Reservoir dam identification method, device and equipment based on satellite interference height measurement and remote sensing image, medium and product

Abstract

The invention discloses a reservoir dam identification method, device and equipment based on satellite interference height measurement and remote sensing images, a medium and a product, and relates to the field of image identification. The method comprises the following steps: acquiring water surface elevation data; preprocessing the water surface elevation data to obtain a longitude and latitude-height sequence, and determining longitude and latitude coordinate information of potential points of the dam based on a preset elevation mutation threshold value; carrying out vectorization processing on the CSV coordinate file, carrying out image interception, buffer area generation and wave band screening processing based on a satellite remote sensing image, and determining an image file; and inputting the image file into a dam body identification model, and outputting a reservoir dam prediction point location so as to realize identification of the reservoir dam. The dam body identification model is determined by adopting a Bayesian optimization algorithm and carrying out hyper-parameter iteration optimization on the improved YOLO11 model according to the image file with the annotation information. The invention aims to efficiently and reliably realize dam automatic identification.

Description

Technical Field

[0001] This application relates to the field of image recognition, and in particular to a method, apparatus, equipment, medium and product for identifying reservoir dams based on satellite interferometric altimetry and remote sensing images. Background Technology

[0002] In core scenarios such as water resource regulation, flood control and disaster reduction, and watershed ecological protection, dams, as critical water conservancy infrastructure, require accurate identification and dynamic monitoring as a core prerequisite for achieving refined watershed management. Currently known technologies suffer from the following problems: The dam database has significant gaps, and the identification of new and small dams is severely inadequate: Mainstream dam datasets generally suffer from structural flaws, prioritizing historical data over newly constructed dams and large dams over small dams. While some databases cover thousands of large dams, their data sources heavily rely on registration information from regional water authorities. Newly constructed dams suffer from delays in registration and a lack of transparency in some areas. Some datasets only record a few thousand dams built after 2000, but the number can be increased several times over after supplementary identification using remote sensing technology, indicating insufficient coverage of new dams. More critically, the lack of coverage for small dams is even more pronounced: while some databases include tens of thousands of dams, statistical models suggest that the actual number of small dams meeting basic monitoring standards far exceeds existing records. These data gaps prevent the hydrological regulation effects and flood control risks of numerous newly constructed and small dams from being assessed in watershed management, significantly increasing the difficulty of watershed water resource allocation and flood control decisions.

[0003] Traditional recognition technologies rely on manual labor and low-precision data, resulting in both low efficiency and low accuracy. Existing dam identification technologies primarily rely on two approaches, both of which have significant limitations. First, there's the manual interpretation of satellite imagery. In some datasets, dam coordinates require professionals to interpret 1°×1° satellite image patches frame by frame. Furthermore, subjective human judgment is susceptible to interference from ground features, leading to missed detections of small or hidden dams. Second, there's the delineation of candidate areas based on low-precision hydrological data. Early studies, for example, used 1km grid precipitation data and rough watershed boundaries to identify potential dam areas. This often results in a large candidate area, requiring subsequent image processing and data storage to handle massive amounts of irrelevant information, causing a surge in platform resource consumption.

[0004] Insufficient utilization of satellite imagery and model optimization, resulting in poor adaptability to complex scenes: At the satellite imagery application level, related technologies generally use RGB (red, green, and blue) bands for identification, without optimizing band combinations based on the spectral differences between the dam and surrounding features. Related research indicates that in densely vegetated areas, RGB bands can easily confuse the dam with surrounding mountain shadows and vegetation cover. At the model performance level, the problem of weak feature extraction capabilities for small dams remains unresolved; basic identification models generally have low accuracy for small dams, making it difficult to meet the actual watershed monitoring requirements for "full-size dam coverage."

[0005] The lack of large-scale watershed validation leads to significant performance degradation of the model during deployment. Current dam identification technologies largely rely on small sample sets and simple scenarios for validation, lacking large-scale testing across regions and diverse terrains. The validation sets of some datasets have limited coverage and are concentrated in simple terrains with minimal interference, such as plains and irrigation areas, failing to cover complex terrains like mountain valleys and high-altitude watersheds. In practical applications, models in mountainous regions suffer from slow inference speeds and increased false detection rates due to terrain occlusion and feature confusion caused by large elevation differences and complex terrain features.

[0006] In summary, the triple shortcomings in "database integrity," "recognition efficiency," and "adaptability to complex scenarios" prevent dam identification from meeting the actual needs of global watershed management for "full-size, full-area, and high-precision" dam identification. Therefore, it is crucial to overcome the core bottlenecks of difficulty in identifying new and small dams, large candidate areas, and low accuracy in complex scenarios, and to provide an efficient and reliable technical path for large-scale, automated dam identification. Summary of the Invention

[0007] The purpose of this application is to provide a method, device, equipment, medium, and product for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery, which can efficiently and reliably achieve automatic dam identification.

[0008] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery, including: Acquire water surface elevation data; the water surface elevation data is the water body elevation data of each observation point at the watershed scale obtained by SWOT satellite using satellite radar interferometry technology; The water surface elevation data is preprocessed to obtain a latitude-longitude sequence, and the latitude-longitude coordinates of potential dam points are determined based on a preset elevation change threshold. The CSV coordinate file is vectorized, and image cropping, buffer generation, and band filtering are performed based on satellite remote sensing imagery to determine the image file. The image file includes information such as the dam number, satellite type, and band combination. The band combination adopts near-infrared enhancement combination. The CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinates of the potential dam points in CSV format. The image files are input into the dam identification model, which outputs predicted dam locations to achieve dam identification. The dam identification model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on image files with labeled information. The labeled information includes dam locations. The improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure. The ACmix attention module enhances the feature extraction capability of linear edges and grayscale differences of the dam body. The BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy. The reparameterized detection head structure is used to improve the regression accuracy and model convergence stability of the dam body.

[0009] In one embodiment, the expression corresponding to the water body elevation data is: ; in, For water body elevation data; Initial elevation data for radar signal inversion; This is the atmospheric delay error correction amount; This is the correction amount for surface roughness error; This is the correction amount for satellite orbit deviation error.

[0010] In one embodiment, the water surface elevation data is preprocessed to obtain a latitude-longitude sequence, and the latitude-longitude coordinates of potential dam points are determined based on a preset elevation change threshold. Specifically, this includes: The watershed range is clipped based on the vector boundary of the target study area to perform interference removal processing on the water surface elevation data, and non-target water body data outside the watershed are removed to obtain the water surface elevation data after interference removal. Based on the water surface elevation data after interference elimination, the discrete elevation observation points are sorted according to the direction of water flow to form a latitude-longitude sequence; For latitude-longitude sequence, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold.

[0011] In one embodiment, for latitude-longitude sequence, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold, specifically including: Determine the preset elevation change threshold based on the dam water level difference. ; For latitude-longitude-height series, when the elevation difference between adjacent elevation observation points along the direction of water flow is greater than the elevation change threshold... At that time, the midpoint of the line connecting the two observation points is taken as the suspected point of sudden drop; Spatial continuity verification is performed on the suspected sudden drop points to determine the latitude and longitude coordinates of potential dam points; wherein, if the elevation difference between the three observation points before and after the suspected sudden drop point does not exceed the elevation change threshold, then the dam is considered to have a potential dam point. In such cases, single-point abrupt changes caused by natural waterfalls or reef obstruction in the river channel are excluded, and these are identified as potential dam points.

[0012] In one embodiment, the CSV coordinate file is vectorized, and image cropping, buffer generation, and band filtering based on satellite remote sensing imagery are performed to determine the image file. Specifically, this includes: The CSV coordinate file is vectorized, and a clipping boundary is generated by setting the buffer radius based on the dam size and the resolution of the satellite remote sensing image, with each potential dam coordinate point as the center. Based on the Sentinel-2 SR Harmonized multispectral dataset, with the dam center coordinates as spatial constraints, a time range and cloud cover threshold were set to filter satellite remote sensing images under cloudless or low-cloud conditions, and near-infrared enhancement was performed for visual enhancement to obtain the filtered near-infrared enhanced images. The near-infrared enhanced images after screening are cropped according to the cropping boundaries, and redundant areas outside the boundaries are removed to determine the image file.

[0013] In one embodiment, it further includes: The dam identification model was verified by identifying dam points.

[0014] Secondly, this application provides a reservoir dam identification device based on satellite interferometric altimetry and remote sensing imagery, comprising: The data acquisition module is used to acquire water surface elevation data; the water surface elevation data is the water body elevation data of each observation point at the watershed scale based on the SWOT satellite using satellite radar interferometry technology. The processing module is used to preprocess the water surface elevation data to obtain a latitude-longitude sequence, and determine the latitude-longitude coordinate information of potential dam points based on a preset elevation change threshold. The image acquisition module is used to vectorize the CSV coordinate file and perform image cropping, buffer generation, and band filtering based on satellite remote sensing imagery to determine the image file. The image file includes information such as the dam number, satellite type, and band combination. The band combination adopts near-infrared enhancement combination. The CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinates of the potential points of the dam in CSV format. The recognition module is used to input the image file into the dam recognition model and output the predicted location of the reservoir dam to achieve the recognition of the reservoir dam. The dam recognition model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on the image file with labeled information. The labeled information includes the dam location. The improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure. Among them, the ACmix attention module is used to enhance the feature extraction capability of linear edges and gray-level differences of the dam body; the BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy; the reparameterized detection head structure is used to improve the regression accuracy of the dam body and the model convergence stability.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing images.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing images.

[0017] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing images.

[0018] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, apparatus, equipment, medium, and product for reservoir dam identification based on satellite interferometric altimetry and remote sensing imagery. The method preprocesses the acquired water surface elevation data to obtain a latitude-longitude sequence, and determines the latitude-longitude coordinates of potential dam points based on a preset elevation abrupt change threshold. In other words, the water surface elevation data ensures centimeter-level accuracy. Based on this high-precision data, by analyzing abrupt changes in the latitude-longitude sequence along the river, sudden drops in water level are automatically detected, thereby pinpointing the dam location. The CSV coordinate file is then vectorized, and image cropping, buffer generation, and band filtering based on satellite remote sensing imagery are performed to determine the image file. The combination of near-infrared enhanced bands enhances the spectral differences between the dam body, vegetation, and water body, improving separability and training accuracy. The image file is input into the dam identification model, outputting the predicted dam locations to achieve dam identification and improve robustness for identifying small dams and complex backgrounds. Therefore, this application can efficiently and reliably achieve automatic dam identification. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart shows a method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery. Figure 2 Flowchart for dam identification; Figure 3 This is a comparison image of RGB combined and near-infrared enhanced satellite imagery; among them, Figure 3 (a) in the image is a schematic diagram of RGB satellite imagery for different scenes; Figure 3 (b) in the figure is a schematic diagram of near-infrared enhanced combined images of different scenes; Figure 4 This is a schematic diagram of the YOLO11 model structure; Figure 5 This is a structural diagram of a reservoir dam identification device based on satellite interferometric altimetry and remote sensing imagery; Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] This application addresses the core bottlenecks of difficult identification of new and small dams, large candidate areas, and low accuracy in complex scenarios by integrating high-resolution satellite elevation data (i.e., water surface elevation data), precise image capture technology, and an optimized identification model (i.e., dam identification model). It provides an efficient and reliable technical path for large-scale, automated dam identification. This application uses satellite interferometry to accurately acquire watershed water surface elevation data and utilizes the characteristics of sudden water level changes to achieve automatic dam location. This method overcomes the limitations of traditional methods that rely on manual image screening or low-precision hydrological data for location, achieving high efficiency and automation in dam location identification.

[0023] After localization, the Google Earth Engine platform was used to automatically perform image cropping, buffer generation, and band selection based on the localization coordinates. By selectively choosing a near-infrared enhancement combination (B8+B4+B3), the dam's features were enhanced. This band combination significantly improved the distinction between the dam and the surrounding water and vegetation through strong contrast of "bright vegetation and dark dam," thus effectively solving the problem of dam features being obscured in complex terrain.

[0024] In the model training stage, this application adopts an improved YOLO11 deep learning model, which integrates the ACmix attention module, BiFPN feature network and reparameterized detection head structure to specifically solve problems such as missed detection of small targets and weak feature transfer. Furthermore, the hyperparameters are intelligently optimized through Bayesian optimization algorithm to improve the model's performance stability and generalization ability.

[0025] This application constructs an integrated identification system that encompasses water level data acquisition, automatic positioning, image capture, band optimization, model training, and large-scale verification, providing a feasible technical solution for watershed water resources management, flood control scheduling, and infrastructure monitoring.

[0026] An automatic dam location method based on high-precision water body elevation data: Interferometric altimetry using SWOT satellites is employed to acquire watershed-scale water body elevation data. This data undergoes atmospheric delay correction, surface roughness correction, and orbital deviation correction to ensure centimeter-level accuracy. Based on this high-precision data, abrupt changes in the elevation sequence along the river are analyzed to automatically detect "sudden water level drops," thereby pinpointing the dam's location. Simultaneously, the algorithm effectively eliminates natural waterfalls and riverbed interference through threshold judgment and spatial continuity verification, achieving precise dam location without human intervention and providing a highly reliable solution for dam positioning.

[0027] Targeted multi-band image extraction and optimization: The Google Earth Engine (GEE) platform is used to automatically extract and standardize the storage of dam images. Near-infrared enhancement bands are combined to enhance the spectral differences between the dam body, vegetation, and water body, thereby improving the model's separability and training accuracy.

[0028] Dam identification model based on improved YOLO11: In terms of model structure, several innovative modules are incorporated into the YOLO11 framework to improve the robustness of identification of small dams and complex backgrounds.

[0029] Intelligent hyperparameter optimization and large watershed verification mechanism: The Bayesian optimization algorithm is used to achieve intelligent optimization of model training parameters, avoiding the randomness of manual parameter tuning, and finally forming a complete dam identification model.

[0030] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] In one exemplary embodiment, such as Figure 1 As shown, a method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery is provided, including: Step 100: Obtain water surface elevation data. The water surface elevation data is the water body elevation data at each observation point at the watershed scale, obtained based on the SWOT satellite using satellite radar interferometry technology.

[0032] The expression corresponding to the water body elevation data is: .

[0033] in, For water body elevation data; Initial elevation data for radar signal inversion; This is the atmospheric delay error correction amount; This is the correction amount for surface roughness error; This is the correction amount for satellite orbit deviation error.

[0034] Step 200: Preprocess the water surface elevation data to obtain a latitude-longitude sequence, and determine the latitude and longitude coordinates of potential dam points based on a preset elevation change threshold.

[0035] The process includes preprocessing water surface elevation data to obtain a latitude-longitude elevation sequence, and determining the latitude and longitude coordinates of potential dam points based on a preset elevation change threshold. Specifically, this includes: The watershed range is clipped based on the vector boundary of the target study area to eliminate interference in the water surface elevation data. Non-target water body data outside the watershed are removed to obtain the water surface elevation data after interference elimination.

[0036] Based on the water surface elevation data after interference elimination, the discrete elevation observation points are sorted according to the direction of water flow to form a latitude-longitude sequence.

[0037] For latitude-longitude sequence, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold.

[0038] As an optional implementation, for latitude-longitude sequence, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold, specifically including: Determine the preset elevation change threshold based on the dam water level difference. For latitude-longitude-height series, when the elevation difference between adjacent elevation observation points along the water flow direction is greater than the elevation change threshold... At this time, the midpoint of the line connecting two observation points is taken as the suspected abrupt drop point; spatial continuity verification is performed on the suspected abrupt drop point to determine the latitude and longitude coordinates of the potential dam point; among them, when the elevation difference between the three observation points before and after the suspected abrupt drop point does not exceed the elevation change threshold, the dam is considered to have a potential abrupt drop point. In such cases, single-point abrupt changes caused by natural waterfalls or reef obstruction in the river channel are excluded, and these are identified as potential dam points.

[0039] Step 300: Vectorize the CSV coordinate file and perform image cropping, buffer generation, and band filtering based on satellite remote sensing imagery to determine the image file. The image file includes: dam number, satellite type, and band combination information; the band combination uses near-infrared enhancement combination; the CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinates of potential dam points using comma-separated values ​​(CSV) format.

[0040] This process involves vectorizing the CSV coordinate files and performing image cropping, buffer generation, and band filtering based on satellite remote sensing imagery to determine the image files. Specifically, this includes: The CSV coordinate file is vectorized, and a clipping boundary is generated by setting the buffer radius based on the dam size and the resolution of the satellite remote sensing image, with each potential dam coordinate point as the center.

[0041] Based on the Sentinel-2 SR Harmonized multispectral dataset, satellite remote sensing images under cloudless or low-cloud conditions were selected by setting the time range and cloud cover threshold, with the dam center coordinates as spatial constraints. Near-infrared enhancement was then used for visual enhancement processing to obtain the selected near-infrared enhanced images.

[0042] The near-infrared enhanced images after screening are cropped according to the cropping boundaries, and redundant areas outside the boundaries are removed to determine the image file.

[0043] Step 400: Input the image file into the dam recognition model and output the predicted locations of the reservoir dam to achieve dam recognition. The dam recognition model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on image files with labeled information; the labeled information includes dam locations; the improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure; among them, the ACmix attention module is used to enhance the feature extraction capability of linear edges and gray-level differences of the dam body; the BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy; the reparameterized detection head structure is used to improve the regression accuracy of the dam body and the model convergence stability.

[0044] As an optional implementation method, the reservoir dam identification method based on satellite interferometric altimetry and remote sensing imagery also includes: verifying the dam body identification model by identifying dam points.

[0045] This application addresses the need for precise dam identification by constructing a comprehensive solution covering data acquisition, location, modeling, and verification. The specific steps are as follows: Figure 2 As shown. Figure 2 The dam detection model mentioned refers to the dam body identification model.

[0046] Step 1: Obtain and process water surface elevation data.

[0047] This application systematically acquired SWOT (Surface Water and Ocean Topography) satellite remote sensing data through NASA's standardized data acquisition channel. This data system not only covers the core characteristics of large-scale natural water bodies (such as rivers, lakes, and reservoirs) such as water level and spatial distribution, but its centimeter-level precision observation data can accurately reflect the dynamic changes of water bodies, providing fundamental support for the subsequent discovery of potential dam locations.

[0048] The SWOT satellite, relying on advanced radar interferometry technology, can accurately retrieve the water body elevation at each observation point (x, y) at a watershed scale. Its basic principle is based on radar altimetry: the satellite's dual-frequency radar transmits microwave signals to the surface water, and the initial elevation is calculated by the round-trip time of the received signals. Subsequently, a series of targeted correction factors are introduced to eliminate various errors in the radar measurement process, including atmospheric delay errors (interference from the ionosphere and troposphere on microwave signal propagation), surface roughness errors (the influence of water surface ripples on signal reflection), and satellite orbital deviation errors (altimetry deviations caused by slight shifts in the satellite's flight trajectory). Finally, highly reliable water body elevation data is output, calculated as follows: .

[0049] Watershed-scale water surface elevation data acquired by SWOT satellites need to be preprocessed to eliminate interference and regularize the data, laying the foundation for subsequent detection: First, the watershed range is clipped by combining the vector boundary of the study area, and data of non-target water bodies (such as isolated lakes) outside the watershed are removed, retaining only the elevation observation points of the main channel of the river and related waters; then, the discrete points are sorted according to the direction of water flow to form a continuous latitude-longitude-elevation sequence, clarifying the directional benchmark for subsequent sudden drop detection.

[0050] Step 2: Obtain the latitude and longitude of potential dam points using elevation data.

[0051] Based on watershed-scale water surface elevation data acquired by SWOT satellites, the latitude and longitude locations of potential dams can be reversed by capturing "sudden drops in water level". The core logic is that dams, as artificial water-retaining structures, will cause significant water level differences between the upstream and downstream water bodies. In elevation data, this is manifested as a "sudden drop feature" along the direction of water flow. That is, in the same river axis direction, there are abrupt changes in the elevation values ​​of adjacent observation points, and the abrupt change is highly likely to correspond to the location of the dam body.

[0052] For the preprocessed latitude-longitude-height sequence, the latitude-longitude coordinates of potential dams are identified through multi-step detection: First, an elevation change threshold is set. (Referencing typical global dam level differences: 1m-5m for small dams, 5m-15m for medium dams, and >15m for large dams, adjusted according to basin size), when the elevation difference between adjacent observation points along the flow direction exceeds the elevation change threshold... At that time, the center point of the line connecting these two observation points was designated as the "suspected abrupt drop point"; the next step was to verify spatial continuity, requiring that no elevation difference greater than [value missing] occurred between the three observation points before and after the suspected abrupt drop point. In this case, the operation can rule out single-point abrupt changes caused by natural waterfalls or reefs blocking the river.

[0053] Step 3: Capture potential point remote sensing satellite images.

[0054] After identifying the latitude and longitude coordinates of potential dams, the coordinate data is standardized and stored in CSV format to meet subsequent data import and management needs. The CSV file must include core fields and ensure accuracy matching with SWOT satellite elevation data to avoid deviations in subsequent image cropping due to insufficient accuracy. An additional dam type / scale field can be added to provide a basis for subsequent buffer settings. All field names should be concise and free of special symbols for easy reading and parsing by the GEE platform.

[0055] Based on the aforementioned CSV coordinate file, the core process of extracting satellite imagery of the dam on the GEE platform consists of four steps: Step 1: Import and vectorize the coordinate file.

[0056] Upload the generated CSV file in the "Assets" module of the GEE interface, specifying the longitude and latitude columns as geometric fields. GEE will automatically convert the tabular data into a vector layer with geographic coordinates. Simultaneously, the attribute information in the CSV will be retained as feature attributes, ensuring a one-to-one correspondence between each dam location and its unique identifier.

[0057] Step 2: Setting the buffer and square boundaries.

[0058] Using each potential dam coordinate point as the center, a buffer radius is set according to the dam size and satellite image resolution to generate an approximately square clipping boundary. This boundary should cover the main body of the dam and key upstream and downstream areas (such as reservoirs and spillways), ensuring complete inclusion of the dam body while avoiding including too much irrelevant background.

[0059] Step 3: Image screening and band processing (determining the scheme).

[0060] In the image data screening stage, this application uses the Sentinel-2 SR Harmonized multispectral dataset as a spatial constraint, sets the time range and cloud cover threshold, and screens high-quality images under cloudless or low-cloud conditions. It also automatically removes residual clouds and fog through a cloud mask function to ensure the clarity and consistency of the input images.

[0061] In terms of band selection, considering the spectral characteristics and recognition requirements of the dam scene, a near-infrared enhancement combination (B8+B4+B3) was adopted as the main band combination. This combination can create a significant spectral contrast between the water body, dam body, and vegetation. The near-infrared band (B8) has strong reflection from vegetation and weak reflection from water, effectively highlighting the characteristic distribution of "bright vegetation, dark dam body, and even darker water body," fundamentally reducing the interference of vegetation occlusion on dam body recognition. Simultaneously, the introduction of the red band (B4) and green band (B3) further enhances the boundary clarity between the dam body and background features, making the dam body outline more prominent in the image. Compared to other solutions such as the true-color combination (RGB), the near-infrared enhancement combination has stronger discrimination ability and universality in valley-type or high-vegetation-coverage areas, can stably highlight the structural features of the dam body in complex terrain environments, and maintains good visual consistency and automatic recognition performance under different seasons and lighting conditions. A comparison diagram of the near-infrared enhancement combination and the RGB combination is shown below. Figure 3 As shown. Among them, Figure 3 (a) in the image is a schematic diagram of RGB satellite imagery for different scenes; Figure 3 (b) in the figure is a schematic diagram of near-infrared enhanced combined images of different scenes.

[0062] Step 4: Image cropping and export.

[0063] The filtered near-infrared enhanced images are cropped according to square boundaries, removing redundant areas outside the boundaries. When exporting, the filename must include the dam number, satellite type, and band combination information to ensure data traceability and ease of batch management. The exported image resolution is consistent with the original accuracy of the selected satellites, and the projection coordinate system adopts World Geodetic System 1984, WGS84 (EPSG: 4326) to ensure spatial reference consistency. The final exported image files will be automatically saved to the specified Google Drive folder and can be directly used for subsequent dam sample annotation, feature extraction, and recognition model training.

[0064] Step 4: Dam dataset creation and optimization.

[0065] It covers four major stages: image annotation, basic training, model improvement, and hyperparameter optimization, forming a complete chain from data to optimized model, as detailed below: (1) Annotation of multi-band satellite images of the dam.

[0066] Based on multi-band satellite imagery acquired via GEE, open-source annotation tools supporting YOLO format output, such as LabelImg and LabelStudio, were used. All image sizes were standardized to 640×640 pixels to match the default input size of YOLO11, and the training, validation, and test sets were randomly divided in an 8:1:1 ratio. Annotation focused only on the dam itself, using the "minimum bounding rectangle" principle to ensure complete coverage of the dam without excessive background. Annotation files adhered to the YOLO standard format (one .txt file per image, containing the category index, normalized x-coordinate of the center, normalized y-coordinate of the center, normalized width, and normalized height). Quality was ensured through a combination of manual sampling and tool verification after annotation, eliminating low-quality images with cloud cover exceeding 30% or blurred dam structures, ultimately forming a complete dam annotation dataset.

[0067] (2) Basic training of dam identification model based on YOLO11.

[0068] YOLO11, as a multi-iteration version of a single-stage real-time object detection algorithm, transforms the object detection task into a single regression problem. Through a single forward propagation of the neural network, it directly predicts the bounding box coordinates and class probability of the target on the input image. Unlike two-stage algorithms (such as Faster R-CNN), it does not require generating candidate regions and then classifying them first, thus achieving higher inference efficiency.

[0069] The YOLO11 network structure consists of three parts: the backbone, the neck, and the head. It employs an improved backbone and neck architecture, enhancing feature extraction capabilities and improving object detection accuracy and performance on complex tasks. Compared to the YOLOv8 model, it replaces the CF2 module with C3K2, adds a C2PSA module after the SPPF module, and incorporates the head concept from YOLOv10 into the YOLO11 head, using a depth-separable approach to improve efficiency. A schematic diagram of the YOLO11 network structure is shown below. Figure 4 As shown.

[0070] During training, a .yaml configuration file is created to specify the dataset path and the "dam" single-class information. "Image-annotation files" are stored according to a standard directory structure. Model performance is evaluated using the following core metrics during training: Precision measures the proportion of samples that the model predicts to be positive, but which are actually positive. It reflects the model's ability to avoid false positives. .

[0071] In this context, TP (True Positives) represents dams that are correctly identified, and FP (False Positives) represents non-dam targets that are misidentified.

[0072] Recall measures the proportion of samples that are actually positive that are successfully identified by the model, reflecting the model's ability to avoid missing detections. .

[0073] Here, FN (False Negatives) represents dams that were missed during the inspection.

[0074] mAP50 refers to the average accuracy (AP) across all classes when the Intersection Over Union (IoU) threshold is 0.5. .

[0075] Where N is the total number of categories, and IoU is the core indicator for measuring the degree of overlap between the predicted bounding box and the ground truth bounding box in object detection. The value range is [0,1]. The closer the value is to 1, the higher the degree of overlap and the more accurate the localization. For the first i The average precision of a class at IoU=0.5 is calculated by integrating the precision at different recall rates for that class: .

[0076] R represents the recall rate, with an integration range of 0 to 1.

[0077] mAP50-95 is calculated by taking the average of the mAP values ​​for each of the 10 IoU thresholds ranging from 0.5 to 0.95 (intervals of 0.05). .

[0078] in, The mAP value is the value when the IoU threshold is t, which comprehensively reflects the performance of the model under different positioning accuracy requirements.

[0079] (3) Targeted improvements to the YOLO11 model.

[0080] After the basic training of the model, to address the challenges of high-resolution satellite imagery and significant ground interference, the YOLO11 model underwent multi-module improvements to enhance dam recognition performance. First, an ACmix module (ACmix attention module) was introduced to the original C2f structure of YOLO11, fusing convolutional and attention features to enhance the extraction of linear edges and grayscale differences in the dam body. Second, the original PANet structure was replaced with BiFPN (Bidirectional Feature Pyramid Network), enabling efficient fusion of multi-scale features and significantly improving the detection accuracy for small-sized dams. Third, a reparameterized detection head (RepHead) structure was adopted in the detection head, combined with the SIoU loss function and Dynamic TaskAlignment mechanism, improving the regression accuracy and convergence stability of rectangular dams while maintaining the model's lightweight design. This improved combination balances detection accuracy and real-time performance, enabling stable identification of dams of different scales in complex aquatic environments.

[0081] (4) Bayesian hyperparameter intelligent optimization.

[0082] The Bayesian optimization algorithm is used to iteratively optimize the key hyperparameters. The specific steps of the algorithm are as follows: ① Initialization: In the hyperparameter search space, a small number of initial parameter combinations are randomly selected for experiments to obtain initial "parameter-performance" data pairs.

[0083] ② Construct a probabilistic model: Use models such as Gaussian processes or tree-structured Parzen estimators to probabilistically model the mapping relationship between "hyperparameters → model performance" and output the mean and variance of the objective function at any parameter point.

[0084] ③ Selecting the acquisition function: By using the acquisition function (such as expected improvement EI, probability improvement PI, confidence upper limit UCB) to weigh "exploration" and "utilization", the hyperparameters to be tested in the next round are generated.

[0085] ④ Iterative update: Conduct experiments with the selected hyperparameters, add the new "parameter-performance" data to the dataset, update the probability model again, and repeat steps ②-③ until the preset number of iterations is reached or the performance converges.

[0086] The parameters to be optimized include learning rate, batch size, regularization coefficient, and data augmentation strength. The optimization objective is to achieve a validation set mAP@0.5:0.95. Finally, the optimal combination of hyperparameters is selected, and the optimized YOLO11 dam recognition model is obtained by retraining based on this combination.

[0087] Step 5: Identification and verification of dam points on a large scale.

[0088] (1) Construction of a validation set of satellite images for large watersheds.

[0089] To ensure the optimized model can work stably in real-world large river basin scenarios, satellite imagery of typical large river basins (such as the middle reaches of the Yangtze River and the Amazon River basin) was selected for validation: imagery covering an area of ​​no less than 100,000 km² was obtained using Google Earth Engine. 2 High-resolution satellite imagery (consistent with the training set data source) was used, and the optimal band combination (B8+B4+B3) was applied. Areas already used in the training set were excluded to avoid data duplication. Dam targets in the imagery were manually labeled, and a large-scale validation set of dam labels was constructed, covering samples of different sizes, dam types (concrete dams, earth-rock dams, and rockfill dams), and different environments (reservoir areas, river valleys, and plain irrigation areas). The optimized YOLO11 model was deployed to the inference platform to perform batch detection on the validation set images, outputting dam locations (bounding boxes), categories ("dam"), and confidence scores (threshold 0.5). Performance was evaluated from three dimensions: overall detection accuracy, dam type adaptability, and environmental robustness. Attribution analysis was performed on missed and false detection samples, ultimately forming an optimized YOLO11 model that can be used for automatic dam identification at the large-scale watershed level, providing technical support for watershed water resource management, flood control, and disaster reduction.

[0090] (2) Validation using public datasets.

[0091] This auxiliary verification selected four high-reliability public datasets, covering key river basins globally and in China, to supplement verification samples for different dam types and scenarios. Specific information is as follows: Georeferenced Global Dams and Reservoirs (GeoDAR): Contains 24,783 geocoded dam points and 21,515 reservoir polygons, covering 155 countries, with small and medium-sized dams accounting for over 60%, providing attributes such as dam height and reservoir capacity; GDAT (Global Dam tracker.pdf): Contains 35,140 dam records, 31,780 of which are geocoded, covering six continents, recording construction year, function, etc., with prominent coverage of emerging dams in Asia and South America; Global Georeferenced Database of Dams (GOODD): Contains 38,667 manually digitized dam points, focusing on mid-to-low latitude regions, providing upstream catchment area attributes, with high coverage of small and medium-sized dams in Southeast Asia and Africa; China's Large Dams, Reservoirs, and Lakes Dataset. Lakes, China-LDRL: Includes 2,418 large dams in China (624 off-river dams and 1,794 river dams), 2,194 reservoirs, and 3,051 lakes. It features high coordinate accuracy and distinguishes between dam types and reservoir boundaries.

[0092] Validation samples were selected from the four datasets above, covering the large watershed of the satellite image validation set in the previous step. The latitude and longitude coordinates of the dam were output using SWOT analysis, and Sentinel-2 near-infrared enhanced combined band images of all coordinate points were acquired via GEE, with a uniform size of 640×640 pixels and cloud cover controlled to ≤10%, consistent with the training data preprocessing workflow. The dam location was then predicted by the model. This location was compared with the acquired dataset. Using the dam coordinates in the dataset as a reference, the Euclidean distance between the center of the model's detection box and the dataset coordinates was calculated. A deviation ≤50m (adapting to Sentinel-2 resolution) was considered a valid location. This method validates the model and enhances the reliability of the dam location data.

[0093] The dam identification method described above, based on SWOT data, GEE image cropping, and YOLO11 optimization, has core advantages over existing technologies in three dimensions: positioning efficiency, training accuracy, and application adaptability. I. Rapid dam location based on sudden drop in water level significantly reduces resource consumption.

[0094] Existing technologies often rely on manual traversal of large-scale watershed imagery to locate dams, or on delineating candidate areas based on low-precision hydrological data. This results in low positioning efficiency, significant resource waste, and a tendency to overlook smaller dams due to subjective judgment. Furthermore, the candidate areas delineated by low-precision data are often too large, requiring the processing of a large amount of irrelevant data for subsequent image cropping and annotation. In contrast, this application directly captures the characteristics of sudden water level drops upstream and downstream of dams using high-precision water surface elevation data from the SWOT satellite, effectively enabling precise zoning for subsequent work. Firstly, it only requires cropping images around the coordinates, avoiding indiscriminate processing of the entire watershed and reducing resource consumption for GEE image acquisition and storage. Secondly, the positioning process requires no manual intervention, significantly improving initial positioning efficiency and saving time and costs for subsequent annotation and training.

[0095] II. Multi-band combination comparison experiment to ensure optimal accuracy of model training.

[0096] Existing dam identification models often use default RGB band imagery for training without considering the feature differences across multiple satellite imagery bands. This results in limited accuracy in complex scenarios (such as densely vegetated areas and reservoirs). For example, in vegetated mountainous areas, the RGB band can easily confuse the dam with the surrounding mountains. This application uses a "near-infrared enhancement combination (B8+B4+B3)". This band combination fundamentally solves the vegetation occlusion problem through strong contrast between "bright vegetation and dark dam". In addition, this combination can stably identify the dam by "distinguishing vegetation with near-infrared and distinguishing water and dam with red / green" – avoiding the color confusion problem of visible light combinations and the misjudgment of embankments and the interference sensitivity of multispectral fusion in short-wave infrared combinations. At the same time, the three-band combination achieves the best balance between "effective feature coverage" and "data efficiency".

[0097] This targeted approach to band selection avoids the limitations of existing technologies that use bands indiscriminately, allowing the model to fully utilize the multispectral information of satellite imagery, thereby improving training accuracy from the source and laying the foundation for high robustness in subsequent large-scale watershed validation.

[0098] Third, combining model improvement with watershed validation enhances the practical application value.

[0099] Existing technologies often stop at model training on small datasets, failing to optimize models for the specific needs of dam identification (such as missed detection of small dams and interference from complex terrain) and lacking practical validation on a large watershed scale, leading to significant performance degradation when the model is deployed. This application addresses the issues of missed detection of small targets and weak feature transfer by integrating the ACmix attention module, BiFPN feature network, and reparameterized detection head structure into YOLO11; simultaneously, it combines Bayesian optimization to achieve intelligent hyperparameter optimization, avoiding the blindness of manual parameter tuning. More importantly, through large watershed validation, it ensures that the model can work stably in different scenarios such as mountainous areas, plains, and reservoir areas, rather than only performing well on small sample sets.

[0100] This closed loop of "optimization-verification-iteration" enables the model to not only have theoretical accuracy but also adapt to the complex environment of real watersheds, providing practical technical solutions for watershed water resource management, flood control and disaster reduction, and making up for the shortcomings of existing technologies that are "high in laboratory accuracy but poor in practical application".

[0101] In one exemplary embodiment, such as Figure 5 As shown, a reservoir dam identification device based on satellite interferometric altimetry and remote sensing imagery is provided, comprising: The data acquisition module is used to acquire water surface elevation data; the water surface elevation data is the water body elevation data of each observation point at the watershed scale based on the SWOT satellite using satellite radar interferometry technology.

[0102] The processing module is used to preprocess the water surface elevation data to obtain a latitude-longitude sequence, and determine the latitude-longitude coordinate information of potential dam points based on a preset elevation change threshold.

[0103] The image acquisition module is used to vectorize the CSV coordinate file and perform image cropping, buffer generation, and band filtering based on satellite remote sensing imagery to determine the image file. The image file includes information such as the dam number, satellite type, and band combination. The band combination adopts near-infrared enhancement combination. The CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinate information of the potential points of the dam in CSV format.

[0104] The recognition module is used to input the image file into the dam recognition model and output the predicted location of the reservoir dam to achieve the recognition of the reservoir dam. The dam recognition model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on the image file with labeled information. The labeled information includes the dam location. The improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure. Among them, the ACmix attention module is used to enhance the feature extraction capability of linear edges and gray-level differences of the dam body; the BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy; the reparameterized detection head structure is used to improve the regression accuracy of the dam body and the model convergence stability.

[0105] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores reservoir and dam identification data based on satellite interferometry and remote sensing imagery. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a reservoir and dam identification method based on satellite interferometry and remote sensing imagery.

[0106] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0107] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0108] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0109] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0110] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0112] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logic devices, etc., and are not limited to these.

[0113] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery, characterized in that, include: Obtain water surface elevation data; The water surface elevation data is based on the water body elevation data of each observation point at the watershed scale obtained by the SWOT satellite using satellite radar interferometry technology. The water surface elevation data is preprocessed to obtain a latitude-longitude sequence, and the latitude-longitude coordinates of potential dam points are determined based on a preset elevation change threshold. The CSV coordinate file is vectorized, and image cropping, buffer generation, and band filtering are performed based on satellite remote sensing imagery to determine the image file. The image file includes information such as the dam number, satellite type, and band combination. The band combination adopts near-infrared enhancement combination. The CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinates of the potential dam points in CSV format. The image files are input into the dam identification model, which outputs predicted dam locations to achieve dam identification. The dam identification model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on image files with labeled information. The labeled information includes dam locations. The improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure. The ACmix attention module enhances the feature extraction capability of linear edges and grayscale differences of the dam body. The BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy. The reparameterized detection head structure is used to improve the regression accuracy and model convergence stability of the dam body.

2. The method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery according to claim 1, characterized in that, The expression corresponding to the water body elevation data is: ； in, For water body elevation data; Initial elevation data for radar signal inversion; This is the atmospheric delay error correction amount; This is the correction amount for surface roughness error; This is the correction amount for satellite orbit deviation error.

3. The method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery according to claim 1, characterized in that, The water surface elevation data is preprocessed to obtain a latitude-longitude sequence, and the latitude-longitude coordinates of potential dam points are determined based on a preset elevation change threshold. Specifically, this includes: The watershed range is clipped based on the vector boundary of the target study area to perform interference removal processing on the water surface elevation data, and non-target water body data outside the watershed are removed to obtain the water surface elevation data after interference removal. Based on the water surface elevation data after interference elimination, the discrete elevation observation points are sorted according to the direction of water flow to form a latitude-longitude sequence; For latitude-longitude sequence, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold.

4. The method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery according to claim 3, characterized in that, For latitude-longitude-height sequences, the latitude and longitude coordinates of potential dam points are determined based on a preset elevation change threshold, specifically including: Determine the preset elevation change threshold based on the dam water level difference. ; For latitude-longitude-height series, when the elevation difference between adjacent elevation observation points along the direction of water flow is greater than the elevation change threshold... At that time, the midpoint of the line connecting the two observation points is taken as the suspected point of sudden drop; Spatial continuity verification is performed on the suspected sudden drop points to determine the latitude and longitude coordinates of potential dam points; wherein, if the elevation difference between the three observation points before and after the suspected sudden drop point does not exceed the elevation change threshold, then the dam is considered to have a potential dam point. In such cases, single-point abrupt changes caused by natural waterfalls or reef obstruction in the river channel are excluded, and these are identified as potential dam points.

5. The method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery according to claim 1, characterized in that, The CSV coordinate file is vectorized, and image cropping, buffer generation, and band filtering based on satellite remote sensing imagery are performed to determine the image file. Specifically, this includes: The CSV coordinate file is vectorized, and a clipping boundary is generated by setting the buffer radius based on the dam size and the resolution of the satellite remote sensing image, with each potential dam coordinate point as the center. Based on the Sentinel-2 SR Harmonized multispectral dataset, with the dam center coordinates as spatial constraints, a time range and cloud cover threshold were set to filter satellite remote sensing images under cloudless or low-cloud conditions, and near-infrared enhancement was performed for visual enhancement to obtain the filtered near-infrared enhanced images. The near-infrared enhanced images after screening are cropped according to the cropping boundaries, and redundant areas outside the boundaries are removed to determine the image file.

6. The method for identifying reservoir dams based on satellite interferometric altimetry and remote sensing imagery according to claim 1, characterized in that, Also includes: The dam identification model was verified by identifying dam points.

7. A reservoir dam identification device based on satellite interferometric altimetry and remote sensing imagery, characterized in that, include: The data acquisition module is used to acquire water surface elevation data; The water surface elevation data is based on the water body elevation data of each observation point at the watershed scale obtained by the SWOT satellite using satellite radar interferometry technology. The processing module is used to preprocess the water surface elevation data to obtain a latitude-longitude sequence, and determine the latitude-longitude coordinate information of potential dam points based on a preset elevation change threshold. The image acquisition module is used to vectorize the CSV coordinate file and perform image cropping, buffer generation, and band filtering based on satellite remote sensing imagery to determine the image file. The image file includes information such as the dam number, satellite type, and band combination. The band combination adopts near-infrared enhancement combination. The CSV coordinate file is obtained by standardizing and storing the latitude and longitude coordinates of the potential points of the dam in CSV format. The recognition module is used to input the image file into the dam recognition model and output the predicted location of the reservoir dam to achieve the recognition of the reservoir dam. The dam recognition model is determined by using a Bayesian optimization algorithm and performing hyperparameter iteration optimization on an improved YOLO11 model based on the image file with labeled information. The labeled information includes the dam location. The improved YOLO11 model integrates an ACmix attention module, a BiFPN feature network, and a reparameterized detection head structure. Among them, the ACmix attention module is used to enhance the feature extraction capability of linear edges and gray-level differences of the dam body; the BiFPN feature network is used to achieve multi-scale feature fusion and improve detection accuracy; the reparameterized detection head structure is used to improve the regression accuracy of the dam body and the model convergence stability.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the reservoir dam identification method based on satellite interferometric altimetry and remote sensing imagery as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the reservoir dam identification method based on satellite interferometric altimetry and remote sensing imagery as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the reservoir dam identification method based on satellite interferometric altimetry and remote sensing imagery as described in any one of claims 1-6.

Images