A Smart Detection and Repair Method and System for Dam Leakage Based on Multi-Source Information Fusion

The intelligent detection method for dam seepage by using multi-source information fusion and CFR coupling algorithm, combined with prefabricated grass-planted concrete components, solves the problems of single perception dimension and disconnect between monitoring and repair in dam seepage detection. It realizes accurate identification of seepage risks and rapid ecological restoration, and improves the modernization level of dam safety operation and maintenance.

CN123117619APending Publication Date: 2026-11-17NANCHANG INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611253674.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing dam seepage detection technologies suffer from problems such as limited sensing dimensions, isolated diagnostic algorithms, and a disconnect between monitoring and repair, making it difficult to achieve real-time, three-dimensional perception of seepage risks, accurate and intelligent diagnosis, and rapid ecological restoration.

Method used

A smart detection method for dam seepage based on multi-source information fusion is adopted. By combining UAV thermal imaging, ground sensor network and high-density electrical resistivity tomography, an integrated three-dimensional monitoring network of "air-ground-interior" is constructed. The CFR coupling algorithm is used for intelligent diagnosis, and prefabricated grass-planted concrete components are used for ecological restoration.

Benefits of technology

It has enabled precise perception and identification of leakage risks, improved information acquisition capabilities and diagnostic accuracy, achieved seamless integration of monitoring and repair, and enhanced the modernization level of dam safety operation and maintenance and the effectiveness of ecological restoration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117619A_ABST
    Figure CN123117619A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for intelligent detection and repair of dam seepage based on multi-source information fusion. The system includes an integrated air-ground-interior sensing module, a CFR coupling algorithm engine, an AI intelligent diagnosis and early warning platform, and an ecological restoration decision-making unit. The system synchronously collects thermal imaging, multi-physics field, and resistivity profile data through the integrated air-ground-interior sensing module; the CFR coupling algorithm engine fuses the data to identify and locate seepage areas; the AI ​​intelligent diagnosis and early warning platform combines the identification results to perform time-series prediction and hazard classification; and the ecological restoration decision-making unit automatically matches prefabricated grass-planted concrete components and generates customized repair plans based on the diagnosis and classification results. This invention achieves a complete technical closed loop from "intelligent perception and accurate diagnosis" to "ecological restoration," solving the problems of low efficiency, delayed early warning, and limited and ecologically inadequate repair methods in traditional dam inspections.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the interdisciplinary technical field of safety monitoring of water conservancy infrastructure, geological disaster early warning and ecological restoration, and specifically involves a method and system for intelligent detection and repair of dam leakage based on multi-source information fusion. Background Technology

[0002] Leakage, as one of the main hidden dangers threatening the stability of dam structures, is characterized by its high degree of concealment, rapid development, and serious consequences. Traditional dam leakage detection mainly relies on manual inspections, but manual methods are inefficient, have limited coverage, and are highly subjective, making it difficult to detect early-stage hidden leakage, especially under adverse weather conditions or complex terrain. With technological advancements, detection methods such as UAV thermal imaging inspections, ground sensor network monitoring, and high-density electrical resistivity tomography (EDT) have gradually emerged, but existing technologies still have significant limitations: UAV thermal imaging inspections can quickly acquire the dam surface temperature field, but can only capture temperature anomalies that have developed to the surface, and cannot detect internal seepage channels, and are easily affected by environmental interference, resulting in a high false alarm rate; ground sensor networks can monitor physical quantities such as seepage pressure and displacement in real time, but are point measurements, making it difficult to cover the entire line, and have limited ability to locate deep seepage sources; high-density EDT can reflect the resistivity distribution inside the dam, but it is mostly a periodic measurement and cannot achieve all-weather real-time monitoring. The three technical methods mentioned above are isolated from each other and lack collaborative perception and cross-verification of multi-dimensional information from "air-ground-interior", resulting in seriously incomplete information acquisition regarding the three-dimensional and hidden danger of leakage.

[0003] At the data analysis and diagnostic level, existing systems mostly rely on manual experience-based interpretation or the isolated use of single algorithms: while the Canny edge detection algorithm is sensitive to temperature gradients, it is prone to misclassifying normal boundaries as leaks, resulting in a high false alarm rate; while semantic segmentation using fully convolutional networks can extract complete suspected leak areas, it is insufficient in characterizing the edges of subtle leak channels; and while the R-CNN object detection algorithm offers high localization accuracy, it performs poorly in identifying early leaks without significant shape features. The lack of intelligent algorithms that organically couple the advantages of edge sensitivity, complete region coverage, and accurate localization means that the identification of potential leaks remains at the level of mere detection, failing to accurately characterize their spatial location and boundary extent.

[0004] At the restoration and treatment level, most existing systems stop at risk warning and lack effective linkage with subsequent engineering restoration. After receiving the warning, managers still need to conduct manual surveys, design restoration plans, and organize construction. There is a huge gap between monitoring results and restoration implementation, which delays the opportunity for emergency response. At the same time, traditional restoration methods often use measures such as pure concrete slope protection, which meet the engineering strength requirements, but have serious problems such as damaging the embankment ecosystem, high carbon emissions, and high energy consumption. On the other hand, some existing ecological restoration materials often cannot meet the requirements of high strength, high erosion resistance and good plant growth compatibility required by the project, resulting in a single restoration effect and making it difficult to achieve a synergistic improvement in safety and ecology.

[0005] In summary, existing dam seepage detection technologies generally suffer from technical deficiencies such as single perception dimensions, isolated diagnostic algorithms, and a disconnect between monitoring and repair. These deficiencies make it difficult to meet the management needs of modern water conservancy projects for real-time three-dimensional perception of seepage risks, accurate intelligent diagnosis, and rapid ecological restoration. Therefore, there is an urgent need to develop an intelligent dam seepage control system that can achieve multi-dimensional collaborative perception across air, ground, and interior dimensions, coupled intelligent diagnosis using multiple algorithms, and seamless integration of monitoring and repair. This system will enhance the modernization level of dam engineering safety operation and maintenance as well as disaster prevention and mitigation. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent detection and repair of dam seepage based on multi-source information fusion. Through an integrated three-dimensional monitoring network of "air-ground-interior", an intelligent diagnostic engine with CFR coupling algorithm, and ecological restoration decision-making of prefabricated grass-planted concrete components, it achieves accurate perception, intelligent identification and location, and rapid ecological restoration of dam seepage risks.

[0007] On the one hand, this invention proposes an intelligent detection and repair method for dam seepage based on multi-source information fusion, which includes the following steps: Step S1: Acquire thermal imaging data of the surface of the target dam area using a drone equipped with a thermal imager; Step S2: Simultaneously acquire multi-physics data related to seepage through a ground sensor network deployed on the dam. The multi-physics data includes at least seepage pressure, water level, humidity and displacement data. Step S3: Obtain resistivity profile data inside the dam using the internal detection unit; Step S4: Perform spatiotemporal registration and fusion processing on the thermal imaging data obtained in step S1, the multiphysics data obtained in step S2, and the resistivity profile data obtained in step S3 to construct a multidimensional feature dataset of dam seepage state. Step S5: The CFR coupling algorithm is used to process the multidimensional feature dataset to identify and locate the leakage area; the specific steps are as follows: Step S51: performing temperature gradient edge extraction on the thermal imaging data by using a Canny edge detection algorithm to obtain an edge feature map; Step S52: performing semantic segmentation on the edge feature map obtained in step S51 by using a fully convolutional network model, extracting contour features of a suspected seepage area, to obtain a suspected seepage area mask; Step S53: performing target recognition and bounding box regression on the suspected seepage area mask obtained in step S52 by using a region convolutional neural network target detection model, to obtain the accurate position, boundary information and confidence of the seepage area; Step S6: based on historical detection data and real-time detection data, combined with the confidence obtained in step S53, analyzing the development trend of the seepage area identified in step S5 by using a long short-term memory neural network time series prediction model, calculating a risk evolution rate, and performing quantitative grading on the danger of the seepage area according to a preset threshold to obtain a danger level; Step S7: based on the accurate position and boundary information of the seepage area obtained in step S5 and the danger level obtained in step S6, automatically matching the corresponding specification parameters of prefabricated grass-planting concrete members from a pre-constructed ecological restoration material database, and generating a restoration scheme report including a construction positioning map, a member layout diagram, a grass seed proportioning scheme, construction procedure suggestions and ecological benefit assessment.

[0008] Further, in step S1, the UAV inspection path adopts a "Jia"-shaped route planning that combines the direction along the dam axis and the direction perpendicular to the dam axis, and integrates oblique photography and orthophotography modes.

[0009] Further, in step S4, the spatio-temporal registration and fusion processing specifically refers to: taking a high-precision geographic information system base map and a dam three-dimensional model as references, performing unified spatial alignment on the thermal imaging data, the coordinate information of sensors in the ground sensor network and the resistivity profile data, and performing data synchronization according to timestamps, so as to form a spatio-temporal correlated "air-ground-internal" integrated data field.

[0010] Further, in step S6, the danger levels are at least divided into four levels: safe, attention, alert, and alarm.

[0011] In another aspect, the present invention proposes a dam seepage intelligent detection and restoration system based on multi-source information fusion, the system comprising: an air-ground-internal integrated perception module, configured to collect multi-source data related to dam seepage, wherein the air-ground-internal integrated perception module comprises: an air inspection unit, which is a UAV equipped with a thermal imager, configured to collect thermal imaging data of the surface of the dam area; The ground sensing unit is a sensor network deployed on and around the embankment to collect multi-physics field data related to seepage; the sensor network includes at least a piezometer, a water level gauge, a soil moisture sensor, and a displacement monitoring point. An internal detection unit, which is a high-density electrical resistivity measurement device, is used to acquire resistivity profile data inside the dam. The CFR coupling algorithm engine is connected to the air-ground-indoor integrated sensing module and is used to perform fusion and leakage identification of the multi-source data; The CFR coupling algorithm engine includes the following steps in sequence according to the processing flow: Canny edge detection submodule: used to extract edge features from the thermal imaging data by temperature gradient; Fully convolutional network semantic segmentation submodule: used to perform semantic segmentation on edge feature maps, extract contour features of suspected leakage areas, and obtain a mask of suspected leakage areas; Regional convolutional neural network target detection submodule: used to perform target recognition and bounding box regression on the suspected leakage region mask obtained by the fully convolutional network semantic segmentation submodule, to obtain the precise location, boundary information and confidence of the leakage region; The AI-powered intelligent diagnostic and early warning platform is connected to the CFR coupled algorithm engine and integrates a long short-term memory neural network time-series prediction model for trend analysis, hazard quantification and classification, and visual early warning of identified leakage areas. The ecological restoration decision-making unit is connected to the AI ​​intelligent diagnosis and early warning platform. It has a built-in database of prefabricated grass-planted concrete components and a restoration scheme model library, which is used to generate customized ecological restoration schemes based on the diagnosis results.

[0012] Furthermore, the inspection path of the UAV in the aerial inspection unit adopts a "A"-shaped route planning that combines the direction along the dam axis and the direction perpendicular to the dam axis, and integrates oblique photography and orthophoto photography modes to achieve full coverage scanning of the dam surface.

[0013] Furthermore, in the ground sensing unit, the sensors are deployed using a cross-sectional monitoring method, with a monitoring section set up every 200 meters along the embankment. Each monitoring section has a piezometer, a water level gauge, a soil moisture sensor, and a Global Navigation Satellite System (GNSS) displacement monitoring point deployed at the top of the embankment, the water-facing slope, the back slope, and the toe of the slope.

[0014] Furthermore, the internal detection unit employs a high-density electrical resistivity measurement device, which deploys an electrode array on the surface of the dam in a periodic detection mode to measure the resistivity distribution of the underground medium and generate a resistivity profile image. By analyzing the resistivity profile image, abnormally low resistivity areas are identified as indicators of water enrichment or seepage channels.

[0015] Furthermore, the multi-source information database is constructed based on geographic information system technology, and its attribute table is used to assign and classify the latitude and longitude, elevation, stratum lithology, permeability coefficient, groundwater level, and sensor deployment location of the dam.

[0016] Furthermore, the AI ​​intelligent diagnosis and early warning platform integrates a long short-term memory neural network prediction model, which is used to predict the risk probability of the identified leakage areas over time series, and analyze the inundation development trend by combining future meteorological data.

[0017] Furthermore, the AI ​​intelligent diagnostic and early warning platform is deployed in the cloud or on an edge server and communicates with the aerial inspection unit, ground sensing unit and internal detection unit via a wireless network, supporting remote real-time monitoring and command issuance.

[0018] Furthermore, the AI ​​intelligent diagnosis and early warning platform classifies the risk level into at least four levels: safe, alert, vigilant, and alarm. It also establishes an emergency response strategy linked to the risk level: the safe level triggers continuous monitoring, the alert level triggers enhanced observation, the vigilant level triggers repair preparation work, and the alarm level triggers emergency repair. The AI ​​intelligent diagnosis and early warning platform releases early warning maps, risk heat maps, and corresponding repair plans in real time through a visual terminal.

[0019] Furthermore, in the ecological restoration decision-making unit, the prefabricated grass-planting concrete component is made of ecological concrete with special admixtures. The special admixtures are used to enhance the concrete's compressive and erosion resistance and reduce alkalinity to facilitate plant growth. The prefabricated grass-planting concrete component has a 30%-40% interconnected porosity, allowing herbaceous roots to penetrate and form a bio-engineering composite protective layer.

[0020] Furthermore, the system also includes a result output module for generating repair solutions and suggestions.

[0021] Compared with the prior art, the present invention has the following beneficial effects: Significantly enhanced perception dimensions and information completeness: The system constructs an integrated "air-ground-internal" three-dimensional monitoring network, for the first time systematically integrating UAV thermal imaging inspection, ground sensor network, and high-density electrical resistivity tomography (EDT) internal detection system into dam seepage monitoring. The aerial inspection unit rapidly scans the surface temperature field, the ground sensor unit monitors changes in the seepage field in real time, and the internal detection unit images the dam structure to invert the seepage channels. The three work together to achieve full-dimensional perception from the surface to the interior, with cross-validation of multi-source data, significantly improving the ability to acquire information on seepage, a three-dimensional and hidden danger, and overcoming the shortcomings of existing technologies that rely on single methods, resulting in incomplete information and high rates of missed and false detections.

[0022] Significantly improving diagnostic accuracy and intelligence: The proposed CFR coupling algorithm innovatively combines the edge sensitivity of Canny edge detection, the region completeness of fully convolutional network semantic segmentation, and the localization accuracy of regional convolutional neural network target detection, forming a progressive recognition process from edge extraction and region segmentation to target localization. Compared to existing methods that use single algorithms in isolation, this invention's algorithm can simultaneously address the edge characterization of temperature anomalies, the complete delineation of leakage areas, and the precise localization of leakage areas. This represents a technological leap from leakage signal detection to accurate interpretation of leakage targets, significantly improving recognition accuracy and robustness.

[0023] Achieving seamless integration of monitoring and restoration with ecological treatment: The system seamlessly integrates intelligent diagnosis with ecological restoration decision-making units. Based on the diagnostic results, it automatically matches the specifications and parameters of prefabricated grass-planted concrete components to generate customized restoration plans, breaking through the technical barriers of existing systems that stop at risk warning and are disconnected from subsequent restoration. At the same time, the prefabricated grass-planted concrete components have both high strength (≥15MPa) and high porosity (30%-40%), achieving a unity of engineering protection and ecological restoration. Herbaceous roots can penetrate the components and combine with the foundation soil layer to form a three-dimensional slope structure that emphasizes both erosion resistance and ecological greening, achieving a synergistic improvement in safety and ecological benefits.

[0024] Significantly improves early warning timeliness and decision support capabilities: The system dynamically analyzes the development trend of leakage areas based on a long short-term memory neural network time-series prediction model, and combines meteorological data to achieve advanced risk warnings; it establishes a four-level hazard level (safe, alert, warning, alarm) and matches differentiated emergency response strategies, realizing full-chain decision support from "monitoring to diagnosis to early warning to repair". Managers can view the distribution of leakage areas, risk heat maps, and repair plans in real time through a visualization platform, significantly improving emergency response efficiency and the scientific nature of decision-making.

[0025] In summary, this invention not only solves the technical problems of single perception dimension, isolated diagnostic algorithm, and disconnect between monitoring and repair in existing dam leakage detection technologies, but also constructs an intelligent management and control system that integrates the entire process from data collection, intelligent diagnosis, risk warning to ecological restoration, significantly improving the dam's safety assurance capabilities and the green sustainability of its operation and maintenance. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram of the internal processing flow of the CFR coupling algorithm engine of the present invention; Figure 3 A plan view of the embedded cross-shaped prefabricated grass-planting concrete block. Figure 4 This is a schematic diagram illustrating the effect of ecological slope protection. In the figure, 1-convex block, 2-concave block. Detailed Implementation

[0027] To make the technical solution of the present invention easier to understand and implement, the specific implementation steps are further described below in conjunction with the embodiments and accompanying drawings, but the embodiments are not intended to limit the present invention.

[0028] like Figure 1 As shown, this embodiment proposes the following technical solution: an intelligent diagnosis and repair system for dam seepage based on multi-source information fusion and thermal imaging, comprising: The integrated air-ground-interior sensing module is used to collect multi-source data related to dam seepage through three-dimensional monitoring methods, and to construct a multi-source information database containing basic dam information, thermal imaging data, sensor data and stratum detection data; the integrated air-ground-interior sensing module specifically includes an aerial inspection unit, a ground sensing unit and an internal detection unit. The CFR coupling algorithm engine is connected to the air-ground-indoor integrated perception module and is used to perform fusion and leakage identification of the multi-source data. The CFR coupling algorithm engine specifically includes a Canny edge detection submodule, a fully convolutional network (FCN) semantic segmentation submodule, and a region convolutional neural network (R-CNN) object detection submodule. The AI-powered intelligent diagnostic and early warning platform is connected to the CFR coupled algorithm engine and integrates a Long Short-Term Memory (LSTM) time-series prediction model for leakage area trend analysis, hazard quantification and classification, and visual early warning. The ecological restoration decision-making unit is connected to the AI ​​intelligent diagnosis and early warning platform. It has a built-in database of prefabricated grass-planted concrete components and a restoration scheme model library. It is used to automatically match component specification parameters based on the diagnosis results and generate customized restoration scheme reports.

[0029] Example 1: Intelligent Diagnosis and Repair of Leakage in a Section of a Dike Step 1: Collect multi-source data through the integrated air-ground-indoor sensing module.

[0030] Step 11: Before starting the inspection, prepare a dike information database containing information such as latitude, longitude, elevation, length, width, and stratum lithology of the target dike section, and load the digital elevation model (DEM) data base plate of the dike to determine the scope and elevation of the drone inspection area as a unified spatial reference for all subsequent data collection and analysis.

[0031] Step 12: Conduct aerial inspection data collection. Using a drone equipped with a thermal imager, crack inspection data is collected according to a pre-set inspection route strategy. The inspection route strategy involves setting the route along two directions: parallel to the dike axis and perpendicular to the dike's central axis. The drone follows the route in a "T" shape, using both oblique and orthophoto photography. The drone, equipped with a thermal imager, conducts a comprehensive inspection of the dike surface, acquiring thermal imaging data (represented in this embodiment as thermal imaging sequence data) and visible light image data of the dike surface.

[0032] Step 13: Deploy the ground-based sensor network. The preferred deployment method is cross-sectional monitoring, with one monitoring section every 200 meters along the embankment. At each monitoring section, piezometers, water level gauges, soil moisture sensors, and GNSS displacement monitoring points are deployed at the embankment crest, upstream slope, downstream slope, and toe. All sensors are connected to the Internet of Things (IoT) to collect multi-physical field data such as seepage pressure, water level, humidity, and displacement in real time and synchronously.

[0033] Step 14: Conduct internal exploration. Use high-density electrical resistivity tomography (EDT) equipment to explore the resistivity distribution inside the dam from the ground surface and obtain resistivity profile data inside the dam.

[0034] Step 15: Construct a multi-source information database and perform spatiotemporal registration. All data collected in the above steps are stored uniformly on a Geographic Information System (GIS) platform. To achieve effective fusion of "air-ground-interior" multi-source data, this invention employs a high-precision spatiotemporal registration strategy: Spatial Registration: Using the National Geodetic Coordinate System 2000 as the unified spatial benchmark, the existing Digital Elevation Model (DEM) of the dam is integrated with the Building Information Model (BIM) to construct a 3D model of the dam as a digital base. UAV thermal imaging data is directly geolocated and orthorectified using its built-in POS (Positioning and Attitude System) information, and mapped onto the surface of the 3D dam model. Ground sensors (such as piezometers and GNSS displacement monitoring points) are deployed using real-time dynamic differential (RTK) technology for centimeter-level precise positioning. High-density electrical resistivity profiling profiles are inverted using the real-time dynamic differential (RTK) coordinates of the survey line endpoints, and the resulting 3D resistivity distribution data is nested within the 3D dam model. Finally, unified spatial alignment of the three types of data—thermal imaging data (surface), sensor data (discrete points), and resistivity data (internal)—is achieved within a Geographic Information System (GIS) platform.

[0035] Time alignment: All data is timestamped in Coordinated Universal Time (UTC). A dynamic time window alignment mechanism is established for data with different sampling frequencies (sensor minute-level, UAV weekly / monthly-level, electrical resistivity tomography monthly / quarterly-level). For stability leaks, a ±3-day window is used for data correlation; for rapidly evolving risks (risk evolution rate)... (High), automatically narrowing to ±6-hour windows to ensure timely analysis. For time-sparse data, time-series interpolation algorithms can be used to fill in the gaps. Ultimately, an integrated air-ground-internal data field with unified spatiotemporal benchmarks and verifiable internal correlations is formed.

[0036] Step 2: Use the CFR coupling algorithm engine to perform fusion processing and leakage identification on multi-source data.

[0037] Step 21: Extract multidimensional feature datasets from the integrated air-ground-indoor data field and input them into the CFR coupling algorithm engine for processing. The CFR coupling algorithm engine specifically includes a Canny edge detection submodule, a fully convolutional network semantic segmentation submodule, and a region convolutional neural network object detection submodule. The internal processing flow of the CFR coupling algorithm engine is as follows: Figure 2 As shown.

[0038] Step 22, Canny edge detection submodule, uses the Canny edge detection algorithm to calculate the temperature gradient of the thermal imaging data, extracts the temperature anomaly edges, and selects edge regions with significant gradient changes as potential leakage boundaries to obtain the edge feature map.

[0039] Step 23: The fully convolutional network semantic segmentation submodule inputs the edge feature map into the fully convolutional network (FCN) model for semantic segmentation, extracts the complete outline of the suspected leakage area, and obtains the suspected leakage area mask.

[0040] Step 24: The Region Convolutional Neural Network (R-CNN) object detection submodule takes the suspected leakage area mask as a high-quality candidate region and inputs it into the R-CNN object detection model. The R-CNN performs object recognition and bounding box regression on each candidate region, outputting the precise location and boundary information of the leakage area, as well as the confidence score.

[0041] Step 3: Conduct trend analysis and risk classification through the AI ​​intelligent diagnosis and early warning platform.

[0042] Step 31: Construction of the Long Short-Term Memory (LSTM) time-series prediction model. Historical seepage detection data for this section of the dike, along with corresponding meteorological and water level data for the specified periods, are collected to construct a training set. The historical seepage detection data includes UAV thermal imaging sequences acquired through an integrated "air-ground-indoor" sensing module, ground sensor monitoring data (seepage pressure, displacement, humidity), and resistivity profile data detected by high-density electrical resistivity tomography (EPT) equipment. The time span is the past three years, with a sampling frequency of once daily. The corresponding meteorological data includes rainfall, temperature, and evaporation; the water level data includes the water level in front of the dike, the water level behind the dike, and the groundwater level. All data undergoes normalization preprocessing to eliminate the influence of dimensions. An LSTM model is used, taking the historical time-series data as input and outputting the predicted risk probability value for future periods.

[0043] Step 32, Risk Evolution Rate Calculation. The risk evolution rate is calculated by combining the confidence level of the leakage area output by the CFR coupling algorithm with the future risk probability predicted by LSTM. The formula is as follows: ; in, For the rate of risk evolution, This represents the confidence level of the leakage area output by the CFR coupling algorithm at the current moment, with a value between 0 and 1. Future predictions for LSTM models The average risk probability over a given time period (e.g., the next 72 hours); representing the forecast time span (unit: days). This rate reflects the development trend of the leakage hazard: if the risk evolution rate... A value greater than 0 indicates an upward trend in risk; if the risk evolution rate is... A value less than 0 indicates a decreasing risk trend; if the risk evolution rate is less than 0, it indicates a decreasing risk trend. A value close to 0 indicates that the risk is stabilizing. The calculated risk evolution rate is compared with a preset dynamic threshold for subsequent quantitative classification of the hazard level. The preset dynamic threshold can be dynamically adjusted according to the importance of different dike sections and the characteristics of historical hazards.

[0044] Step 33: Quantitative Classification of Risk Levels. Based on preset dynamic thresholds, leakage risks are classified into four levels: safe, caution, alert, and alarm. In this embodiment, once a leakage point reaches the corresponding threshold, the system automatically generates an early warning message.

[0045] Step 4: The ecological restoration decision-making unit generates a customized restoration plan.

[0046] Step 41: Construct a database of prefabricated grass-planted concrete components. These components utilize a specialized admixture that enhances strength, reduces alkalinity, improves durability, and resists erosion. They have a porosity of 30%-40%, a compressive strength ≥15MPa, and allow herbaceous roots to penetrate the concrete and bond with the foundation soil, forming a three-dimensional slope structure. The specialized admixture comprises three main components: an alkalinity-reducing component, a strengthening and durability component, and a vegetation and water-retention component. The alkalinity-reducing component aims to lower the alkalinity within the concrete pores, creating a suitable environment for plant roots. The strengthening and durability component compensates for potential strength loss due to the use of alkalinity-reducing materials and the pursuit of high porosity, and improves erosion resistance. The vegetation and water-retention component provides initial nutrients and water for plant germination and growth within the concrete pores.

[0047] Step 42, Repair Solution Matching. Based on the precise location and hazard level of the leakage area, match prefabricated grass-covered concrete components of the appropriate specifications from the prefabricated grass-covered concrete component database.

[0048] The repair plan uses a rule engine that combines hierarchical decision trees with multi-factor weighted scoring to automatically match the optimal component specifications and grass species from a pre-built database of prefabricated grass-planted concrete components.

[0049] (a) First layer: The main rules driven by the level of danger.

[0050] Depending on the severity of the emergency, different levels of repair strategies will be triggered directly:

[0051] (ii) Second layer: Sub-rules for refining the physical environment parameters of the leakage area.

[0052] After determining the repair strength, specific specifications and models are further matched from the component database based on the physical environmental parameters of the leakage area. Matching factors include:

[0053] Weighted scoring is performed based on matching factors and their corresponding weights, and the component specification with the highest comprehensive score is selected from the prefabricated grass-planted concrete component database.

[0054] For example, a section of the Yangtze River main dike (chainage 15+200) is at "alarm level" ( Taking a seepage of 0.72 as an example, it is located at the toe of the back slope with a water flow velocity of 2.8 m / s. The system triggers a "high-level" repair intensity and retrieves candidate components from the prefabricated grass concrete component database for scoring. The scores are calculated based on the above multi-factor weighted rules. The embedded cross-shaped component (C001) is selected because it meets the safety requirements for erosion velocity (3.5 m / s) and has excellent embedding performance, with the highest matching score (92 points). The through-hole regular hexagonal component (C015) is eliminated because its erosion resistance (2.5 m / s) is insufficient. At the same time, according to Appendix A of Jiangxi Provincial Local Standard "Applicable Grass Species" and the Gramineae Grass Seed Quality Grading Standard GB 6142, the system matches the following grass species combination: Main grass species: Tall fescue (60%) - strong root penetration and erosion resistance; Associated grass species: Bermuda grass (30%) - creeping growth and fast coverage; Sprinkler grass species: White clover (10%) - nitrogen fixation and soil improvement. The embedded cross-shaped type refers to a cross-shaped prefabricated block with a protruding block 1 and a concave block 2, and its planar layout is as follows: Figure 3 As shown, multi-directional splicing and embedding can be achieved.

[0055] Step 43, Remediation Plan Generation. The system automatically generates a customized remediation plan report, including a construction location map, component layout diagram, grass seed ratio scheme, construction procedure suggestions, and ecological benefit assessment. After implementing this remediation plan, the following can be achieved: Figure 4 The ecological slope protection effect shown. Figure 4 The map includes elements such as shrubs, bank top elevation, normal water level, and riverbed to illustrate the overall ecological environment after the restoration project is completed.

[0056] Step 5: Results Output and Emergency Response.

[0057] Step 51, Visual Release. Through the human-computer interaction interface of the AI-powered intelligent diagnostic and early warning platform, the leak point distribution map, risk heat map, long short-term memory network prediction curve, and remediation plan are displayed in real time. Early warning information is sent to management personnel via the platform.

[0058] Step 52, Emergency Response Execution. Based on the four-level risk mechanism, trigger the corresponding response strategy: Safe level corresponds to continuous monitoring, Attention level to enhanced observation, Alert level to preparation for repair, and Warning level to emergency repair. Management personnel organize emergency response and construction according to the warning level and plan.

[0059] Step 53: System closed-loop feedback. After the repair is completed, the ground sensor network continues to monitor changes in seepage pressure and displacement in the area, while the drone patrols and tracks vegetation recovery. New data is automatically imported into a multi-source information database for model retraining and parameter adjustment, enabling system self-optimization.

[0060] Through the above steps, this invention realizes intelligent management of the entire process from "air-ground-indoor perception, CFR algorithm coupled diagnosis, LSTM trend early warning to ecological restoration scheme generation", which significantly improves the efficiency of handling dam seepage risks and the level of ecological protection.

[0061] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent detection and repair of dam seepage based on multi-source information fusion, characterized in that, Includes the following steps: Step S1: Acquire thermal imaging data of the surface of the target dam area using a drone equipped with a thermal imager; Step S2: Simultaneously acquire multi-physics data related to seepage through a ground sensor network deployed on the dam. The multi-physics data includes at least seepage pressure, water level, humidity and displacement data. Step S3: Obtain resistivity profile data inside the dam using the internal detection unit; Step S4: Perform spatiotemporal registration and fusion processing on the thermal imaging data obtained in step S1, the multiphysics data obtained in step S2, and the resistivity profile data obtained in step S3 to construct a multidimensional feature dataset of dam seepage state. Step S5: The CFR coupling algorithm is used to process the multidimensional feature dataset to identify and locate the leakage area; the specific steps are as follows: Step S51: Use the Canny edge detection algorithm to extract the temperature gradient edges from the thermal imaging data to obtain an edge feature map; Step S52: Use a fully convolutional network model to perform semantic segmentation on the edge feature map obtained in step S51, extract the contour features of the suspected leakage area, and obtain the mask of the suspected leakage area. Step S53: Use a region convolutional neural network target detection model to perform target recognition and bounding box regression on the suspected leakage area mask obtained in step S52, so as to obtain the precise location, boundary information and confidence level of the leakage area; Step S6: Based on historical and real-time detection data, and combined with the confidence level obtained in step S53, the development trend of the leakage area identified in step S5 is analyzed using a long short-term memory neural network time-series prediction model. The risk evolution rate is calculated, and the risk level of the leakage area is quantified and classified according to a preset threshold to obtain the risk level. Step S7: Based on the precise location and boundary information of the leakage area obtained in Step S5 and the hazard level obtained in Step S6, the corresponding prefabricated grass-planted concrete component specifications are automatically matched from the pre-built ecological restoration material database, and a restoration plan report is generated that includes a construction positioning map, component layout map, grass seed ratio scheme, construction procedure suggestions and ecological benefit assessment.

2. The intelligent detection and repair method for dam seepage based on multi-source information fusion according to claim 1, characterized in that, In step S1, the UAV inspection path adopts a "A"-shaped route planning that combines the direction along the dam axis and the direction perpendicular to the dam axis, and integrates oblique photography and orthophoto photography modes.

3. The intelligent detection and repair method for dam seepage based on multi-source information fusion according to claim 1, characterized in that, In step S4, the spatiotemporal registration and fusion process specifically involves: using the high-precision geographic information system base map and the dam 3D model as a reference, uniformly aligning the thermal imaging data, the coordinate information of the sensors in the ground sensor network, and the resistivity profile data, and synchronizing the data according to the timestamp to form a spatiotemporally correlated "air-ground-interior" integrated data field.

4. The intelligent detection and repair method for dam seepage based on multi-source information fusion according to claim 1, characterized in that, In step S6, the danger level is divided into at least four levels: safe, alert, vigilant, and alarm.

5. A smart dam leakage detection and repair system based on multi-source information fusion, used to execute the smart dam leakage detection and repair method based on multi-source information fusion as described in any one of claims 1-4, characterized in that, The system includes: An integrated air-ground-indoor sensing module is used to collect multi-source data related to dam seepage. The integrated air-ground-indoor sensing module includes: An aerial inspection unit, which is a drone equipped with a thermal imager, is used to collect thermal imaging data of the surface of the dam area. The ground sensing unit is a sensor network deployed on and around the embankment to collect multi-physics field data related to seepage; the sensor network includes at least a piezometer, a water level gauge, a soil moisture sensor, and a displacement monitoring point. An internal detection unit, which is a high-density electrical resistivity measurement device, is used to acquire resistivity profile data inside the dam. The CFR coupling algorithm engine is connected to the air-ground-indoor integrated sensing module and is used to perform fusion and leakage identification of the multi-source data; The CFR coupling algorithm engine includes the following steps in sequence according to the processing flow: Canny edge detection submodule: used to extract edge features from the thermal imaging data by temperature gradient; Fully convolutional network semantic segmentation submodule: used to perform semantic segmentation on edge feature maps, extract contour features of suspected leakage areas, and obtain a mask of suspected leakage areas; Regional convolutional neural network target detection submodule: used to perform target recognition and bounding box regression on the suspected leakage area mask obtained by the fully convolutional network semantic segmentation submodule, to obtain the precise location and boundary information of the leakage area and the confidence level; The AI-powered intelligent diagnostic and early warning platform is connected to the CFR coupled algorithm engine and integrates a long short-term memory neural network time-series prediction model for trend analysis, hazard quantification and classification, and visual early warning of identified leakage areas. The ecological restoration decision-making unit is connected to the AI ​​intelligent diagnosis and early warning platform. It has a built-in database of prefabricated grass-planted concrete components and a restoration scheme model library, which is used to generate customized ecological restoration schemes based on the diagnosis results.

6. The intelligent detection and repair system for dam seepage based on multi-source information fusion according to claim 5, characterized in that, In the ground sensing unit, the sensors are deployed using a cross-sectional monitoring method.

7. The intelligent detection and repair system for dam seepage based on multi-source information fusion according to claim 5, characterized in that, The AI-powered intelligent diagnostic and early warning platform classifies the risk level into at least four levels: safe, alert, vigilant, and alarm. It also establishes an emergency response strategy linked to the risk level: the safe level triggers continuous monitoring, the alert level triggers enhanced observation, the vigilant level triggers repair preparation work, and the alarm level triggers emergency repair. The AI-powered intelligent diagnostic and early warning platform releases early warning maps, risk heat maps, and corresponding repair plans in real time through a visual terminal.

8. The intelligent detection and repair system for dam seepage based on multi-source information fusion according to claim 5, characterized in that, The AI ​​intelligent diagnostic and early warning platform is deployed in the cloud or on an edge server and communicates with the aerial inspection unit, ground sensing unit and internal detection unit via a wireless network, supporting remote real-time monitoring and command issuance.

9. The intelligent detection and repair system for dam seepage based on multi-source information fusion according to claim 5, characterized in that, The prefabricated grass-planting concrete component is made of ecological concrete with special admixtures. The special admixtures are used to enhance the concrete's compressive and erosion resistance and reduce alkalinity to facilitate plant growth. The prefabricated grass-planting concrete component has a 30%-40% interconnected porosity, allowing herbaceous roots to penetrate and form a bio-engineering composite protective layer.