit 32 is connected to boundary extraction unit 31 and is used to parameterize the discrete boundary point set into a continuous curve. Parameterization is the process of representing the boundary as a parametric equation. The process, where s is the arc length parameter, and These are the x and y coordinates of the boundary points, respectively. In practical applications, parameterization can be achieved through cubic spline interpolation, i.e.:

[0134] ,

[0135] ,

[0136] in: Let x be the x-coordinate of the parametric curve at the parameter value s. Let y be the y-coordinate of the parametric curve at the parameter value s. The coefficients of the cubic spline interpolation for the x-coordinate are... Here, represents the coefficients of the cubic spline interpolation for the y-coordinate, and n is the number of nodes. Let be the arc length parameter value of the i-th node, where i ranges from 1 to 2. These coefficients are obtained by solving a system of linear equations, satisfying the conditions for the curve's position outside the nodes and the continuity of its derivative. Preferably, to ensure the uniformity of parameterization, equal arc length sampling can be used, that is, the arc length between adjacent nodes remains consistent, and the sampling interval can be set to 5-10 pixels. This range can achieve a balance between preserving boundary details and computational efficiency.

[0137] The curvature calculation unit 33 is connected to the boundary parameterization unit 32 and is used to calculate the local curvature on the boundary manifold, generating curvature distribution characteristics. On the parametric curve, the formula for calculating curvature is:

[0138] ,

[0139] in: Let be the curvature at parameter s. for The first derivative with respect to s, for The first derivative with respect to s, for The second derivative with respect to s, for The second derivative with respect to s, The modulus of the cross product result. This is the 3 / 2 power of the velocity vector length. In practical applications, the first and second derivatives can be approximated using finite difference methods:

[0140] ,

[0141] ,

[0142] in: This is the parameter increment, typically taken as 1 / 10 of the sampling interval. The parameter value is indicated. The x-coordinate at that location The parameter value is indicated. The x-coordinate at that location This represents the x-coordinate at a parameter value of s. The derivative of the y-coordinate is calculated in the same way as the x-coordinate. The calculated curvature sequence... The curvature distribution characteristics constituting the boundary of a water body, among which The parameter value is indicated. The curvature value at point n, where n is the number of sampling points.

[0143] Topology preservation unit 34 is connected to curvature calculation unit 33 to maintain the connectivity and directionality of the boundary and construct a discrete exterior differential form. Topology preservation unit 34 maintains the connectivity and directionality of the boundary by constructing the discrete exterior differential form. Specifically, a method based on discrete differential geometry can be used to construct a discrete curvature flow, whose evolution equation is:

[0144] ,

[0145] in: For the boundary curve, i.e. A curve representing the parameter change over time t. This represents the x and y derivatives of the curve with respect to the parameter s. By solving this equation, the boundary can be smoothly evolved while preserving the topological structure. In numerical solutions, an explicit Euler method can be used, with a time step typically set to 0.01-0.05 and 10-20 iterations. These parameter settings ensure numerical stability and computational efficiency.

[0146] Boundary descriptor generation unit 35 is connected to topology preservation unit 34 to construct boundary representation triples containing a vertex set, a connectivity set, and a curvature value set. This is a complete mathematical description of the water body boundary, where V is the set of boundary vertices, denoted as... , Let be the coordinates of the i-th vertex; E is the set of boundary connectivity relations, denoted as . , Indicates connection to vertices and The edge; K is the set of curvature values ​​at each point, denoted as , As vertex The curvature value at each vertex. Preferably, the vertex set V is obtained by uniform sampling, and the number of sampling points is generally 1 / 10 to 1 / 5 of the boundary perimeter; the connection relationship set E describes the connection relationship between vertices and can be represented as an adjacency matrix or adjacency list; the curvature value set K contains the curvature value at each vertex and is an important feature of the boundary morphology.

[0147] Reference Figure 5 The curvature flow dynamic monitoring module 4 includes a multi-temporal boundary extraction unit 41, a boundary mapping unit 42, a curvature difference analysis unit 43, an adaptive threshold unit 44, and a change region identification unit 45.

[0148] The multi-temporal boundary extraction unit 41 is used to extract water body boundary representations from different temporal phases. In practical applications, water body boundaries change over time, such as seasonal variations, expansion due to rainfall, or contraction due to drought. The multi-temporal boundary extraction unit 41 processes satellite images acquired at different times to extract water body boundary representation triplets for each temporal phase. , where i represents the phase index. Let be the vertex set of the i-th time phase. For the set of connection relationships of the i-th time phase, Let be the set of curvature values ​​for the i-th time phase. To ensure consistency in boundary extraction, the same preprocessing parameters and boundary extraction algorithm can be used to process images from different time phases. Preferably, in acquiring multi-time phase images, the consistency of image acquisition conditions should be considered, such as similar imaging times (to avoid excessive differences in solar illumination angles) and similar atmospheric conditions, in order to reduce the influence of non-water body change factors.

[0149] Boundary mapping unit 42 is connected to multi-temporal boundary extraction unit 41 and is used to construct the optimal transport mapping between boundaries of different temporal phases, establishing a point-to-point correspondence. Water body boundaries of different temporal phases typically have different shapes and lengths, making direct comparison difficult; therefore, it is necessary to establish a correspondence between boundaries. Boundary mapping unit 42 employs a boundary mapping method based on optimal transport theory. This method finds the optimal correspondence between two boundaries by solving the optimal transport problem. Preferably, a mapping algorithm based on Earth Mover's Distance (EMD) is used, which transforms the boundary mapping problem into minimizing the following objective function:

[0150] ,

[0151] in: Let be the flow (mapping intensity) from point i on boundary 1 to point j on boundary 2, representing the degree of correspondence between the two points. Let be the cost between points i and j, represent the dissimilarity between the two points, n be the number of points on boundary 1, and m be the number of points on boundary 2. This represents summing over all points on boundary 1. This represents the summation over all points on boundary 2. Cost It is usually defined as a weighted sum of Euclidean distance and curvature difference:

[0152] ,

[0153] in: Let i be the coordinates of the i-th point on boundary 1. Let j be the coordinates of the j-th point on boundary 2. The Euclidean distance between two points is expressed by the formula: ,in and Points x and y coordinates and Points The x and y coordinates; Let be the curvature of the i-th point on boundary 1. Let be the curvature of the j-th point on boundary 2. It represents the absolute value of the difference in curvature between two points; and These are weighting coefficients used to balance the importance of spatial distance and curvature differences; they are generally taken as... , This configuration can ensure spatial correspondence while taking into account morphological similarity.

[0154] The curvature difference analysis unit 43 is connected to the boundary mapping unit 42 and is used to calculate the curvature difference between corresponding points to generate a difference curvature field. Curvature difference is an important indicator for measuring changes in water body boundaries and can reflect local changes in boundary morphology. The curvature difference analysis unit 43 constructs a difference curvature field by calculating the curvature difference between mapped corresponding points. ,in:

[0155] ,

[0156] in: Let be the curvature difference value at the i-th point on boundary 1. Let be the curvature of the i-th point on boundary 1. Let i be the curvature of the point on boundary 2 that is mapped from point i on boundary 1. Let i be the index of the point on boundary 2 that is mapped to point i on boundary 1. This represents the absolute value of the curvature difference. To reduce the influence of noise, the difference curvature field can be smoothed, such as by using a Gaussian weighted average.

[0157] ,

[0158] in: This represents the smoothed curvature difference value. Here, w represents the original curvature difference value, and w is the window radius, indicating the neighborhood range considered in the smoothing operation; it is typically taken as 5-10. The standard deviation of the Gaussian kernel controls the degree of smoothing; it is typically taken as a fraction of the window radius. The Gaussian weights represent the influence weight of point j on point i. This represents summation within the neighborhood of point i. This smoothing process eliminates local noise and highlights the main changes. The adaptive threshold unit 44 is connected to the curvature difference analysis unit 43 and is used to dynamically calculate the change detection threshold based on the statistical characteristics of curvature differences.

[0159] The adaptive threshold unit 44 dynamically determines the change detection threshold by analyzing the statistical characteristics of the difference curvature field. For optimal performance, an adaptive thresholding method based on the OTSU algorithm is used, which determines the optimal threshold by maximizing the inter-class variance.

[0160] ,

[0161] in: The optimal threshold is t, and the candidate threshold is t. This represents the value of t that maximizes the expression that follows. The proportion of the first class (points with curvature differences less than t) under the threshold t is calculated using the following formula: ,in t represents the number of points with a curvature difference less than t, and n represents the total number of points; The proportion of the second type (points with curvature differences greater than or equal to t) under the threshold t is calculated using the following formula: ,in The number of points with a curvature difference greater than or equal to t; The mean of the first category is calculated using the following formula: ; The mean of the second category is calculated using the following formula: ; This represents the square of the difference between the two class means, indicating the magnitude of the inter-class difference. In practical applications, to improve the stability of detection, the final threshold can be set to 0.8-1.2 times the OTSU threshold, with the specific coefficient adjusted according to the application scenario. For scenarios requiring the detection of minute changes, a coefficient of 0.8 can be used; for scenarios focusing only on significant changes, a coefficient of 1.2 can be used.

[0162] The change region identification unit 45 is connected to the adaptive threshold unit 44 and is used to locate significantly changing boundary segments and calculate the area and rate of change of these segments. The change region identification unit 45 identifies significantly changing boundary segments by comparing the difference curvature field with a threshold. Specifically, a set of points that continuously exceed the threshold can be defined as a change boundary segment. ,in Let be the starting parameter for the i-th change segment. This is the end parameter for the i-th change segment. To avoid over-segmentation caused by scattered change points, change segments can be merged. If the distance between two change segments is less than a preset threshold (e.g., 10 points), they are merged into one change segment. Based on the identified change boundary segments, the change area and change rate can be calculated. The formula for calculating the change area is:

[0163] ,

[0164] in: For changes in water body area, Let i be the area of ​​the i-th water body in phase 1. Let be the area of ​​the j-th water body in phase 2, n be the number of water bodies in phase 1, and m be the number of water bodies in phase 2. This represents the summation of the areas of all water bodies in phase 1. This represents the summation of the areas of all water bodies in phase 2. The formula for calculating the rate of change is:

[0165] ,

[0166] Where R is the rate of change, representing the percentage change in area relative to time phase 1. This represents the absolute value of the area change. This represents the sum of the areas of all water bodies in phase 1. The area of ​​change and the rate of change are important quantitative indicators of dynamic changes in water bodies, and can be used to assess the degree and rate of change in water resources.

[0167] Reference Figure 6 The multi-scale evolution prediction module 5 includes a time series data processing unit 51, a discrete Ricci stream unit 52, a multi-scale representation unit 53, a boundary spectrum analysis unit 54, an evolution pattern recognition unit 55, and a boundary prediction unit 56.

[0168] The time-series data processing unit 51 is used to collect satellite images from multiple periods and extract time-series boundary representation sequences. Time-series data is the foundation for predicting water body boundary evolution, containing historical information about how water body boundaries change over time. The time-series data processing unit 51 processes multiple periods of satellite images acquired over a period of time, extracting water body boundary representations for each period to form a time-series boundary representation sequence. ,in The boundary representation triple of the i-th time phase is represented by... For vertex set, To connect sets of relations, Let n be the set of curvature values, and n be the number of time phases. Preferably, the time series data should include images from different seasons and years to capture the seasonal changes and long-term trends of the water body. In time series data acquisition, the time interval is generally 15 to 30 days, and the total duration should cover at least one complete seasonal variation cycle (e.g., one year).

[0169] Discrete Ricci flow unit 52 is connected to time-series data processing unit 51 and is used to construct a discrete Ricci flow model and generate boundary curvature evolution equations. Ricci flow is a partial differential equation describing the geometric evolution of a surface; in this invention, discrete Ricci flow is used to describe the geometric evolution of water body boundaries. Discrete Ricci flow unit 52 discretizes the continuous Ricci flow equations by constructing a discrete Ricci flow model, generating curvature evolution equations applicable to water body boundaries. The basic equations of discrete Ricci flow are:

[0170] ,

[0171] in: Let Riemannian metric tensor represent the distance metric on the surface. Let be the Ricci curvature tensor, representing the degree of curvature of the surface, and t be the flow parameter, representing the evolution time. This represents the rate of change of the metric tensor over time. On a one-dimensional manifold such as a water body boundary, the equation simplifies to:

[0172] ,

[0173] in: Let t be the curvature and t be the flow parameter. This represents the rate of change of curvature with respect to flow parameters. Let be the Laplace-Beltrami operator, representing the second spatial derivative of curvature. Let be the cube of the curvature, representing the nonlinear term. In discretization, a finite difference method can be used, such as:

[0174] ,

[0175] in: Let i be the curvature of point i at time t. Let i be the curvature at time t+1. The time step controls the speed of evolution. The spatial step size represents the parameter distance between adjacent sampling points. and Let be the curvatures of two adjacent points i at time t, respectively. The discrete Laplace operator for curvature approximates the continuous case. , Let be the cube of the curvature, representing the nonlinear term. Preferably, the time step... The space step size is set to 0.01-0.05. These parameters, set as the sampling interval, ensure numerical stability and computational accuracy.

[0176] Multi-scale representation unit 53 is connected to discrete Ricci flow unit 52 to construct the scale space of the boundary and generate multi-scale geometric features. Multi-scale representation unit 53 obtains the boundary representation under different flow parameters t by solving the discrete Ricci flow equation, thus constructing the scale space of the boundary. ,in This represents the boundary curve when the flow parameter is t. Let T be the range of the flow parameters, where T is the maximum flow parameter. Preferably, the flow parameter t can be set as a geometric sequence, such as... This setup allows for sufficient sampling points at both small and large scales. At each scale, geometric features of the boundary, such as curvature distribution and curvature moments, are extracted to form multi-scale geometric features.

[0177] Boundary spectrum analysis unit 54 is connected to multi-scale representation unit 53 and is used to calculate the characteristic spectrum of the boundary manifold and extract invariant geometric features. Boundary spectrum analysis unit 54 calculates the characteristic spectrum of the boundary manifold by solving the discrete Laplace-Beltrami eigenvalue problem. ,in Let be the i-th eigenvalue, and m be the number of eigenvalues. In a concrete implementation, a matrix representation of the discrete Laplace-Beltrami operator can be constructed:

[0178] ,

[0179] in: Let i be the element in the i-th row and j-th column of the discrete Laplace-Beltrami operator matrix. Let be the area of ​​the Voronoi region of vertex i, and let represent the area weight of vertex i. Let be the set of neighbors of vertex i, and let be the set of vertices directly connected to vertex i. and For the edge The angles on both sides and These are the cotangent values ​​for these two angles, Representing an edge The weight, This represents the summation of all neighbors of vertex i, where i=j represents the diagonal elements of the matrix. Let 'j' indicate that vertex j is a neighbor of vertex i, and 'otherwise' indicate other cases. Solving the eigenvalue problem of L yields the eigenspectrum. Preferably, the first 10 to 20 eigenvalues ​​are taken as the eigenspectrum; this range strikes a balance between descriptive power and computational efficiency.

[0180] The evolution pattern recognition unit 55 is connected to the boundary spectrum analysis unit 54 and is used to identify boundary evolution patterns and learn the dynamic laws of evolution. The evolution pattern recognition unit 55 identifies the evolution patterns of the boundary by analyzing the time-series boundary representation sequence and multi-scale geometric features. Preferably, time series analysis methods, such as autoregressive models and seasonal decomposition, can be used to extract the periodic and trend characteristics of boundary evolution. In specific implementation, boundary features (such as area, perimeter, curvature distribution, etc.) can be constructed as a time series and subjected to seasonal decomposition.

[0181] ,

[0182] in: This represents the original time series, indicating the feature value at time t. The trend component represents the long-term trend of change. The seasonal component indicates periodic changes. The residual component represents random fluctuations, and t is the time index. Through analysis... and It can identify long-term trends and seasonal patterns. Furthermore, the dynamic characteristics of time series can be described using the Autoregressive Moving Average (ARMA) model.

[0183] ,

[0184] in: The eigenvalue at time t, For time The eigenvalue at point i represents the value after lag i. The order of the autoregressive term represents the number of historical data periods considered. The autoregressive coefficient represents the weight of the value lagged by i periods. The order of the moving average term represents the number of historical error periods considered. The moving average coefficient represents the weight of the error with lag j periods. For time The white noise at that location represents random perturbation. For time The white noise at point j represents a random disturbance with a lag of j. This indicates summing over the autoregressive terms. This represents the summation of the moving average terms. Preferably, the order of the ARMA model can be determined by the AIC or BIC criterion, and generally both p and q do not exceed 3.

[0185] Boundary prediction unit 56 is connected to evolution pattern recognition unit 55 and is used to predict the boundary morphology at future times and assess prediction uncertainty. Boundary prediction unit 56 predicts the water body boundary morphology at future times based on the identified evolution patterns and learned dynamic laws. Preferably, a two-stage prediction method can be used: first, the boundary features (such as area, perimeter, etc.) at future times are predicted based on a time series model; then, the boundary morphology is reconstructed using the Ricci flow equation. Specifically, an ARMA model can be used for feature prediction.

[0186] ,

[0187] in: Let h be the ahead prediction value, representing the predicted feature value for the h-th period in the future. For the predicted future The characteristic value of the period, For the estimated autoregressive coefficients, For the estimated moving average coefficients, Here, represents the estimated error term, h is the prediction step size, indicating the number of future periods to be predicted, p is the order of the autoregressive term, and q is the order of the moving average term. This indicates summing over the autoregressive terms. This represents the summation of the moving average term. Based on the predicted features, the boundary shape is reconstructed by solving the inverse Ricci flow problem:

[0188] ,

[0189] in: The boundary curve is denoted by t, where t is the flow parameter. This represents the rate of change of boundary parameters with respect to flow. Let n be the curvature, and n be the unit normal vector. This represents the contraction velocity along the normal direction. Solving this equation yields the boundary shape that satisfies the prediction characteristics. To assess prediction uncertainty, the bootstrap method can be used to generate prediction intervals, or multiple possible prediction results can be generated through Monte Carlo simulation, and the confidence interval of the prediction can be calculated.

[0190] Reference Figure 7 The multi-source data fusion module 6 includes a land use module 61, a meteorological data module 62, and an environmental parameter module 63.

[0191] The land use module 61 is used to analyze the spatial location and characteristic parameters of various types of water bodies, such as rivers, reservoirs, lakes, wetlands, and farmland, using land surveying data. Land use data provides the spatial distribution of water body types and land cover types. The land use module 61 processes land surveying data to extract spatial distribution information of different types of water bodies and calculates characteristic parameters such as area, shape, and connectivity of each type of water body. Preferably, remote sensing image classification and GIS spatial analysis methods can be used, combining satellite imagery and land use data to generate a water body type distribution map. During the classification process, machine learning methods such as random forests or support vector machines can be used, utilizing multi-source features (such as spectral features, texture features, and topographic features) for classification. The classification accuracy should generally reach above 85%, and the producer accuracy and user accuracy of water body categories in the confusion matrix should both exceed 90%.

[0192] The meteorological data module 62 is connected to the land use module 61 and is used to calculate multi-source meteorological data using meteorological satellite data. Through spatiotemporal correlation, it analyzes water volume and its changing trends. The meteorological data module 62 processes meteorological satellite data and ground observation data to calculate multi-source meteorological parameters, such as rainfall, evaporation, and temperature, and analyzes the relationship between meteorological factors and water body changes through spatiotemporal correlation. Preferably, a spatiotemporal interpolation method, such as Kriging interpolation, can be used to interpolate discrete meteorological observation data into a continuous meteorological field, achieving a spatial correspondence between meteorological data and water body data. In terms of temporal correlation, the lag effect of meteorological factors can be considered; for example, the lag time of rainfall's impact on water body area is generally 1–7 days, depending on the watershed characteristics and rainfall intensity.

[0193] The environmental parameter module 63 is connected to the meteorological data module 62 and is used to combine environmental information surrounding the water body to establish a buffer zone analysis on the GIS platform to assess the impact of the environment on water resources. The environmental parameter module 63 establishes a buffer zone around the water body on the GIS platform, analyzes environmental parameters within the buffer zone, such as land cover type, population density, and industrial distribution, and assesses the impact of environmental factors on water resources. Preferably, the buffer zone can be set at 1-5 kilometers outside the water body boundary, flexibly adjusted according to the water body type and research objectives. Within the buffer zone, the area proportion and spatial distribution characteristics of various environmental parameters can be calculated, and the correlation between these parameters and changes in the water body can be analyzed.

[0194] The decision support module 7 is connected to the multi-scale evolution prediction module 5 and the multi-source data fusion module 6, and is used to generate risk assessment and early warning information based on the water change prediction results and fused data. The decision support module 7 includes a risk indicator calculation unit, a risk level assessment unit, an early warning rule engine unit, and an early warning information generation unit.

[0195] The risk indicator calculation unit receives the predicted boundary morphology and prediction uncertainty output by the multi-scale evolution prediction module 5, and the spatial distribution characteristics of water resources output by the multi-source data fusion module 6, and calculates multi-dimensional risk indicators. The risk indicators calculated by the risk indicator calculation unit include area change risk indicators, boundary stability risk indicators, capacity risk indicators, and environmental impact risk indicators.

[0196] The area change risk index is calculated by quantifying the deviation between the predicted rate of change in water area and the historical average rate of change for the same period. Specifically, the formula for calculating the area change risk index is:

[0197] ,

[0198] in: This is an area change risk indicator, representing the degree of abnormality in water body area changes. It is a dimensionless scalar with a value range of [value range missing]. It is obtained through calculation using a formula; The predicted rate of change of water area represents the percentage change in water area relative to the current area at a future time, output by the multi-scale evolution prediction module. It is a scalar and its value range is... The unit is a percentage, and it comes from the output of the boundary prediction unit 56; The historical average area change rate represents the average change rate of water body area in the same season over the past three years. It is a scalar quantity with a range of values. The units are percentages, obtained through statistical analysis of historical data; The standard deviation of the historical area change rate represents the dispersion of the historical change rate. It is a positive scalar and its value range is [value range missing]. The units are percentages, obtained through statistical analysis of historical data; This is the area risk weighting coefficient, a positive scalar, with a value range of [value range missing]. Dimensionless, set according to water type: 0.3 for rivers, 0.4 for lakes, and 0.5 for reservoirs.

[0199] The boundary stability risk index is calculated by quantifying the magnitude and uncertainty of the predicted boundary curvature change. Specifically, the formula for calculating the boundary stability risk index is:

[0200] ,

[0201] in: This is a boundary stability risk index, representing the degree of risk of changes in the morphology of water body boundaries. It is a dimensionless scalar with a value range of [range missing]. , is obtained by calculation through formula; n is the number of boundary sampling points, which is a positive integer and its value range is determined according to the circumference of the water body, usually from 100 to 500; For the first The predicted curvature change at each sampling point represents the difference between the future curvature and the current curvature at that point. It is a scalar and its value range is... The unit is The output comes from the boundary prediction unit 56; To predict the uncertainty coefficient, the inverse index representing the confidence level of the boundary prediction is a non-negative scalar with a value range of [value range missing]. , dimensionless, is derived from the normalized value of the prediction interval width calculated by the bootstrap method in boundary prediction unit 56; This is the boundary risk weighting coefficient, of type positive scalar, with a value range of [value range missing]. Dimensionless, set according to the type of water body: 0.4 for rivers, 0.3 for lakes, and 0.2 for reservoirs.

[0202] The capacity risk index is calculated by quantifying the relationship between predicted changes in water body capacity and warning capacity. Specifically, the formula for calculating the capacity risk index is:

[0203] ,

[0204] in: This is a capacity risk indicator, representing the degree of risk of insufficient water capacity. It is a dimensionless scalar quantity with a value range of [missing value]. It is obtained through calculation using a formula; The warning capacity threshold represents the minimum capacity required for the normal function of a water body. It is a positive scalar, and its value range is determined according to the type and use of the water body. The unit is cubic meters, and it is derived from the water body function parameters provided by the land use module 61 in the multi-source data fusion module 6. For predicted capacity, it represents the future water volume estimated by combining predicted area and depth. It is a positive scalar and its value range is [range missing]. The unit is cubic meters, which is derived from the joint calculation of the output of the water volume estimation module 8 and the output of the boundary prediction unit 56; This is the capacity risk weighting coefficient, a positive scalar, with a value range of [value range missing]. Dimensionless, set according to the purpose of the water body: 0.5 for drinking water sources, 0.3 for agricultural irrigation, and 0.2 for ecological water use.

[0205] The calculation method for environmental impact risk indicators is as follows: quantifying the potential impact of water body changes on surrounding land use and ecology. Specifically, the formula for calculating environmental impact risk indicators is:

[0206] ,

[0207] in: This is an environmental impact risk indicator, representing the comprehensive impact of water body changes on the surrounding environment. It is a dimensionless scalar quantity with a value range of [range missing]. It is obtained through a formula; m is the number of land use types within the buffer zone, and the types are positive integers, usually 5 to 8; Let be the impact sensitivity of the j-th land use type, representing the degree of sensitivity of this type to changes in water bodies. The type is a dimensionless scalar, and its value range is [value range missing]. The values ​​are 0.8 for farmland, 0.9 for residential areas, 0.6 for industrial areas, 0.4 for forest land, and 0.2 for unused land, which are derived from the settings of the environmental parameter module 63 in the multi-source data fusion module 6. Let be the area proportion of the j-th land use type within the buffer zone, be a scalar, and have a value range of . Dimensionless, derived from the spatial analysis results of environmental parameter module 63; Let J be the environmental weight for the j-th land use type, be a positive scalar, and have a value range of 1. Dimensionless, set according to regional environmental protection priorities.

[0208] The risk level assessment unit is connected to the risk indicator calculation unit and is used to comprehensively assess the risk level of the water body based on the calculated multi-dimensional risk indicators. The risk level assessment unit uses a weighted comprehensive scoring method to calculate the comprehensive risk score, and the calculation formula is as follows:

[0209] ,

[0210] in: The comprehensive risk score represents the overall risk level of the water body. It is a dimensionless scalar with a value range of [range missing]. It is obtained through calculation using a formula; This is the comprehensive weighting coefficient for each risk indicator, a positive scalar, with a value range of [value range missing]. And satisfy Dimensionless, preferably , , , .

[0211] The risk level assessment unit classifies risks into four levels based on the comprehensive risk score: when When the risk level is low, it indicates that the water body is in normal condition and no special attention is required; when At that time, the risk level was low to medium, indicating a slight anomaly in the water body, and continued monitoring was recommended; when At that time, the risk level is medium to high, indicating that there are obvious anomalies in the water body and preventive measures need to be taken; when At this time, the risk level is high, indicating that there is a serious anomaly in the water body and immediate intervention is required.

[0212] The early warning rule engine unit connects to the risk level assessment unit and is used to trigger corresponding early warning rules based on the risk level and specific risk indicators. The early warning rule engine unit has multiple built-in early warning rules, including:

[0213] The first early warning rule is for warnings of rapid changes in area: when... and At that time, a warning of a sharp change in area was triggered, and the warning type was orange.

[0214] The second early warning rule is the boundary abnormal deformation early warning: when more than 30% of the points in the boundary sampling points meet the condition... When the boundary abnormal deformation warning is triggered, the warning type is yellow warning. This represents the standard deviation of the historical curvature variation.

[0215] The third early warning rule is a capacity shortage warning: when... When this occurs, a severe capacity shortage warning is triggered, and the warning type is red; when When this occurs, an insufficient capacity warning is triggered, and the warning type is orange.

[0216] The fourth early warning rule is the environmental impact early warning: when Furthermore, if there are residential areas or farmland within the buffer zone, an environmental impact warning will be triggered, and the warning type will be an orange warning.

[0217] The fifth early warning rule is a comprehensive high-risk warning: when At any given time, regardless of the specific indicators, a comprehensive high-risk warning will be triggered, and the warning type will be a red warning.

[0218] The early warning information generation unit is connected to the early warning rule engine unit and is used to generate structured early warning information based on the triggered early warning rules. The early warning information generated by the early warning information generation unit includes the early warning time, the water body identification, the early warning type, the early warning level, the risk indicator value, the triggering rule number, the recommended measures, and the effective period.

[0219] The early warning information generation unit generates corresponding recommended measures based on different early warning types: For early warnings of rapid area changes, recommended measures include increasing the monitoring frequency to once a day, verifying upstream water inflow and downstream water outflow, and assessing whether there are any human intervention factors; for early warnings of abnormal boundary deformation, recommended measures include conducting on-site investigations of the boundary change area, assessing the stability of the bank slope, and taking engineering reinforcement measures when necessary; for early warnings of insufficient capacity, recommended measures include initiating emergency water source allocation, restricting non-essential water use, and notifying downstream water users to prepare; for early warnings of environmental impact, recommended measures include notifying residents or farmers in the affected area, assessing ecological water replenishment needs, and coordinating with relevant departments to take protective measures; for comprehensive high-risk early warnings, recommended measures include activating the emergency response mechanism, organizing expert consultations, and developing a comprehensive response plan.

[0220] The early warning information generation unit outputs the generated early warning information in a standardized format, supporting multiple notification methods such as SMS, email, and APP push. At the same time, it stores the early warning information in the early warning history database for subsequent early warning effect evaluation and rule optimization.

[0221] Reference Figure 8 The water volume estimation module 8 is connected to the differential geometry water boundary characterization module 3 and the curvature flow dynamic monitoring module 4, and is used to estimate the water volume and assess the impact of water changes on the surrounding environment.

[0222] The water capacity estimation module 8 uses DEM data to calculate the elevation differences of various types of water bodies such as rivers, lakes, and reservoirs. It combines the water body area and water depth to analyze the water body capacity, assess the impact of water body changes on other surrounding environments, classify the capacity, and output the water resource capacity detection results.

[0223] The processing steps of water volume estimation module 8 include the following:

[0224] First, based on DEM data, the difference between the water level and the riverbed height is calculated to estimate the water depth. DEM (Digital Elevation Model) data provides surface elevation information, which can be used to estimate water depth. Specifically, the elevation values ​​at the water body boundary can be extracted as the water surface elevation, and the lowest elevation value inside the water body can be extracted as the riverbed / lake bottom elevation; the difference between the two is the water depth. Preferably, to improve estimation accuracy, multi-point sampling and statistical analysis methods can be used, such as extracting the elevation values ​​of multiple points on the water body boundary and taking the median as the water surface elevation. This method can reduce the influence of outliers. For large water bodies, the non-horizontal nature of the water surface can be considered; for example, in flowing rivers, there are differences in water surface elevation between upstream and downstream, and a segmented estimation method can be used.

[0225] Secondly, considering the water body's surface area, calculate its volume. Water volume is a function of both surface area and depth, and can be calculated using the following formula:

[0226] ,

[0227] Where: V is the water volume in cubic meters, A is the water area in square meters, D is the average water depth in meters, and C is the shape correction factor, dimensionless, used to correct for the influence of the water body's cross-sectional shape on volume calculation. The shape correction factor reflects the influence of the water body's cross-sectional shape on volume calculation; for example, C is approximately 0.5 for a V-shaped cross-section, approximately 0.67 for a U-shaped cross-section, and approximately 1.0 for a rectangular cross-section. In practical applications, an appropriate shape correction factor can be selected based on the water body type and topographic features, or calibration can be performed using measured data.

[0228] Then, water volume statistics are performed on various types of water bodies. Different types of water bodies (such as rivers, lakes, and reservoirs) have different characteristics and functions, and classified statistics can provide more targeted information for water resource management. Preferably, statistics can be performed according to water body type (rivers, lakes, reservoirs, wetlands, etc.), administrative region (province, city, county, etc.), or watershed unit to generate multi-dimensional capacity statistics results.

[0229] Finally, the capacity is classified, and the water resource capacity detection results are output. Water body capacity classification helps to intuitively express the size and importance of water bodies, facilitating water resource management decisions. Preferably, the classification can be based on the statistical distribution of water body capacity, using the natural breakpoint method, the equal interval method, or the quantile method, generally divided into 5 to 7 levels. The results can be presented in various forms, such as classification symbol maps, statistical reports, and capacity change curves, to facilitate user understanding and use.

[0230] The satellite image dynamic monitoring system for water resources based on deep learning of the present invention can be applied to a variety of scenarios, such as watershed water resources assessment and urban water resources management.

[0231] In the context of watershed water resources assessment, the implementation process of this invention includes:

[0232] First, acquire satellite time-series images of each water body within the monitoring area. Use satellite remote sensing platforms (such as Sentinel-2, Landsat, etc.) to acquire satellite images of the monitoring area at different times, and construct a time-series dataset. Preferably, the time series should cover different seasons to capture the seasonal variation patterns of the water bodies.

[0233] Secondly, obtain internet mapping data for the monitoring area. Obtain mapping data from publicly available internet databases (such as OpenStreetMap and Google Earth) as supplementary information to improve monitoring accuracy.

[0234] Then, the water images are analyzed to identify rivers, lakes, reservoirs, wetlands, and farmland, generating change detection maps and statistically analyzing the changed area and rate of change. Utilizing the deep learning feature extraction module and differential geometry water boundary representation module of this invention, water information is extracted from satellite images, and water changes are detected through a curvature flow dynamic monitoring module to generate change detection maps.

[0235] Next, land use and environmental parameter information is acquired to calculate the capacity of various water bodies and changes in the surrounding environment. Using a multi-source data fusion module and a water body capacity estimation module, combined with land use data and environmental parameters, the capacity of various water bodies is calculated, and the impact of changes in the surrounding environment on the water bodies is analyzed.

[0236] Finally, a watershed water resources assessment layer is generated for watershed water resources management and planning decisions. Integrating monitoring and analysis results into this watershed water resources assessment layer provides a scientific basis for watershed management and planning.

[0237] In the context of urban water resource management, the implementation process of this invention includes:

[0238] First, acquire satellite time-series images of various types of water bodies in the monitoring area. For urban areas, select high-resolution satellite images (such as GF-2, WorldView, etc.) to capture changes in small water bodies within the city.

[0239] Secondly, the changes in water bodies are analyzed by combining publicly available internet databases and real-world mapping data. Changes in urban water bodies are often influenced by human activities, requiring comprehensive analysis incorporating data from urban planning and land use changes.

[0240] Then, the changes in the water body are analyzed, and the water volume is output. Using the water volume estimation module of this invention, the changes in the volume of urban water bodies are calculated, and the status of urban water resources is assessed.

[0241] Next, an urban water resource management layer is generated to support urban planning and decision-making. By combining water monitoring results with urban planning, an urban water resource management layer is created to support urban planning and construction decisions.

[0242] In summary, the satellite imagery-based dynamic water resources monitoring system provided by this invention, by integrating deep learning and differential geometry theory, achieves high-precision identification, accurate change detection, and scientific trend prediction of water bodies, providing comprehensive and accurate data support for water resources management and decision-making, and has broad application prospects and significant socio-economic value.

[0243] Although the present invention has been described in detail above with general descriptions, specific embodiments, and experiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for dynamic monitoring of water resources based on satellite imagery using deep learning, characterized in that, Includes the following steps: Acquire multi-channel satellite images of the detection area and perform image denoising, band adjustment, and geometric correction to enhance image quality; The input image is converted to a grayscale image, and features are extracted. The extracted water body boundary is regarded as a two-dimensional manifold, the boundary curvature characteristics are calculated, and the boundary manifold representation is constructed. Construct the optimal transport mapping between water body boundaries in different time phases, calculate the curvature field difference, determine the adaptive threshold, identify the changing region, and calculate the changing area and rate of change; We construct a multi-scale representation of water body boundaries using Ricci flow, analyze boundary evolution patterns, and predict future water body boundary morphology. By integrating satellite imagery, publicly available internet data, and public basic data, spatial distribution characteristics of water resources are extracted. Estimate the water volume and assess the impact of water changes on the surrounding environment; Risk assessment and early warning information are generated based on water body change prediction results and fused data; The extracted water body boundary is treated as a two-dimensional manifold, and the boundary curvature features are calculated to construct a boundary manifold representation. Specifically: A contour tracking algorithm is used to extract the set of boundary points from the water body mask; the water body mask is a binary image generated by converting a grayscale image and performing feature extraction steps, wherein the white area represents water and the black area represents non-water. Parameterize the discrete boundary point set into a continuous curve; Calculate local curvature on the boundary manifold to generate curvature distribution characteristics; By constructing the discrete exterior differential form of the boundary, the connectivity and directionality of the boundary are preserved; Construct a boundary representation triple containing a vertex set, a connectivity set, and a curvature value set; The process of constructing the optimal transport mapping between water body boundaries at different time phases, calculating curvature field differences, determining adaptive thresholds, identifying changing regions, and calculating the changing area and rate of change are specifically as follows: Extracting water body boundary representations at different time phases: By processing satellite images acquired at different times, extracting water body boundary representation triplets for each time phase; Construct the optimal transmission mapping between different temporal boundaries and establish point-to-point correspondences; Calculate the curvature difference between corresponding points and generate a difference curvature field; By analyzing the statistical characteristics of the differential curvature field, the change detection threshold is dynamically determined; Locate the boundary segments with significant changes, and calculate the area and rate of change of the changes; The construction of a multi-scale Ricci flow representation of the water body boundary, analysis of boundary evolution patterns, and prediction of future water body boundary morphology are specifically as follows: By processing multiple satellite images acquired over a period of time, water body boundary characteristics are extracted for each period to form a temporal boundary characterization sequence. By constructing a discrete Ricci flow model, the continuous Ricci flow equation is discretized, generating curvature evolution equations applicable to water body boundaries; By solving the discrete Ricci flow equation, the boundary representation under different flow parameters is obtained, the scale space of the boundary is constructed, and multi-scale geometric features are generated. Calculate the feature spectrum of the boundary manifold and extract invariant geometric features; By analyzing the temporal boundary characterization sequence and multi-scale geometric features, we can identify the evolution patterns of the boundaries and learn the laws of evolution dynamics. Based on the identified evolution patterns and learned dynamics, the water boundary morphology at future moments is predicted, and the prediction uncertainty is assessed.

2. The satellite imagery-based dynamic water resources monitoring method according to claim 1, characterized in that, The process of acquiring multi-channel satellite images of the detection area and performing image denoising, band adjustment, and geometric correction to enhance image quality specifically involves: The received satellite images are denoised to remove noise and interference: an improved nonlocal mean filtering algorithm is used to denoise the images by searching for similar blocks in the images and using the similarity between blocks to suppress noise. Adjusting multi-channel satellite images to achieve multi-channel image fusion: enhancing water body characteristics by weighting and fusing different bands; Using satellite image data, DEM image data, and a reference base map as input, correction control points are generated through coordinate matching, and new coordinates are established using bilinear interpolation. The satellite image data is then registered onto the reference base map: first, feature points are extracted from the satellite image, and then matched with corresponding points in the reference base map to establish a conversion model.

3. The satellite imagery-based dynamic water resource monitoring method according to claim 1, characterized in that, The process of converting the input image into a grayscale image and extracting features specifically involves: The input image is converted into a grayscale image by weighted averaging of the RGB channels of the color image. Perform a two-dimensional Fourier transform on the grayscale image to extract the low-frequency, mid-frequency, and high-frequency features of the image; The low-frequency features represent the average brightness, mediumness, roughness, contrast, and histogram of the image; the mid-frequency features represent the line texture and edge information in different directions of the image; and the high-frequency features represent the detail features and noise of the image. The transformed grayscale image is normalized, Euclidean distance is calculated, and cluster centers are generated to obtain the feature map set of the image.

4. The satellite imagery-based dynamic water resource monitoring method according to claim 1, characterized in that, The fusion of satellite imagery, publicly available internet data, and public basic data extracts spatial distribution characteristics of water resources, specifically as follows: Using land survey data, we analyze the spatial location and characteristic parameters of various types of water bodies; using meteorological satellite data, we calculate multi-source meteorological data and analyze the water body capacity and changing trends through spatiotemporal correlation; combining the environmental information around the water body, we establish buffer analysis on the GIS platform to assess the impact of the environment on water resources.

5. The satellite imagery-based dynamic water resources monitoring method according to claim 1, characterized in that, The estimation of water body capacity and assessment of the impact of water body changes on the surrounding environment specifically include: Based on DEM data, calculate the difference in elevation between the water level and the riverbed to estimate the water depth. Calculate the water volume by combining the water depth and water area; Perform water volume statistics on various types of water bodies; The capacity is classified and the results of water resource capacity testing are output.

6. A system for implementing the deep learning-based satellite imagery-based dynamic water resource monitoring method as described in any one of claims 1-5, characterized in that, The system includes: The satellite image processing module is used to receive multi-channel satellite images, perform image denoising, band adjustment, and geometric correction to enhance image quality. A deep learning feature extraction module is connected to the satellite image processing module and is used to extract features from the images processed by the satellite image processing module to form a feature map set. The differential geometry water body boundary representation module is connected to the deep learning feature extraction module. It is used to treat the extracted water body boundary as a two-dimensional manifold, calculate the boundary curvature features, and construct the boundary manifold representation. The curvature flow dynamic monitoring module is connected to the differential geometry water body boundary characterization module. It is used to construct the optimal transport mapping between water body boundaries in different time phases, calculate the curvature field difference, determine the adaptive threshold, identify the changing region, and output the changing area and rate of change. The multi-scale evolution prediction module, connected to the curvature flow dynamic monitoring module, is used to construct a multi-scale representation of the Ricci flow of the water body boundary, analyze the boundary evolution pattern, and predict the future water body boundary morphology. A multi-source data fusion module, connected to the curvature flow dynamic monitoring module, is used to fuse satellite images, publicly available Internet data, and public basic data to extract spatial distribution characteristics of water resources. The decision support module, connected to the multi-scale evolution prediction module and the multi-source data fusion module, is used to generate risk assessment and early warning information based on the water body change prediction results and fused data. The water volume estimation module, connected to the differential geometry water boundary characterization module and the curvature flow dynamic monitoring module, is used to estimate the water volume and assess the impact of water changes on the surrounding environment.

7. The system according to claim 6, characterized in that, The satellite image processing module includes: The image denoising unit is used to denoise the received satellite images, removing noise and interference; A band adjustment unit, connected to the image denoising unit, is used to adjust multi-channel satellite images to achieve multi-channel image fusion; The geometric correction unit, connected to the band adjustment unit, is used to take satellite image data, DEM image data and reference base map as input, generate correction control points through coordinate matching, establish new coordinates using bilinear interpolation, and register the satellite image data onto the reference base map. The differential geometry water body boundary characterization module includes: The boundary extraction unit is used to extract the initial set of water body boundary points from the water body mask; the water body mask is a binary image generated by the deep learning feature extraction module, in which white areas represent water bodies and black areas represent non-water bodies. A boundary parameterization unit, connected to the boundary extraction unit, is used to parameterize a discrete set of boundary points into a continuous curve. The curvature calculation unit, connected to the boundary parameterization unit, is used to calculate the local curvature on the boundary manifold and generate curvature distribution features; A topology preservation unit, connected to the curvature calculation unit, is used to maintain the connectivity and directionality of the boundary and construct a discrete exterior differential form; A boundary descriptor generation unit, connected to the topology maintenance unit, is used to construct a boundary representation triplet containing a vertex set, a connection relation set, and a curvature value set. The curvature flow dynamic monitoring module includes: Multi-temporal boundary extraction unit is used to extract water body boundary characteristics at different temporal phases; The boundary mapping unit, connected to the multi-temporal boundary extraction unit, is used to construct the optimal transmission mapping between different temporal boundaries and establish a point-to-point correspondence. A curvature difference analysis unit, connected to the boundary mapping unit, is used to calculate the curvature difference between corresponding points and generate a difference curvature field. An adaptive threshold unit, connected to the curvature difference analysis unit, is used to dynamically calculate the change detection threshold based on the statistical characteristics of curvature differences. A change region identification unit, connected to the adaptive threshold unit, is used to locate significantly changing boundary segments and calculate the change area and change rate. The multi-scale evolution prediction module includes: The time-series data processing unit is used to collect satellite images from multiple periods and extract time-series boundary characterization sequences. A discrete Ricci flow unit, connected to the time-series data processing unit, is used to construct a discrete Ricci flow model and generate boundary curvature evolution equations. A multi-scale representation unit, connected to the discrete Ricci flow unit, is used to construct the scale space of the boundary and generate multi-scale geometric features; The boundary spectrum analysis unit, connected to the multi-scale representation unit, is used to calculate the feature spectrum of the boundary manifold and extract invariant geometric features; An evolution pattern recognition unit, connected to the boundary spectrum analysis unit, is used to identify boundary evolution patterns and learn evolution dynamics. A boundary prediction unit, connected to the evolution pattern recognition unit, is used to predict the boundary shape at future moments and assess prediction uncertainty. The multi-source data fusion module includes: The land use module is used to analyze the spatial location and characteristic parameters of various types of water bodies using land surveying data; The meteorological data module, connected to the land use module, is used to calculate multi-source meteorological data using meteorological satellite data, and analyze water volume and change trends through spatiotemporal correlation. The environmental parameter module, connected to the meteorological data module, is used to combine information about the surrounding environment of the water body to establish a buffer analysis on the GIS platform and assess the impact of the environment on water resources.

Images