Lake terrain prediction method based on machine learning fused with multi-source satellite data

Abstract

The invention discloses a lake terrain prediction method based on machine learning fused multi-source satellite data, and relates to the technical field of machine learning, and the method comprises the steps: extracting the optical features of a target lake from an optical satellite image; extracting water features of each water body pixel in the target lake from the SWOT satellite data; performing space-time alignment and fusion on the optical features and the water features to obtain a fusion feature vector; the actually measured water depth point of the real water depth is matched to the geographic position of the corresponding pixel, and then the real water depth and the fusion feature vector of the corresponding pixel form a training sample; selecting a machine learning model and initializing the structure or parameters of the machine learning model; training the initialized machine learning model by using the training sample; utilizing the trained machine learning model to predict the water depth value of the corresponding pixel according to the fusion characteristic vector of each water body pixel; and determining the terrain of the target lake according to each water depth value. According to the method, the terrain of a large-scale lake can be efficiently predicted at low cost.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a lake terrain prediction method based on machine learning and fusion of multi-source satellite data. Background Art

[0002] Lakes are a vital component of Earth's surface system. Their underwater topography (also known as bathymetry) data is crucial for water resource assessment and management, water storage calculations, water environment simulations, ecosystem research, flood prevention and mitigation, shipping planning, aquaculture, and lake evolution studies. While traditional lake topography methods offer high accuracy, they still suffer from high costs and low efficiency. Summary of the Invention

[0003] The main purpose of the embodiments of the present application is to propose a lake terrain prediction method based on machine learning and fusion of multi-source satellite data to improve the prediction efficiency of lake terrain and reduce costs.

[0004] To achieve the above objectives, an embodiment of the present application proposes a lake terrain prediction method based on machine learning and fusion of multi-source satellite data, the method comprising the following steps:

[0005] Extract the optical characteristics of the target lake from optical satellite images;

[0006] Extracting water characteristics of each water pixel in the target lake from SWOT satellite data;

[0007] Performing spatiotemporal alignment and fusion of the optical feature and the water feature to obtain a fusion feature vector;

[0008] Matching the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then forming a training sample with the real water depth and the fused feature vector corresponding to the pixel;

[0009] Select a machine learning model and initialize the structure or parameters of the machine learning model;

[0010] Using the training samples to train the initialized machine learning model;

[0011] Using the trained machine learning model to predict the water depth value of each water body pixel according to the fusion characteristic vector of each water body pixel;

[0012] The topography of the target lake is determined according to each of the water depth values.

[0013] In some embodiments, extracting the optical characteristics of the target lake from the optical satellite image comprises the following steps:

[0014] The optical characteristics including the band reflectance, band ratio, logarithmic conversion characteristics and water body index of the target lake are extracted from the optical satellite image.

[0015] In some embodiments, extracting water characteristics of each water pixel in the target lake from SWOT satellite data includes the following steps:

[0016] The water characteristics including the water surface elevation, water surface slope, shoreline distance and SWOT water body information of each water body pixel in the target lake are extracted from the SWOT satellite data.

[0017] In some embodiments, selecting a machine learning model and initializing the structure or parameters of the machine learning model include the following steps:

[0018] According to the characteristics of the training samples and computing resources, the machine learning model is selected from a random forest regression model, a gradient boosting regression model, an artificial neural network model, or a convolutional neural network variant model for fusing neighborhood information, and the structure or parameters of the machine learning model are initialized.

[0019] In some embodiments, the step of training the initialized machine learning model using the training samples comprises the following steps:

[0020] Dividing the training samples to obtain a training set;

[0021] Perform parameter learning on the machine learning model using the training set;

[0022] The target hyperparameters of the machine learning model are optimized using cross-validation methods combined with grid search, random search, or Bayesian optimization strategies.

[0023] In some embodiments, determining the topography of the target lake according to each of the water depth values ​​comprises the following steps:

[0024] Converting each of the water depth values ​​into lake bottom elevation using the water surface elevation of the water body pixel in the SWOT satellite data;

[0025] generating a raster map of the lake bottom elevation;

[0026] The raster image is subjected to spatial smoothing filtering, outlier removal and repair, and depth contour generation and mapping to obtain a topographic map of the target lake.

[0027] In some embodiments, before extracting the optical features of the target lake from the optical satellite image, the method further includes the following steps:

[0028] Performing radiometric correction, geometric correction and registration, water body extraction, and cloud, cloud shadow and other interference removal on the optical satellite image;

[0029] Before extracting the water characteristics of each water pixel in the target lake from the SWOT satellite data, the method further includes the following steps:

[0030] The SWOT satellite data were subjected to outlier removal, spatiotemporal matching and grid unification, and water body boundary optimization.

[0031] To achieve the above objectives, another aspect of the present application provides a lake terrain prediction device based on machine learning and fusion of multi-source satellite data, the device comprising:

[0032] An optical feature extraction unit, used to extract the optical features of the target lake from the optical satellite image;

[0033] A water feature extraction unit is used to extract the water features of each water pixel in the target lake from the SWOT satellite data;

[0034] a feature fusion unit, configured to align and fuse the optical feature and the water feature in time and space to obtain a fused feature vector;

[0035] A sample construction unit is used to match the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then form a training sample with the real water depth and the fused feature vector corresponding to the pixel;

[0036] A model initialization unit, configured to select a machine learning model and initialize the structure or parameters of the machine learning model;

[0037] A model training unit, configured to train the initialized machine learning model using the training samples;

[0038] a water depth prediction unit, configured to predict the water depth value corresponding to each water body pixel according to the fusion characteristic vector of each water body pixel using the trained machine learning model;

[0039] A terrain determination unit is used to determine the terrain of the target lake according to each of the water depth values.

[0040] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method when executing the computer program.

[0041] To achieve the above-mentioned purpose, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above-mentioned method when executed by a processor.

[0042] The embodiments of the present application include at least the following beneficial effects:

[0043] This application can extract the optical characteristics of the target lake from optical satellite images; extract the water characteristics of each water pixel in the target lake from SWOT satellite data; align and fuse the optical characteristics and water characteristics in time and space to obtain a fusion characteristic vector; match the measured water depth point of the real water depth to the geographical location of the corresponding pixel, and then form a training sample with the fusion characteristic vector of the real water depth and the corresponding pixel; select a machine learning model and initialize the structure or parameters of the machine learning model; use the training samples to train the initialized machine learning model; use the trained machine learning model to predict the water depth value of the corresponding pixel based on the fusion characteristic vector of each water pixel; and determine the topography of the target lake based on each water depth value. This application uses the trained machine learning model to predict the topography of the entire lake based on the extracted characteristic values, and can efficiently and cost-effectively predict the underwater topography of a large area of ​​lakes, providing important data support for water resource management and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 A flow chart of a lake terrain prediction method based on machine learning and fusion of multi-source satellite data provided in an embodiment of the present application;

[0046] Figure 2 This is an example flow chart of a lake terrain prediction method based on machine learning and fusion of multi-source satellite data provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of the structure of a lake terrain prediction device based on machine learning and fusion of multi-source satellite data provided in an embodiment of the present application;

[0048] Figure 4 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are merely examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0050] It will be understood that the terms "first", "second", etc. used in this application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0051] The terms "at least one", "plurality", "each", "any", etc. used in this application include "at least one", "two" or more, "plurality" or "each", "any" or "any one", "each" or "any one" as used herein.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0053] Before describing the embodiments of the present application in detail, some of the related technologies involved in the embodiments of the present application are first described as follows:

[0054] With the development of remote sensing technology, it has become possible to conduct large-scale, periodic lake monitoring using satellite data. In recent years, the emergence of high-resolution, multi-source satellites (such as the Sentinel and Landsat series) and satellite missions dedicated to surface hydrological observations (such as the SWOT satellite) has provided an unprecedented data foundation for lake topography inversion. At the same time, machine learning technology, due to its powerful nonlinear fitting and complex pattern recognition capabilities, has shown great potential in processing remote sensing big data and Earth science problems. Therefore, combining multi-source satellite data with advanced machine learning algorithms to efficiently and accurately predict lake topography has become a research hotspot in hydrology and remote sensing.

[0055] At present, there are many methods for lake topography measurement:

[0056] (1) Traditional underwater topography measurement:

[0057] Traditional underwater topographic surveys are mainly divided into two categories: direct measurement and acoustic detection. Depth is measured directly in shallow water using a graduated sounding rod; and plumb bob sounding is used, where a weighted measuring line is lowered to the bottom and the water depth is determined by reading the line length. Acoustic detection is usually carried out using a ship-borne platform (including manned vessels or unmanned vessels). For example, a single-beam echo sounder transmits sound waves vertically from the hull to the bottom of the water, calculating the water depth at a single point based on the round-trip time of the sound waves, thereby obtaining a topographic profile directly below the ship's trajectory. For example, a multi-beam echo sounder simultaneously transmits multiple sound beams arranged in a fan shape, measuring the water depth within a wide strip under the ship's path at once, capable of obtaining high-precision, full-coverage underwater topographic data. Side-scan sonar is also used to generate acoustic images of underwater topography, assisting in identifying underwater objects, bottom types, and understanding overall topographic characteristics.

[0058] (2) Satellite optical remote sensing water depth inversion based on empirical models:

[0059] This method uses imagery acquired by satellite optical sensors (such as the multispectral imagers carried by Landsat and Sentinel-2) to estimate water depth by analyzing the selective absorption and scattering properties of water at different spectral bands. The core principle is that light attenuates with increasing depth as it propagates through water, particularly in short-wavelength bands such as blue and green, which have relatively strong penetration, while red and near-infrared bands are rapidly absorbed by water. Empirical models aim to establish a mathematical relationship between the apparent reflectance of satellite imagery (or atmospherically corrected surface reflectance) and actual water depth. This typically requires a sufficient number of measured depth points as calibration data for regression analysis. Common models include single-band log-linear models (which assume a linear relationship between water depth and the logarithm of the reflectance of a particular band) and multi-band log-band ratio models (such as the model proposed by Stumpf et al., which uses the logarithmic ratio of the blue and green band reflectances to mitigate the influence of variations in water optical parameters and some bottom sediments).

[0060] (3) Remote sensing water depth inversion based on physical model (semi-analytical model):

[0061] Compared to empirical models, remote sensing water depth inversion methods based on physical or semi-analytical models attempt to more deeply describe the radiative transfer of light in water. These models not only consider water depth but also explicitly incorporate intrinsic water optical parameters (IOPs, such as absorption coefficient and backscattering coefficient) and bottom reflectance properties (substrate type and reflectivity) into the model framework. The basic idea is to separate the contributions of water depth, water components (such as chlorophyll, suspended sediment, and colored soluble organic matter (CDOM)), and bottom reflectance from the total above-surface reflectance signal observed by satellite by solving the inverse problem of the complex underwater radiative transfer equation (or its simplified form). For example, through iterative optimization, lookup table (LUT) matching, or spectral decomposition techniques, water depth can be inverted given water IOPs and a bottom spectral library. These methods are theoretically more robust, more adaptable to environmental changes, and have the potential to simultaneously infer water optical parameters.

[0062] (4) Water depth estimation based on a small amount of satellite altimetry data combined with images:

[0063] This method utilizes discrete but precise water surface elevation (WSE) anchor points provided by high-precision satellite altimetry (such as the ICESat-2 laser altimeter) and combines them with water body boundaries (shorelines) delineated by quasi-simultaneously acquired optical or radar satellite imagery to estimate water storage changes and local depth characteristics in water bodies such as lakes. ICESat-2's photon-counting lidar can penetrate some water bodies and even directly detect the bottom in shallow areas, providing true water depths in some areas. Even if it fails to penetrate the bottom, the precise WSE it provides is a valuable elevation benchmark. By acquiring multiple periods of water elevation and the corresponding water body extent, a lake area-elevation curve (i.e., a lake curve or reservoir capacity curve) can be constructed to estimate changes in total water storage at different water levels. For a specific shoreline location, its elevation is the water level at that time and can be considered the "zero water depth" reference for that point. Combined with this elevation benchmark, the relative water depth estimated by other remote sensing water depth inversion methods (such as optical empirical models) can be converted into absolute water depth with actual elevation significance.

[0064] (5) Preliminary exploration of water body characteristics research using SWOT satellite data alone:

[0065] The SWOT satellite is designed to provide high-precision global elevation, extent, and slope of surface water bodies. Current research focuses on using SWOT data for water level monitoring, dynamic analysis of water extent, and river flow estimation. Directly retrieving detailed lake underwater topography (not just surface elevation), particularly when integrating it with other types of satellite data, such as multispectral imagery, is still in its early stages.

[0066] (6) Applying machine learning to water depth inversion from a single remote sensing data source:

[0067] In recent years, machine learning algorithms have garnered increasing attention and application in remote sensing water depth retrieval due to their powerful nonlinear modeling capabilities and ability to automatically learn features from complex data. Numerous studies have attempted to utilize various machine learning models (e.g., random forests, support vector machines / regression, gradient boosting trees, artificial neural networks, and even deep learning models such as convolutional neural networks) to predict water depth using single-source optical satellite imagery (e.g., Landsat or Sentinel-2 imagery). These methods typically use raw imagery band reflectance, calculated spectral indices (e.g., various water indices and color indices), and texture features as input features, and measured water depth data as training labels. Models are trained through supervised learning, aiming to establish a highly nonlinear mapping between image features and water depth. Compared to traditional empirical models, machine learning methods often better account for the influence of complex water conditions and bottom sediment variations, sometimes achieving higher prediction accuracy, and without the need for predefined mathematical formulas.

[0068] Disadvantages of existing technology:

[0069] (1) Traditional underwater topography measurement:

[0070] While highly accurate, these systems are expensive, labor-intensive, and limited in scope, resulting in low efficiency and timeliness, making it difficult to rapidly and repeatedly conduct dynamic monitoring. This is especially true in shallow waters, such as in some urban landscape lakes, where measurements can be affected by obstacles like passing tourist boats and aquatic plants.

[0071] (2) Empirical / semi-analytical optical remote sensing water depth inversion method:

[0072] ① Strong dependence on water quality: The accuracy is seriously affected by the optical properties of the water body (such as turbidity, chlorophyll concentration, and bottom type). The model has poor universality and usually requires local calibration for different lakes or different periods.

[0073] ②High requirements for atmospheric correction: Atmospheric scattering and absorption have a significant impact on water body signals. Accurate atmospheric correction is very critical, but it is often difficult to achieve perfectly.

[0074] ③ Limited water depth detection range: In deep or turbid water, the optical signal decays rapidly, making detection difficult. In extremely shallow water, the reflection from the bottom has a significant impact.

[0075] ④ Dependence on measured data: Most empirical models and some semi-analytical models still require a certain amount of measured water depth data for model parameter calibration and verification, which limits their effective application in areas with no or little data.

[0076] Limitations of using SWOT data alone for topographic inversion: SWOT directly measures water surface elevation and extent, not underwater topography. While its data can be used to infer changes in water volume or constrain water boundaries, directly and accurately retrieving detailed lakebed topography is difficult. Furthermore, SWOT's spatial resolution (e.g., its gridded water surface elevation product) is still limited compared to fine-grained topographic depictions.

[0077] ⑥ Machine learning methods based on a single remote sensing data source: While machine learning improves modeling capabilities, the information provided by a single data source is limited. For example, optical imagery alone cannot accurately distinguish differences in reflected signals caused by changes in water depth and water composition, nor can it directly obtain accurate absolute water surface elevation as a benchmark. This limits further improvements in model accuracy and robustness in complex hydrological conditions.

[0078] ⑦ Insufficient data fusion: Existing technologies are rarely able to effectively integrate the unique advantages of structural hydrological information such as the spectral information of optical images and the high-precision water surface elevation and water body range provided by SWOT satellites, and fail to fully tap the collaborative potential of multi-source data to overcome the limitations of a single data source.

[0079] In response to the problems of the above-mentioned existing technologies, such as high cost, low efficiency, significant impact of water quality, poor universality, limited deep-water detection capabilities, and failure to fully integrate the advantages of multi-source satellite data, this application aims to solve the following technical problems:

[0080] (1) How to effectively integrate the rich spectral information provided by satellite multispectral / hyperspectral band images with the high-precision water surface elevation, water body range and water surface slope data provided by SWOT satellites to more comprehensively characterize the water-light-topography characteristics of lakes.

[0081] (2) How to utilize the powerful nonlinear modeling capabilities of machine learning to learn and accurately predict the underwater topography of lakes from the fused multi-source data features, thereby reducing dependence on the assumptions of traditional empirical models and improving the universality and prediction accuracy of the model.

[0082] (3) How to use the precise water surface elevation provided by SWOT data as a benchmark to assist in distinguishing signal variations in optical images caused by changes in water depth and water composition, and to help constrain the boundary conditions and elevation references for terrain prediction.

[0083] (4) Provide a lake topography prediction technology solution that is more efficient and less costly than traditional measurement methods, and more accurate and more applicable than methods based on a single remote sensing data source. It is particularly suitable for rapid topographic mapping of lakes that lack measured data or cover large areas.

[0084] Based on this, this application proposes a lake topography prediction method based on multi-source satellite data and machine learning. First, satellite band imagery and SWOT satellite data are acquired and preprocessed to extract multidimensional features related to lake topography. Next, the model is trained and adjusted using water depth data measured by unmanned vessels and other sources. Finally, the trained machine learning model is used to predict the entire lake's topography based on the extracted feature values. This application can efficiently and cost-effectively predict the underwater topography of large lakes, providing important data support for water resource management and environmental protection.

[0085] The embodiment of the present application provides a lake terrain prediction method based on machine learning fusion of multi-source satellite data, which relates to the field of machine learning technology. The lake terrain prediction method based on machine learning fusion of multi-source satellite data provided by the embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network; the software can be an application that implements a lake terrain prediction method based on machine learning fusion of multi-source satellite data, etc., but is not limited to the above forms.

[0086] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0087] Reference Figure 1The present application embodiment provides a lake terrain prediction method based on machine learning and fusion of multi-source satellite data. The method may include but is not limited to S1 to S8, as follows:

[0088] S1: Extract the optical characteristics of the target lake from optical satellite images;

[0089] S2: Extracting water characteristics of each water pixel in the target lake from SWOT satellite data;

[0090] S3: performing spatiotemporal alignment and fusion of the optical feature and the water feature to obtain a fusion feature vector;

[0091] S4: Matching the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then forming a training sample with the real water depth and the fused feature vector corresponding to the pixel;

[0092] S5: Select a machine learning model and initialize the structure or parameters of the machine learning model;

[0093] S6: Using the training samples to train the initialized machine learning model;

[0094] S7: using the trained machine learning model to predict the water depth value corresponding to each water pixel according to the fusion characteristic vector of each water pixel;

[0095] S8: Determine the topography of the target lake according to each of the water depth values.

[0096] Optionally, extracting the optical characteristics of the target lake from the optical satellite image comprises the following steps:

[0097] The optical characteristics including the band reflectance, band ratio, logarithmic conversion characteristics and water body index of the target lake are extracted from the optical satellite image.

[0098] Optionally, extracting the water characteristics of each water pixel in the target lake from the SWOT satellite data comprises the following steps:

[0099] The water characteristics including the water surface elevation, water surface slope, shoreline distance and SWOT water body information of each water body pixel in the target lake are extracted from the SWOT satellite data.

[0100] Optionally, selecting a machine learning model and initializing a structure or parameters of the machine learning model comprises the following steps:

[0101] According to the characteristics of the training samples and computing resources, the machine learning model is selected from a random forest regression model, a gradient boosting regression model, an artificial neural network model, or a convolutional neural network variant model for fusing neighborhood information, and the structure or parameters of the machine learning model are initialized.

[0102] Optionally, the training of the initialized machine learning model using the training samples comprises the following steps:

[0103] Dividing the training samples to obtain a training set;

[0104] Perform parameter learning on the machine learning model using the training set;

[0105] The target hyperparameters of the machine learning model are optimized using cross-validation methods combined with grid search, random search, or Bayesian optimization strategies.

[0106] Optionally, determining the topography of the target lake according to each of the water depth values ​​comprises the following steps:

[0107] Converting each of the water depth values ​​into lake bottom elevation using the water surface elevation of the water body pixel in the SWOT satellite data;

[0108] generating a raster map of the lake bottom elevation;

[0109] The raster image is subjected to spatial smoothing filtering, outlier removal and repair, and depth contour generation and mapping to obtain a topographic map of the target lake.

[0110] Optionally, before extracting the optical features of the target lake from the optical satellite image, the method further comprises the following steps:

[0111] Performing radiometric correction, geometric correction and registration, water body extraction, and cloud, cloud shadow and other interference removal on the optical satellite image;

[0112] Before extracting the water characteristics of each water pixel in the target lake from the SWOT satellite data, the method further includes the following steps:

[0113] The SWOT satellite data were subjected to outlier removal, spatiotemporal matching and grid unification, and water body boundary optimization.

[0114] Next, the solution of the embodiment of the present application will be introduced and explained in detail with reference to specific application examples.

[0115] Reference Figure 2, the embodiment of the present application aims to propose a method for predicting the underwater topography of lakes. The key technical solutions include deep fusion of multi-source satellite remote sensing data (optical images and SWOT satellite observations) and combination with advanced machine learning algorithms. The basic idea is to make full use of the optical response characteristics of optical satellite images to water components and water column attenuation effects, combined with the centimeter-level water surface elevation (WSE), high-precision water body range and water surface slope and other multi-dimensional information provided by SWOT satellites, to construct a comprehensive feature set with richer information and more significant physical constraints. Subsequently, the complex nonlinear mapping relationship between these geophysical features and the actual water depth is mined and learned through a machine learning model, ultimately achieving high-precision and wide-coverage inversion of the underwater topography of the lake.

[0116] The detailed steps of this embodiment are as follows:

[0117] ①Data acquisition.

[0118] Optical satellite image acquisition:

[0119] Collect archived or newly acquired multispectral / hyperspectral satellite imagery covering the target lake area. Data sources may include, but are not limited to, publicly available data such as Landsat series (such as OLI / TIRS), Sentinel-2 (MSI), and Sentinel-3 (OLCI), or high-resolution satellite imagery such as Planet, WorldView, and GeoEye.

[0120] SWOT satellite data acquisition:

[0121] Acquire SWOT (Surface Water and Ocean Topography) satellite data products that are as consistent as possible with the optical imagery or within the same hydrological cycle. Key data products include: high-precision water surface elevation (WSE) raster data provided by L2_HR_KaRIn (Ka-band Radar Interferometer High Rate), water body classification data, and potential water surface slope (WSS) products.

[0122] Auxiliary data acquisition (optional):

[0123] Digital Elevation Models (DEMs): These include ASTER GDEM and SRTM DEM, used to assist in shoreline delineation, analyze surrounding terrain, and correct results. Historically measured water depth data: Small amounts of high-precision historical water depth data acquired through sonar sounding and LiDAR (Light Detection and Ranging) sounding are primarily used as training labels for machine learning models and / or for independent accuracy verification. Hydrological station observation data: Synchronous or quasi-synchronous water level observations from hydrological stations in the target lake or adjacent areas are used to cross-validate or bias-correct the WSE data provided by SWOT.

[0124] ②Data preprocessing.

[0125] Optical image preprocessing:

[0126] Radiometric correction: converts the original DN (Digital Number) value into the reflectance or radiance value of the top of the atmosphere (TOA). Atmospheric correction: eliminates the influence of atmospheric scattering and absorption on the reflectance of ground objects and obtains accurate surface reflectance. Mature algorithms or tools such as FLAASH, QUAC, Sen2Cor, LaSRC, etc. can be used. Geometric correction and registration: perform high-precision geometric correction on the image to ensure that it is accurately spatially registered with SWOT data, DEM and other auxiliary data, and the error is controlled within one pixel. Removal of clouds, cloud shadows and other interference: use quality assessment bands or specific algorithms (such as Fmask, CFmask) to identify and mask invalid pixels such as clouds, cloud shadows, and dense fog. Accurate water body extraction: use a combination of the improved normalized difference water index (MNDWI), the automated water extraction index (AWEI), object-oriented classification methods or deep learning segmentation models to extract the high-precision lake water range.

[0127] SWOT data preprocessing:

[0128] Outlier processing removes significant outliers or outliers from data such as WSE and WSS. Spatiotemporal matching and grid unification reprojects SWOT rasterized products (such as WSE) to a geographic coordinate system consistent with the optical imagery and resamples them to a unified analysis grid resolution (e.g., consistent with the optical imagery resolution or setting an optimal analysis scale). Water body boundary optimization utilizes the high-precision water body extent classification data provided by SWOT to further optimize or constrain water body boundaries extracted from optical imagery.

[0129] ③ Feature extraction and multi-source data fusion.

[0130] Feature extraction based on optical images:

[0131] Band reflectance directly uses preprocessed surface reflectance values ​​for each band (blue, green, red, near-infrared, shortwave infrared, etc.). Band ratios calculate depth-sensitive band ratios, such as the green / blue band ratio, the red / green band ratio, and the near-infrared / green band ratio, to mitigate the influence of the bottom matrix and some water components. Logarithmic transformation features utilize logarithmically transformed single bands or band ratios (e.g., ln(blue), ln(green), ln(red), ln(NIR / Green), ln(blue / green)). These features are widely used in classic semi-theoretical / empirical water depth inversion models and can linearize the relationship between water depth and radiation signals. Water indices, such as MNDWI and NDWI, may exhibit a certain correlation with water depth under certain conditions (e.g., extremely shallow and clear water bodies).

[0132] Feature extraction based on SWOT data:

[0133] High-precision water surface elevation (WSE), the high-precision WSE value corresponding to each water body pixel, is one of the core inputs of this method and provides a key reference plane for water depth inversion. Water surface slope (WSS), the slope of the water surface in the X and Y directions directly provided or calculated by SWOT data. WSS can reflect local hydrodynamic conditions and may indirectly indicate the undulations of underwater terrain. Shoreline distance, based on the precise lake shoreline extracted by SWOT or optical imagery, calculates the Euclidean distance from each water body pixel to the nearest shoreline. This feature utilizes the prior knowledge that water depth generally increases with increasing distance from the shore. SWOT water body information, the water body pixel classification or water body range information provided by SWOT is used to accurately define the effective area for model training and prediction.

[0134] Feature vector construction and fusion:

[0135] Various features extracted from optical images and SWOT data are spatially aligned and fused at a unified pixel (or grid unit) scale to construct a high-dimensional feature vector for each water body pixel.

[0136] ④Machine learning model construction and water depth inversion.

[0137] Training sample construction:

[0138] Labeled Data (True Water Depth): Ideally, use measured water depth data (e.g., sonar or LiDAR measurements) covering a representative portion of the study area as the true label. Accurately match the measured water depth points to the geographic location of the corresponding pixel, and extract the fused feature vector for that pixel to form a training sample pair of (feature vector, measured water depth).

[0139] Machine learning model selection and configuration:

[0140] Use a machine learning model suitable for handling high-dimensional inputs and complex nonlinear regression tasks. Alternative models include: Random Forest Regression (RFR), Gradient Boosting Regression (GBR) such as XGBoost, LightGBM, and CatBoost, Support Vector Regression (SVR), Artificial Neural Networks (ANN) such as the Multilayer Perceptron (MLP), or Convolutional Neural Network (CNN) variants that incorporate neighborhood information (if local pixel neighborhoods are used as image patch input). Select an appropriate model based on data characteristics and computing resources, and perform preliminary architecture design and parameter configuration.

[0141] Model training and optimization:

[0142] Dataset partitioning involves splitting the constructed training sample set into a training set and independent validation / test sets according to a specific ratio (e.g., 70 / 30 or 80 / 20). Model training involves using the training set to learn the parameters of the selected machine learning model. Hyperparameter optimization involves systematically optimizing the model's key hyperparameters using cross-validation techniques combined with strategies such as grid search, random search, or Bayesian optimization to achieve optimal generalization performance.

[0143] ⑤Generation and output of lake underwater terrain.

[0144] Model-based water depth prediction:

[0145] The trained and optimized optimal machine learning model is applied to the fused feature vectors of all water pixels in the target lake area to predict the water depth value of each pixel. Using the high-precision water surface elevation (WSE_SWOT) of the corresponding pixel provided by SWOT, the model-predicted water depth (Depth_predicted) is converted to the lakebed elevation (Bed_Elevation):

[0146] Bed_Elevatio=WSE_SWOT-Depth_predicted;

[0147] Output:

[0148] Generate a lake bottom elevation model raster map.

[0149] Post-processing of results:

[0150] Perform necessary post-processing on the generated lake bottom elevation map or water depth map, such as spatial smoothing filtering, outlier removal and repair, and depth contour generation and mapping.

[0151] ⑥ Accuracy verification and evaluation.

[0152] Independent validation dataset:

[0153] The final lake topography product was rigorously evaluated for accuracy using independent measured water depth data (the "test set") that was not involved in model training and optimization.

[0154] Quantitative evaluation indicators:

[0155] Root mean square error, mean absolute error, coefficient of determination, mean relative error, or percent deviation.

[0156] In summary, this embodiment includes the following technical solutions:

[0157] Key technical points:

[0158] (1) Multi-source heterogeneous data fusion strategy: How to effectively integrate the optical spectral characteristics at the pixel level with the higher-level structural hydrological information provided by SWOT (such as WSE grid, water body extent vector / grid, water surface slope) at the feature level or model level to maximize information complementarity.

[0159] (2) Innovative feature extraction and application based on SWOT data: How to transform SWOT data such as WSE, water body extent, and water surface slope into the most effective input features for machine learning models to predict lake bottom topography. For example, using WSE as an absolute elevation reference to constrain relative water depth, or using multi-temporal WSE and water body extent changes to infer lake storage characteristics to assist in topographic inversion.

[0160] (3) Optimization of machine learning models for fusion features: Select or design a machine learning model that can effectively process the fusion of multiple types of features such as spectrum, elevation, slope, distance, etc., and perform targeted structural optimization, loss function design or training strategy adjustment to improve the learning efficiency and prediction accuracy of the model.

[0161] (4) Model adaptability and transferability in the case of sparse or no ground truth: How to train a lake topography prediction model with certain generalization capabilities through semi-supervised learning, transfer learning, or generating pseudo-labels using SWOT data when the measured water depth data is extremely limited?

[0162] More specifically:

[0163] (1) A lake topography prediction method obtains satellite (multispectral / hyperspectral) band images and SWOT satellite data (including at least water surface elevation and / or water body range) of the target lake, preprocesses and extracts features from the two types of data, fuses the extracted band image features with the SWOT data features, and uses the fused feature set to predict the underwater topography of the lake through a machine learning model.

[0164] (2) Specific method of feature fusion: Protect the technical solution of combining the spectral reflectance, water depth correlation index and other features of satellite band images with the water surface elevation, water surface slope, distance to the shoreline and other features of SWOT at the pixel (or grid) level into a multi-dimensional feature vector for machine learning model input.

[0165] (3) Specific application of machine learning models: The protection method adopts specific categories or improved machine learning models (such as random forests, gradient boosting machines, neural networks, etc. optimized for this type of fusion data) to achieve terrain prediction by learning the mapping relationship between fusion features and water depth (or lake bottom elevation).

[0166] (4) Methods of using SWOT data to constrain models or calibrate results: Protect the method of using the high-precision water surface elevation provided by SWOT as part of the training label (for example, if the lake bottom elevation is predicted, the lake bottom elevation = SWOT WSE - predicted water depth), or as an absolute elevation benchmark for calibration of the prediction results.

[0167] The lake topography prediction system based on the above method includes an integrated system consisting of a data acquisition module, a preprocessing module, a feature extraction and fusion module, a machine learning model training and prediction module, and a topography result generation and output module. In the absence or sparseness of measured water depths, SWOT data can be used to assist in generating training samples or improving model robustness.

[0168] Beneficial effects of this embodiment:

[0169] Originality: This embodiment addresses the bottleneck problems of traditional lake underwater topography measurement methods, such as high cost, low operating efficiency, poor timeliness, and difficulty in application to vast, complex, or sparsely populated waters. It focuses on major demands such as sustainable use of water resources, precise management of water environment, and flood prevention and disaster reduction, and proposes a lake underwater topography intelligent inversion technology that deeply integrates multiple types of satellite-borne remote sensing observation data with advanced machine learning algorithms. This technology takes improving the efficiency, accuracy, and coverage of underwater topography mapping as its core goal, and innovatively constructs a collaborative inversion framework for optical satellite image features (reflecting the optical properties of water bodies and water column attenuation) and SWOT satellite unique observation data (high-precision water surface elevation WSE, water body range, and water surface slope). By constructing a multi-source feature set with richer information and stronger physical constraints, and using machine learning models to mine the complex nonlinear relationship between these features and underwater topography, a revolutionary breakthrough in traditional methods has been achieved. This method changes the previous limitations of insufficient information from a single remote sensing data source or overly strong physical model assumptions, and provides a new technical approach for quickly and accurately obtaining underwater topography information of large-scale lakes. It has significant originality and technological leadership.

[0170] Value: The technical framework proposed in this embodiment closely addresses the urgent need to improve the ability to obtain basic geographic information about lakes, and its value is reflected in multiple aspects. First, by efficiently and cost-effectively utilizing multi-source satellite remote sensing data to replace or supplement traditional measurement methods, the economic and time costs of lake underwater topography mapping are greatly reduced, making it possible to conduct topographic surveys of lakes that lack field data. Second, the high-precision, timely lake underwater topography data (DBEM / DBM) produced by this technology are indispensable key input parameters for hydrology, hydrodynamics, water environment, and ecological models. They can significantly improve the simulation accuracy and prediction reliability of related models, thereby providing solid data support for accurate reservoir operation, water balance analysis, flood risk assessment, water quality simulation, ecosystem health assessment, waterway planning, and water-related engineering design. This technology has the advantages of non-contact, large-scale, and repeatable observation, making it particularly suitable for monitoring remote areas, environmentally sensitive areas, or dynamically changing water bodies. It provides core technical support for the refined, intelligent management and sustainable development of lake water resources, and has important strategic significance and broad application prospects for addressing water security challenges in the context of climate change.

[0171] Effectiveness: The lake underwater topography prediction technology based on multi-source satellite data fusion and machine learning proposed in this embodiment has undergone sufficient methodological design and case verification (in the future or already conducted), and has demonstrated strong practical application effectiveness. This technology can effectively overcome the bottleneck of insufficient information from a single remote sensing data source. The high-precision water surface elevation information provided by the SWOT satellite provides a direct vertical benchmark constraint for water depth inversion, significantly improving the prediction accuracy. The introduction of the machine learning model enables it to adaptively learn complex relationships from the data, reduces dependence on prior water body optical parameters, and enhances the universality of the method to lakes with different water quality conditions. The underwater topography results produced by this method (such as RMSE, MAE, R 2 The technology is expected to achieve or surpass the accuracy of some existing remote sensing inversion methods (measured by indicators such as geography, geography, and geography), and offers unparalleled advantages in data acquisition cost, mapping efficiency, and coverage. This technology has a clear workflow, and key steps (such as data preprocessing, feature extraction, model training, and terrain generation) are easily automated and standardized. It has great potential for transitioning from research to operational applications, enabling rapid response to emergency monitoring needs (such as terrain assessment within flood inundation areas), and possesses significant practical benefits and operability.

[0172] Systematic aspects: The lake underwater terrain prediction technology proposed in this embodiment is not a simple superposition of a single technology, but a complete technical system from multi-source heterogeneous data acquisition, collaborative preprocessing, deep feature engineering, intelligent model construction to final product generation and verification. This system emphasizes the collaborative observation advantages of optical remote sensing and radar remote sensing (SWOT's KaRIn), and systematically designs feature extraction and fusion strategies for water column optical properties, water surface geometry and shoreline spatial relationships to maximize information utilization. The application of machine learning models constitutes the core of intelligent inversion, realizing end-to-end mapping from high-dimensional features to underwater terrain. Furthermore, this method ensures that the predicted "water depth" can be accurately converted into "lake bottom elevation" with a clear elevation benchmark by directly linking with the WSE data provided by SWOT, forming a systematic three-dimensional spatial information cognition solution from the water surface to the bottom. This technical system not only focuses on the accuracy of terrain prediction, but also takes into account the automation potential and promotion applicability of the method. It provides a key link for building digital twin lakes and realizing normalized remote sensing monitoring and intelligent management of lakes. It is a systematic and innovative solution in the field of lake information science.

[0173] Reference Figure 3 The present application also provides a lake terrain prediction device based on machine learning and fusion of multi-source satellite data, which can implement the above-mentioned lake terrain prediction method based on machine learning and fusion of multi-source satellite data. The device includes:

[0174] An optical feature extraction unit, used to extract the optical features of the target lake from the optical satellite image;

[0175] A water feature extraction unit is used to extract the water features of each water pixel in the target lake from the SWOT satellite data;

[0176] a feature fusion unit, configured to align and fuse the optical feature and the water feature in time and space to obtain a fused feature vector;

[0177] A sample construction unit is used to match the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then form a training sample with the real water depth and the fused feature vector corresponding to the pixel;

[0178] A model initialization unit, configured to select a machine learning model and initialize the structure or parameters of the machine learning model;

[0179] A model training unit, configured to train the initialized machine learning model using the training samples;

[0180] a water depth prediction unit, configured to predict the water depth value corresponding to each water body pixel according to the fusion characteristic vector of each water body pixel using the trained machine learning model;

[0181] A terrain determination unit is used to determine the terrain of the target lake according to each of the water depth values.

[0182] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0183] The present application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of the present application. The electronic device can be any smart terminal, such as a tablet computer or an in-vehicle computer.

[0184] It can be understood that the contents of the above method embodiments are all applicable to the embodiments of the present device, the functions specifically implemented by the embodiments of the present device are the same as those of the method of the present application, and the beneficial effects achieved are also the same as those achieved by the method of the present application.

[0185] See also Figure 4 , Figure 4 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0186] The processor 401 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0187] The memory 402 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 402 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 402 and is called by the processor 401 to execute the methods of the embodiments of this application.

[0188] Input / output interface 403, used to implement information input and output;

[0189] Communication interface 404, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0190] Bus 405 , which transmits information between various components of the device (e.g., processor 401 , memory 402 , input / output interface 403 , and communication interface 404 );

[0191] The processor 401 , the memory 402 , the input / output interface 403 and the communication interface 404 are connected to each other in communication within the device via a bus 405 .

[0192] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method of the present application is implemented.

[0193] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiment, the functions specifically implemented by the present storage medium embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0194] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0195] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0196] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0197] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0198] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0199] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0200] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0201] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0202] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0203] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0204] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: various media that can store programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0205] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A lake topography prediction method based on machine learning and fusion of multi-source satellite data, characterized in that: The method comprises the following steps: Extract the optical characteristics of the target lake from optical satellite images; Extracting water characteristics of each water pixel in the target lake from SWOT satellite data; Performing spatiotemporal alignment and fusion of the optical feature and the water feature to obtain a fusion feature vector; Matching the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then forming a training sample with the real water depth and the fused feature vector corresponding to the pixel; Select a machine learning model and initialize the structure or parameters of the machine learning model; Using the training samples to train the initialized machine learning model; Using the trained machine learning model to predict the water depth value of each water body pixel according to the fusion characteristic vector of each water body pixel; The topography of the target lake is determined according to each of the water depth values.

2. The lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to claim 1 is characterized in that: The method of extracting the optical characteristics of the target lake from the optical satellite image comprises the following steps: The optical characteristics including the band reflectance, band ratio, logarithmic conversion characteristics and water body index of the target lake are extracted from the optical satellite image.

3. The lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to claim 1 is characterized in that: The method of extracting water characteristics of each water pixel in the target lake from the SWOT satellite data comprises the following steps: The water characteristics including the water surface elevation, water surface slope, shoreline distance and SWOT water body information of each water body pixel in the target lake are extracted from the SWOT satellite data.

4. The lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to claim 1 is characterized in that: The step of selecting a machine learning model and initializing the structure or parameters of the machine learning model comprises the following steps: According to the characteristics of the training samples and computing resources, the machine learning model is selected from a random forest regression model, a gradient boosting regression model, an artificial neural network model, or a convolutional neural network variant model for fusing neighborhood information, and the structure or parameters of the machine learning model are initialized.

5. The lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to claim 1 is characterized in that: The step of training the initialized machine learning model using the training samples comprises the following steps: Dividing the training samples to obtain a training set; Perform parameter learning on the machine learning model using the training set; The target hyperparameters of the machine learning model are optimized using cross-validation methods combined with grid search, random search, or Bayesian optimization strategies.

6. The lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to claim 1 is characterized in that: Determining the topography of the target lake according to each of the water depth values ​​comprises the following steps: Converting each of the water depth values ​​into lake bottom elevation using the water surface elevation of the water body pixel in the SWOT satellite data; generating a raster map of the lake bottom elevation; The raster image is subjected to spatial smoothing filtering, outlier removal and repair, and depth contour generation and mapping to obtain a topographic map of the target lake.

7. A lake terrain prediction method based on machine learning and fusion of multi-source satellite data according to any one of claims 1 to 6, characterized in that: Before extracting the optical features of the target lake from the optical satellite image, the method further includes the following steps: Performing radiometric correction, geometric correction and registration, water body extraction, and cloud, cloud shadow and other interference removal on the optical satellite image; Before extracting the water characteristics of each water pixel in the target lake from the SWOT satellite data, the method further includes the following steps: The SWOT satellite data were subjected to outlier removal, spatiotemporal matching and grid unification, and water body boundary optimization.

8. A lake topography prediction device based on machine learning and fusion of multi-source satellite data, characterized in that: The device comprises: An optical feature extraction unit, used to extract the optical features of the target lake from the optical satellite image; A water feature extraction unit is used to extract the water features of each water pixel in the target lake from the SWOT satellite data; a feature fusion unit, configured to align and fuse the optical feature and the water feature in time and space to obtain a fused feature vector; A sample construction unit is used to match the measured water depth points of the real water depth to the geographical locations of the corresponding pixels, and then form a training sample with the real water depth and the fused feature vector corresponding to the pixel; A model initialization unit, configured to select a machine learning model and initialize the structure or parameters of the machine learning model; A model training unit, configured to train the initialized machine learning model using the training samples; a water depth prediction unit, configured to predict the water depth value corresponding to each water body pixel according to the fusion characteristic vector of each water body pixel using the trained machine learning model; A terrain determination unit is used to determine the terrain of the target lake according to each of the water depth values.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images