A method and system for monitoring dynamic embankment danger hidden trouble

Abstract

The application discloses a kind of embankment danger hidden danger dynamic monitoring method and system, belong to hidden danger monitoring technical field, its method specifically includes: obtaining embankment apparent image, and establish uniform coordinate datum in combination with embankment control point, the subpixel level space-time registration is carried out to the embankment apparent image, based on embankment section standard form library automatically generates embankment structure template, and is obtained by water surface reflection suppression and vegetation shelter perspective reconstruction, obtains embankment structure enhancement image, utilizes the preset double branch image analysis model, jointly extracts the apparent feature of embankment structure enhancement image danger hidden danger, and abnormal area is carried out pixel level change detection and connected domain growth, forms candidate danger target set, carries out grading determination to danger target, outputs embankment danger hidden danger dynamic monitoring result;The application can reduce the influence of environmental interference and shooting condition change on monitoring result, realize continuous identification and evolution tracking to embankment concealment danger.

Description

Technical Field

[0001] This invention belongs to the field of hidden danger monitoring technology, specifically a method and system for dynamic monitoring of potential dangers to dikes. Background Technology

[0002] With the continuous improvement of comprehensive watershed management and flood control engineering systems, dikes, as key infrastructure in the flood control and disaster reduction system, are directly related to the safety of people's lives and property and the stability of regional economy and society through their long-term operation. Affected by factors such as water level fluctuations, seepage, uneven foundation settlement, and material aging, dikes are prone to gradually developing potential hazards such as leakage, cracks, local heave, or collapse during their service life. These hazards are often characterized by strong concealment, gradual evolution, and discontinuous spatial distribution. If they are not detected and dealt with in a timely manner, they can easily evolve into dangerous situations during the flood season or under extreme conditions.

[0003] Current methods for monitoring dikes mainly rely on manual patrols, fixed-point sensor monitoring, or single-image comparison. Manual patrols are heavily influenced by subjective experience, making it difficult to achieve continuous, quantitative, and traceable monitoring. Fixed-point sensors are costly to deploy, have limited coverage, and cannot reflect overall changes in the dike's surface structure. Monitoring methods based on single images or simple differences often ignore the impact of differences in the dike's structural morphology and changes in shooting conditions on the images, easily misjudging environmental disturbances, changes in lighting, or water surface reflections as potential hazards, resulting in insufficient stability and reliability of monitoring results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a dynamic monitoring method and system for dike hazards. This method integrates prior information about the dike structure with multi-temporal image data for dynamic monitoring. Under a unified spatial reference, the apparent structure of the dike is constrained, and multiple types of hazards are jointly identified and graded from a temporal evolution perspective, thus achieving continuous, objective, and dynamic monitoring of dike hazards.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for dynamic monitoring of potential dangers to dikes includes:

[0007] Acquire the surface image of the dike, and establish a unified coordinate reference based on the dike control points, and perform sub-pixel-level spatiotemporal registration on the surface image of the dike;

[0008] Based on the standard morphology library of dike cross sections, dike structure templates are automatically generated. The registered images are then subjected to structural constraint normalization. Finally, through water surface reflection suppression and vegetation occlusion perspective reconstruction, an enhanced image of the dike structure is obtained.

[0009] Using a pre-defined bi-branch image analysis model, the apparent features of potential hazards in the enhanced image of the embankment structure are jointly extracted, and pixel-level change detection and connected component growth are performed on abnormal areas to form a set of candidate potential hazards. The apparent features of potential hazards include changes in seepage wet spots, micro-deformation of local bulges / collapses, and crack evolution.

[0010] A dynamic risk index is constructed based on the time-series image features of candidate potential hazards, and the potential hazards are classified and judged, outputting the dynamic monitoring results of dike hazards.

[0011] Specifically, the step of acquiring the apparent image of the dike, establishing a unified coordinate reference based on the dike control points, and performing sub-pixel-level spatiotemporal registration on the apparent image of the dike includes:

[0012] The surface images of the dikes acquired by different imaging terminals at different times are obtained, and the surface images of the dikes are classified according to the imaging terminal identifier and the time identifier to form a sequence of surface images of the dikes.

[0013] In the sequence of apparent images of the embankment, a stable structural line is identified, and the embankment control points corresponding to the stable structural line are extracted. The embankment control points are then filtered to obtain a set of effective embankment control points. The stable structural line includes the embankment top, the embankment toe, and the slope boundary.

[0014] Based on the set of effective dike control points, a unified coordinate reference associated with the dike axis is established, and the appearance images of each dike are mapped to the unified coordinate reference.

[0015] Under the unified coordinate reference, the appearance images of each dike are divided into local regions, the feature position offset relationship within the corresponding region is determined, and sub-pixel level displacement information for image registration is obtained;

[0016] Using the levee appearance image at the reference time as a reference, coordinate mapping and geometric adjustment are performed on the levee appearance images at other times in chronological order, so that the levee appearance images are spatiotemporally registered under a unified coordinate reference.

[0017] Specifically, the process of using the levee appearance image at a reference time as a reference, and performing coordinate mapping and geometric adjustment on the levee appearance images at other times in chronological order, so that all levee appearance images are spatiotemporally registered under a unified coordinate reference, includes:

[0018] The images of the dike are sorted according to the time stamp, and the images of the dike within the preset reference time window are selected as the reference time images. Their spatial position under the unified coordinate reference is determined as the reference mapping state.

[0019] For each levee appearance image at a non-reference time, an initial coordinate mapping relationship corresponding to the reference time image is generated based on the positional relationship of the corresponding levee control points under a unified coordinate reference.

[0020] Based on the initial coordinate mapping relationship, the apparent image of the dike at the non-reference time is divided into multiple continuous sub-regions, and the local geometric adjustment parameters of each sub-region relative to the reference time image are determined respectively.

[0021] Based on the local geometric adjustment parameters, the apparent image of the dike at the non-reference time is subjected to partitioned coordinate correction, and the corrected apparent image of the dike is uniformly projected onto the unified coordinate reference to complete the spatiotemporal registration of the apparent image of the dike.

[0022] Specifically, the process involves automatically generating a dike structure template based on a standard dike cross-section morphology library, performing structural constraint normalization on the registered image, and obtaining an enhanced dike structure image through water surface reflection suppression and vegetation occlusion perspective reconstruction, including:

[0023] Call the pre-built standard morphology library of dike cross sections, automatically match the corresponding cross section morphology under a unified coordinate reference, and generate a dike structure template corresponding to the currently monitored dike section accordingly.

[0024] The dike structure template is projected onto the dike appearance image that has been spatiotemporally registered. Based on the spatial distribution range of the dike top, dike slope and dike toe defined by the template, the dike appearance image is divided into structural constraint regions.

[0025] Within the structural constraint area, the image grayscale distribution and spatial continuity of each region are uniformly scaled to ensure that the apparent images of the dike under different times and imaging conditions meet consistent normalization rules within the corresponding structural regions.

[0026] For image content located in the area adjacent to the water surface, based on the spatial constraints of the water-shore boundary position in the structural template, the reflection interference area is identified, and the image information of the reflection interference area is reconstructed and replaced.

[0027] For structurally constrained areas covered or obscured by vegetation, perspective reconstruction is performed on the obscured areas based on the cross-sectional morphology of the corresponding areas in the embankment structure template to generate an enhanced image of the embankment structure.

[0028] Specifically, the dike structure template is projected onto the spatiotemporally registered dike appearance image. Based on the spatial distribution range of the dike crest, slope, and toe defined by the template, the dike appearance image is divided into structurally constrained regions, including:

[0029] Under a unified coordinate reference, the scale and orientation matching processing of the embankment structure template is performed so that its cross-sectional position corresponds to the embankment axis in the spatiotemporally registered embankment appearance image.

[0030] Based on the positional relationship of the predefined dike top line, dike slope line and dike toe line in the dike structure template, the corresponding structural boundary trajectory is generated in the dike appearance image;

[0031] Using the structural boundary trajectory as a constraint, the apparent image of the embankment is spatially segmented to form sets of structural constraint regions corresponding to the top region, slope region, and toe region of the embankment, respectively.

[0032] The set of structural constraint regions is identified and sequentially numbered, and each structural constraint region is associated with the corresponding cross-sectional region in the embankment structural template.

[0033] Specifically, the method utilizes a pre-defined bi-branch image analysis model to jointly extract the apparent features of potential hazards from the enhanced image of the levee structure, and performs pixel-level change detection and connected component growth on abnormal areas to form a candidate hazard target set, including:

[0034] The enhanced image of the embankment structure is simultaneously input into the first feature analysis branch and the second feature analysis branch according to the preset analysis rules. The first feature analysis branch is used to process the image region that reflects changes in surface state, and the second feature analysis branch is used to process the image region that reflects changes in structural morphology.

[0035] In the first feature analysis branch, based on the gray-scale distribution changes and regional continuity of the enhanced image of the embankment structure, the apparent features related to the changes in seepage wet spots are extracted, and the corresponding first set of abnormal features is formed.

[0036] In the second feature analysis branch, based on the spatial positional relationship of the structural constraint area in the enhanced image of the embankment structure, morphological change features related to local uplift, collapse and crack evolution are extracted, and a corresponding second set of abnormal features is formed.

[0037] The first set of abnormal features and the second set of abnormal features are fused under a unified coordinate reference. Abnormal features that have spatial overlap or temporal correlation are merged, and pixel-level change detection and connected component growth are performed on the fused abnormal region.

[0038] Based on the range of the abnormal region after the connected domain is grown and its distribution position in the structural constraint region, corresponding candidate hazard targets are generated, and the candidate hazard targets are summarized to form a candidate hazard target set.

[0039] Specifically, the step of generating corresponding candidate hazard targets based on the range of the abnormal region after connected component growth and its distribution position in the structural constraint region, and summarizing the candidate hazard targets to form a candidate hazard target set, includes:

[0040] The abnormal regions obtained by connected domain growth are traversed one by one to determine the spatial range of each abnormal region under a unified coordinate reference, and the spatial inclusion relationship between each abnormal region and the corresponding structural constraint region is recorded.

[0041] Based on the spatial inclusion relationship, each abnormal region is mapped to at least one structural category in the top region, slope region, or toe region of the dike, and a corresponding structural location identifier is assigned to each abnormal region.

[0042] By combining the spatial extent, structural location identifiers, and order of appearance of the abnormal regions in the time series, each abnormal region is converted into a candidate hazard target with an independent identifier;

[0043] The generated candidate hazard targets are uniformly numbered and archived, and summarized according to their structural location identifiers and time sequence to form a candidate hazard target set.

[0044] Specifically, the process of constructing a dynamic risk index based on the time-series image features of candidate potential hazards, classifying and determining the potential hazards, and outputting dynamic monitoring results of dike hazards includes:

[0045] For each candidate hazard target in the candidate hazard target set, extract the image segments corresponding to it at multiple times according to the time identifier, and establish a time series segment set of the candidate hazard target under a unified coordinate reference;

[0046] In the set of time series segments, based on the structural location identifiers of candidate hazardous targets, time series features for characterizing changes in seepage wet spots, changes in local uplift / collapse, and crack evolution are extracted respectively, and the time series features are grouped into target feature sequences according to type;

[0047] The target feature sequence is processed for temporal consistency to determine the persistence, discontinuity and migration relationships of each type of feature between adjacent time points, and an evolutionary trajectory description corresponding to the candidate hazard target is generated.

[0048] Based on the evolutionary trajectory description, each type of feature is jointly assigned a value according to a preset set of risk factors, and the joint assignment results are aggregated to form a dynamic risk index corresponding to the candidate risk target.

[0049] Based on the dynamic risk index and preset grading rules, candidate risk targets are graded and judged, and the grading results are associated with the spatial range, structural location and time of the candidate risk targets to form dynamic monitoring results of dike risks and hidden dangers.

[0050] Specifically, based on the evolutionary trajectory description, each type of feature is jointly assigned a value according to a preset set of risk factors, and the joint assignment results are aggregated to form a dynamic risk index corresponding to the candidate risk target, including:

[0051] For each candidate hazard target, read its corresponding evolution trajectory description, and generate a sequence of feature events for joint assignment according to the structural position identifier, feature type identifier and time order recorded in the evolution trajectory description;

[0052] A preset set of risk factors is invoked, and the set of risk factors is divided into location-based risk factors, evolution-based risk factors, and coupling-based risk factors according to applicable conditions. Location-based risk factors are used to limit the assignment entry points for different structural positions, evolution-based risk factors are used to limit the assignment rules corresponding to the continuity, discontinuity, and migration of the feature event sequence, and coupling-based risk factors are used to limit the linkage assignment rules between different feature types.

[0053] Based on the location-based risk factors, the candidate hazard targets are assigned values ​​through a channel selection process. Within the assigned value channel, the characteristic event sequences corresponding to changes in seepage wet spots, changes in local uplift / collapse, and crack evolution are assigned values ​​in segments based on the evolution-based risk factors. At the same time, different characteristic event sequences that have temporal overlap or spatial adjacency are assigned values ​​through cross-joint processing based on the coupling-based risk factors.

[0054] The joint assignment results are aggregated, and the aggregated assignment entries are mapped in chronological order to the dynamic risk index corresponding to the candidate risk target.

[0055] A dynamic monitoring system for potential dangers to dikes, used to implement the aforementioned dynamic monitoring method for potential dangers to dikes, includes: an image processing module, an enhancement module, a danger target generation module, and a monitoring result output module;

[0056] The image processing module is used to acquire the surface image of the dike, and establish a unified coordinate reference in combination with the dike control points, and perform sub-pixel-level spatiotemporal registration on the surface image of the dike.

[0057] The enhancement module is used to automatically generate a dike structure template based on a standard dike cross-section morphology library, perform structural constraint normalization processing on the registered image, and obtain an enhanced image of the dike structure through water surface reflection suppression and vegetation occlusion perspective reconstruction.

[0058] The hazard target generation module is used to jointly extract the hazard appearance features of the enhanced image of the levee structure using a preset bi-branch image analysis model, and to perform pixel-level change detection and connected component growth on abnormal areas to form a candidate hazard target set.

[0059] The monitoring result output module is used to construct a dynamic risk index based on the time series image features of candidate potential danger targets, classify and determine the potential danger targets, and output the dynamic monitoring results of potential dangers to the dikes.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] This invention proposes a dynamic monitoring method and system for dike hazards. By performing spatiotemporal registration of multi-temporal dike appearance images under a unified coordinate reference, introducing a standard dike cross-section morphology library to generate structural templates and applying structural constraints to the images, and combining bi-branch image analysis and time series evolution modeling, the method achieves joint identification and dynamic classification of multiple types of hazards such as seepage, deformation, and cracks, thereby avoiding misjudgments caused by single image or single feature analysis. This method can reduce the impact of environmental interference and changes in shooting conditions on monitoring results without relying on high-density sensor deployment, and achieve continuous identification and evolution tracking of hidden dike hazards, effectively improving the overall consistency, traceability, and engineering applicability of dike monitoring. Attached Figure Description

[0062] Figure 1 A flowchart of a dynamic monitoring method for potential dangers to dikes provided by this invention;

[0063] Figure 2 A schematic diagram of candidate hazardous targets provided by the present invention;

[0064] Figure 3 This invention provides an architecture diagram of a dynamic monitoring system for potential dangers to dikes. Detailed Implementation

[0065] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0067] It should be noted that, unless there is conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order than the module division in the device or the order in the flowchart. In addition, the "

[0068] The terms "first," "second," and "third" do not limit the data or execution order; they are merely used to distinguish identical or similar items with essentially the same function and purpose.

[0069] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0070] Example 1

[0071] Please see Figures 1-2 The present invention provides an embodiment of a method for dynamic monitoring of potential dangers to dikes, comprising the following specific steps:

[0072] Step S1: Obtain the apparent image of the dike, and establish a unified coordinate reference based on the dike control points, and perform sub-pixel-level spatiotemporal registration on the apparent image of the dike.

[0073] The specific steps of step S1 are as follows:

[0074] Step S101: Acquire surface images of the dikes collected by different imaging terminals at different times, and classify the surface images of the dikes according to the imaging terminal identifier and time identifier to form a sequence of surface images of the dikes.

[0075] In this embodiment, the surface images of the dike are acquired at different times by fixed imaging terminals deployed along the dike and mobile imaging terminals that are periodically inspected. Since the shooting positions, viewing angles, and imaging parameters of different imaging terminals differ, and the surface state of the dike exhibits significant temporal evolution characteristics, directly mixing these images could easily lead to confusion in temporal relationships or incomparability of features across terminals. Based on these considerations, after acquiring the surface images of the dike, the images are first distinguished by their unique identifiers to ensure consistency in spatial perspective for images acquired by the same imaging terminal. Then, the images from each source are sorted chronologically based on their time identifiers, thereby establishing a clear organizational relationship in both the terminal and time dimensions, forming a sequence of surface images of the dike with clear temporal attributes. This method enables subsequent steps to analyze changes in the surface of the dike from the same terminal perspective and accurately track the evolution of the same dike section on the timeline.

[0076] Step S102: In the apparent image sequence of the dike, identify stable structural lines, extract the dike control points corresponding to the stable structural lines, and filter the dike control points to obtain a set of effective dike control points. The stable structural lines include the dike top, dike toe, and slope protection boundary.

[0077] In this embodiment, the dike top, dike toe, and slope protection boundaries contained in the dike appearance image sequence have stronger structural stability and positional continuity compared to the water surface, vegetation, or temporary attachments, thus serving as a reliable basis for establishing spatial references. Specifically, in the dike appearance images at each time point, linear regions with continuous brightness changes and directions consistent with the dike axis are first detected, identifying stable structural lines corresponding to the dike top edge, slope toe turning point, and slope protection boundaries. Subsequently, several candidate dike control points are extracted along these stable structural lines at preset intervals. It should be noted that due to local occlusion, changes in imaging angle, or short-term construction activities causing some structural lines to shift or be interrupted, the candidate dike control points are further screened, eliminating control points with abnormal positional changes in adjacent time points or those that do not satisfy structural continuity relationships. Only control points that maintain a stable distribution across multiple time nodes are retained, ultimately forming an effective set of dike control points for subsequent modeling and registration.

[0078] Step S103: Based on the set of effective dike control points, establish a unified coordinate reference associated with the dike axis, and map the appearance image of each dike to the unified coordinate reference.

[0079] In this embodiment, since the dike extends in a strip along the axis, its apparent structural changes are mainly manifested as continuous changes along the axis and local offsets in the direction perpendicular to the axis. Therefore, establishing a unified coordinate reference using the dike axis as a spatial reference is beneficial to eliminating spatial differences caused by different imaging terminals and shooting positions. In specific implementation, firstly, based on the relative positional relationship of each control point in the effective dike control point set in the image, a dike axis consistent with the dike's direction is fitted, and this axis is used as the main direction reference. Subsequently, with the axis as a reference, the distribution of each control point in the dike top direction and dike slope direction is expanded to construct a coordinate expression method that matches the dike's structural morphology. It should be noted that after the unified coordinate reference is established, by matching the correspondence between the control point positions in each dike appearance image and the unified coordinate reference, dike appearance images acquired at different times and from different terminals are mapped to the same spatial reference, thereby making each image comparable and continuous in subsequent processing.

[0080] Step S104: Under the unified coordinate reference, the appearance images of each dike are divided into local regions, the feature position offset relationship within the corresponding region is determined, and sub-pixel level displacement information for image registration is obtained.

[0081] In this embodiment, even though the surface images of the dike acquired at different times have been mapped to a unified coordinate reference, there are still local misalignments due to slight differences in shooting posture or changes in imaging conditions. Relying solely on overall mapping is insufficient to meet the requirements for fine registration. Based on this understanding, each surface image of the dike is divided into several interrelated local regions along the dike axis and perpendicular to the axis under the unified coordinate reference. Feature positions with stable texture distribution are selected within each local region as comparison references. Subsequently, the feature positions corresponding to the same local region at different times are aligned one by one. By comparing their relative offsets under the unified coordinate reference, the displacement relationship within the local region is determined. It should be noted that, to avoid errors caused by integer pixel alignment, when determining the displacement relationship, a refined interpolation analysis is performed on the spatial distribution of the feature positions to obtain sub-pixel-level displacement information reflecting subtle local offsets, enabling subsequent image registration to be adjusted at a finer spatial scale.

[0082] Step S105: Using the levee appearance image at the reference time as a reference, perform coordinate mapping and geometric adjustment on the levee appearance images at other times in chronological order, so that the levee appearance images are spatiotemporally registered under a unified coordinate reference.

[0083] The specific steps of step S105 are as follows:

[0084] Step S1051: Sort the dike appearance images according to the time markers, and select the dike appearance images within the preset reference time window as the reference time images, and determine their spatial positions under the unified coordinate reference as the reference mapping state.

[0085] In this embodiment, the apparent images of the dike are acquired at different times. Their apparent state reflects both the actual structural changes and inevitably includes short-term environmental disturbances or differences in shooting conditions. Therefore, before subsequent mapping and registration, a stable and representative reference time needs to be determined. In practice, the apparent images of the dike are first sorted according to time markers, and images within a preset reference time window are selected from the sorting results. This time window is used to avoid images acquired under obviously abnormal working conditions or extreme environmental conditions. It should be noted that within this time window, the overall structural state of the dike is relatively continuous, and the differences between images mainly come from shooting conditions rather than structural evolution. The selected apparent image of the dike is determined as the reference time image, and its spatial position under a unified coordinate reference is used as the reference mapping state for subsequent image mapping and geometric adjustment, thereby providing a unified alignment reference for images at other times. It should be noted that the preset reference time window is such as 1 hour, 24 hours, or determined according to the dike monitoring frequency.

[0086] Step S1052: For each levee appearance image at a non-reference time, based on the positional relationship of its corresponding levee control points under a unified coordinate reference, generate an initial coordinate mapping relationship corresponding to the reference time image.

[0087] In this embodiment, although the levee appearance image acquired at a non-reference time is under the same coordinate reference as the reference time image, their corresponding positions still have an overall offset due to differences in imaging posture and viewing angle. Based on this, when generating the initial coordinate mapping relationship, the levee control points that maintain a stable distribution across multiple times are preferentially used as spatial anchor points. In the specific implementation process, each levee control point in the non-reference time image is matched one by one with the corresponding levee control point in the reference time image under the same coordinate reference, and the spatial alignment of the non-reference time image relative to the reference time image is determined based on the relative positional relationship between the two. It should be noted that this mapping relationship does not depend on local texture details, but is based on the overall structural form of the levee reflected by the control points, thereby forming an initial coordinate mapping relationship for subsequent fine-tuning, so that the non-reference time image is consistent with the reference time image at the overall level.

[0088] Step S1053: Based on the initial coordinate mapping relationship, the apparent image of the dike at the non-reference time is divided into multiple continuous sub-regions, and the local geometric adjustment parameters of each sub-region relative to the reference time image are determined.

[0089] In this embodiment, the initial coordinate mapping relationship eliminates the overall deviation between the non-reference time levee appearance image and the reference time image. However, due to factors such as the large length of the levee along the axis, local topographical undulations, and slight differences in shooting angle, inconsistent local misalignments still exist at different locations. Based on this, after the initial coordinate mapping relationship is established, the non-reference time levee appearance image is divided into multiple continuous and interconnected sub-regions along the levee axis, with each sub-region corresponding to a local structure of the levee. Subsequently, feature positions reflecting the structural stability of each sub-region are selected and compared with the feature positions of the corresponding sub-regions in the reference time image. By analyzing the relative offset relationship between the two under a unified coordinate reference, the local geometric adjustment parameters of each sub-region relative to the reference time image are determined. It should be noted that this method establishes local adjustments on the basis of overall mapping, maintaining the continuity between regions and providing a regional adjustment basis for subsequent fine registration.

[0090] Step S1054: Based on the local geometric adjustment parameters, perform partitioned coordinate correction on the levee appearance image at the non-reference time, and project the corrected levee appearance image onto the unified coordinate reference to complete the spatiotemporal registration of the levee appearance image.

[0091] In this embodiment, after obtaining the local geometric adjustment parameters corresponding to each continuous sub-region, the levee appearance image at non-reference times is partitioned according to the sub-regions. Coordinate correction is performed on each sub-region to ensure that the spatial position of the sub-region is consistent with the corresponding region in the reference time image. It should be noted that during the partitioned coordinate correction process, a preset connection constraint is maintained between adjacent sub-regions to avoid discontinuity in region boundaries caused by independent adjustments. Subsequently, the levee appearance image after partitioned coordinate correction is projected onto a unified coordinate reference, so that it forms a consistent spatial representation with the reference time image and other time images. Through the above processing, the levee appearance images acquired at different times achieve a consistent representation of their corresponding positions under a unified coordinate reference, thereby completing the spatiotemporal registration of the levee appearance image.

[0092] Step S2: Automatically generate a dike structure template based on the standard morphology library of dike cross sections, perform structural constraint normalization processing on the registered image, and obtain an enhanced image of the dike structure through water surface reflection suppression and vegetation occlusion perspective reconstruction.

[0093] The specific steps of step S2 are as follows:

[0094] Step S201: Call the pre-built standard morphology library of dike cross sections, automatically match the corresponding cross section morphology under a unified coordinate reference, and generate a dike structure template corresponding to the currently monitored dike section.

[0095] In this embodiment, considering the differences in design standards, construction years, and protection methods among different embankment sections, their cross-sectional structures have relatively stable but diverse characteristics. Therefore, introducing prior structural information through the cross-sectional standard morphology library helps to constrain the subsequent image processing process. In specific implementation, under a unified coordinate reference, based on the basic geometric information such as the axial direction, top width, and slope ratio of the current monitored embankment section, the cross-sectional type matching the embankment cross-sectional standard morphology library is selected. Subsequently, the selected cross-sectional type is unfolded along the embankment axis, and the cross-sectional position is aligned and adjusted according to the spatial distribution of effective embankment control points under a unified coordinate reference, thereby generating an embankment structure template consistent with the overall morphology of the current monitored embankment section.

[0096] It should be noted that the standard morphology library of levee cross sections is not an abstract collection of concepts, but a structural morphology parameter library built based on existing levee engineering design specifications, historical as-built drawings and measured cross section data. In this standard morphology library, each cross section morphology uses geometric parameters such as levee top width, slope ratio, levee height and levee toe position as structural description elements, and stores them parametrically under a unified coordinate reference.

[0097] Step S202: Project the levee structure template onto the spatiotemporally registered levee appearance image, and divide the levee appearance image into structural constraint regions based on the spatial distribution range of the levee top, levee slope and levee toe defined by the template.

[0098] The specific steps of step S202 are as follows:

[0099] Step S2021: Under a unified coordinate reference, the scale and orientation matching processing of the embankment structure template is performed so that its cross-sectional position corresponds to the embankment axis in the spatiotemporally registered embankment appearance image.

[0100] In this embodiment, since the levee structure template originates from a standard cross-sectional shape, its initial scale and orientation are of a general form of expression. However, the levee appearance image after spatiotemporal registration reflects the actual spatial orientation of the specific monitored levee section under a unified coordinate reference. Therefore, a consistent spatial correspondence needs to be established between the two. In specific implementation, firstly, the determined levee axis position in the levee appearance image is read under a unified coordinate reference, and the orientation of the levee structure template is adjusted using this axis as a direction reference, so that the normal direction of the template cross-section is consistent with the perpendicular direction of the levee axis. Subsequently, based on the distribution range of effective levee control points in the levee top and toe directions, the scale of the levee structure template in the direction perpendicular to the axis is adjusted so that the levee top, slope, and toe positions defined by the template cover the corresponding areas in the levee appearance image. Through the above scale and orientation matching processing, the levee structure template maintains spatial consistency with the levee appearance image after spatiotemporal registration under a unified coordinate reference.

[0101] Step S2022: Based on the positional relationship of the predefined dike top line, dike slope line and dike toe line in the dike structure template, generate the corresponding structural boundary trajectory in the dike appearance image.

[0102] In this embodiment, the predefined top line, slope line, and toe line in the dike structure template are used to describe the relative relationships between the key structural positions in the dike cross-section. This relative relationship is spatially stable and serves as the basis for locating the dike's apparent structure. In specific implementation, after completing the scale and direction matching, the dike structure template is projected onto the dike's apparent image along the dike axis. Based on the relative spacing and arrangement order of the structural lines in the template on the cross-section, each segment is mapped to the corresponding position in the dike's apparent image. It should be noted that when generating the structural boundary trajectory, the relative positional relationship between the structural lines is maintained first, rather than simply relying on grayscale or texture changes in the image, thereby avoiding structural line interruptions caused by water reflection, vegetation obstruction, or local damage. Through the above method, continuous structural boundary trajectories corresponding to the top, slope, and toe of the dike are formed in the dike's apparent image.

[0103] Step S2023: Using the structural boundary trajectory as a constraint, spatially segment the embankment appearance image to form a set of structural constraint regions corresponding to the embankment top region, embankment slope region, and embankment toe region, respectively.

[0104] In this embodiment, the structural boundary trajectory clearly identifies the spatial boundaries of the levee top, slope, and toe in the levee appearance image. It possesses continuity and traceability under a unified coordinate reference, thus serving as a constraint for region division. Specifically, the structural boundary trajectory is used as the segmentation boundary to spatially divide the levee appearance image along a direction perpendicular to the levee axis, ensuring that each pixel is assigned to only a single structural region. It should be noted that during spatial segmentation, pixel grayscale or texture similarity is not relied upon; instead, the structural boundary trajectory defines the region range, ensuring that the same structural region maintains a consistent spatial definition under different times and imaging conditions. Through the above segmentation process, the levee appearance image is divided into sets of structurally constrained regions corresponding to the levee top, slope, and toe regions, respectively.

[0105] Step S2024: Identify and number the structural constraint regions in the set, and establish an association between each structural constraint region and the corresponding cross-sectional region in the embankment structure template.

[0106] In this embodiment, to ensure the callability and consistency of the structural constraint regions in subsequent processing, it is necessary to uniformly identify and organize the divided structural constraint regions. Specifically, firstly, each region in the set of structural constraint regions is sequentially numbered according to the levee axis direction, ensuring a consistent region numbering order in images of the same levee segment acquired at different times. Then, each structural constraint region is assigned a corresponding region identifier to characterize its levee top, slope, or toe structure category. It should be noted that after completing the region identification and numbering, the position range of each structural constraint region under a unified coordinate reference is matched with the corresponding cross-sectional region in the levee structural template, establishing a clear association between each structural constraint region and the specific cross-sectional location in the structural template. Through the above processing, the structural constraint regions are consistent with the levee structural template in both spatial location and structural semantics.

[0107] Step S203: Within the structural constraint area, the image grayscale distribution and spatial continuity relationship of each area are uniformly scaled to ensure that the apparent images of the dike under different times and imaging conditions meet the same normalization rules within the corresponding structural areas.

[0108] In this embodiment, since the apparent images of the dike acquired at different times and under different imaging conditions differ in terms of illumination intensity, imaging angle, and environmental background, their grayscale distribution and spatial representation are not directly comparable even within the same structural region. Therefore, a unified normalization processing method needs to be introduced within the structural constraint region. Specifically, for each structural constraint region, the grayscale distribution characteristics of the pixels within that region are statistically analyzed, and combined with the continuous spatial distribution relationship of the pixels within the region, the grayscale variation range is uniformly adjusted to ensure that images of the same structural region at different times have a consistent grayscale representation benchmark. It should be noted that this adjustment process uses the structural constraint region as the smallest processing unit, avoiding the problem of mutual interference between different structural regions caused by uniform adjustment of the entire image. Through the above processing, the apparent images of the dike at different times and under different imaging conditions satisfy a consistent normalization rule within the corresponding structural region.

[0109] Step S204: For the image content located in the adjacent area of ​​the water surface, based on the spatial constraints of the water-shore boundary position in the structure template, the reflection interference area is identified, and the image information of the reflection interference area is reconstructed and replaced.

[0110] In this embodiment, the area near the water surface in the apparent image of the dike is easily affected by water reflection, and its brightness distribution and texture morphology differ significantly from the actual structure of the dike. If directly involved in subsequent analysis, it can easily interfere with structural identification. Therefore, based on this, the spatial range of the adjacent area of ​​the water surface is first defined under a unified coordinate reference according to the predefined water-shore boundary position in the dike structure template. Within this range, the image content is checked segment by segment to identify reflection interference areas that are inconsistent with the continuity of the dike structure. It should be noted that the determination of reflection interference areas is based on whether their spatial position falls within the water-shore boundary constraint range, rather than simply relying on brightness anomalies, thereby avoiding misjudging changes in the dike's appearance as reflection interference. Subsequently, for the identified reflection interference areas, the image information is reconstructed and replaced by referring to the spatial continuity relationship of their adjacent non-interference structural areas, so that the structural expression of the area is consistent with the actual contour of the dike, thereby reducing the impact of water reflection on subsequent structural analysis and anomaly identification.

[0111] Step S205: For structural constraint areas covered or obscured by vegetation, based on the cross-sectional morphology relationship of the corresponding area in the embankment structure template, perform perspective reconstruction processing on the obscured area to generate an enhanced image of the embankment structure.

[0112] In this embodiment, some structurally constrained areas in the embankment's apparent image cannot fully represent the true outline of the embankment due to vegetation growth, attachments, or temporary occlusion. Directly including these areas in subsequent analysis could easily lead to missing or discontinuous structural information. Therefore, firstly, areas with occlusion features are identified within the set of structurally constrained areas. Based on their correlation with the corresponding cross-sectional areas in the embankment structural template under a unified coordinate reference, the theoretical position of these occluded areas in the cross-sectional morphology is determined. It should be noted that the cross-sectional morphology described in the embankment structural template reflects the stable geometric relationships between the various structural elements of the embankment, serving as the constraint basis for reconstruction. Subsequently, considering the spatial continuity of adjacent unoccluded areas within the occluded area, perspective reconstruction of the structural outline of the occluded area is performed to ensure its spatial representation remains consistent with the cross-sectional morphology. Through the above processing, an enhanced embankment structural image containing complete structural information is generated.

[0113] Step S3: Using a preset bi-branch image analysis model, jointly extract the apparent features of potential hazards from the enhanced image of the embankment structure, and perform pixel-level change detection and connected domain growth on abnormal areas to form a candidate hazard target set. The apparent features of potential hazards include changes in seepage wet spots, micro-deformation of local bulges / collapses, and crack evolution.

[0114] In this embodiment, the dual-branch image analysis model is used to extract different types of potential hazards from the enhanced image of the levee structure. Its model structure includes a common input module, a first feature analysis branch, a second feature analysis branch, and a feature fusion module. The first feature analysis branch and the second feature analysis branch are set in parallel after the common input module.

[0115] The common input module receives an enhanced image of the embankment structure as input. The enhanced image of the embankment structure has completed spatiotemporal registration and structural constraint region division, and contains corresponding structural region identification information.

[0116] The first feature analysis branch is used to analyze the image region reflecting the changes in the surface state of the embankment. In this branch, the region related to the leakage phenomenon is selected based on the structural constraint region identifier. The gray-scale distribution changes and regional continuity relationships within the region are analyzed, and abnormal features related to the changes in leakage wet spots are extracted to form the first abnormal feature set.

[0117] The second feature analysis branch is used to analyze the image region reflecting the changes in the morphology of the embankment structure. In this branch, based on the spatial relationship of the structural constraint region, the morphological changes of the same structural region at different times are compared, and abnormal features related to local uplift, collapse and crack evolution are extracted to form a second set of abnormal features.

[0118] The feature fusion module is used to fuse the first set of abnormal features and the second set of abnormal features under a unified coordinate reference, merge abnormal features that have spatial overlap or temporal correlation, and output the fused abnormal region.

[0119] Before deployment, the dual-branch image analysis model is trained using sample data derived from historical levee inspection images and corresponding engineering records. The training process includes: labeling seepage spots, bulges, collapses, and cracks in the images based on historical inspection results; inputting the labeled images into the model, allowing the first feature analysis branch to learn surface state change features and the second feature analysis branch to learn structural morphology change features; through multiple rounds of training, the model can stably distinguish between normal and abnormal structural areas under structural constraints. It should be noted that the training process maintains independent learning for both branches and unified verification of the fusion module to ensure that different types of hazards do not interfere with each other.

[0120] like Figure 3 As shown, the specific steps of step S3 are as follows:

[0121] Step S301: Input the enhanced image of the embankment structure into the first feature analysis branch and the second feature analysis branch simultaneously according to the preset analysis rules. The first feature analysis branch is used to process the image area reflecting changes in surface state, and the second feature analysis branch is used to process the image area reflecting changes in structural morphology.

[0122] In this embodiment, the enhanced image of the embankment structure contains both surface state change information and structural morphology change information. These two types of information differ significantly in spatial distribution characteristics and evolution patterns. If a single path is used for unified analysis, it is easy to cause feature mixing or unclear focus. Based on the above understanding, when analyzing the enhanced image of the embankment structure, the image content is first functionally divided according to preset analysis rules. Regions reflecting changes in color, brightness, and humidity are classified as surface state analysis objects, while regions reflecting contour offset, boundary changes, and morphological continuity changes are classified as structural morphology analysis objects. Subsequently, the enhanced image of the embankment structure is simultaneously fed into the first feature analysis branch and the second feature analysis branch under a unified coordinate reference, so that the two types of analysis are carried out in parallel on the same time scale. It should be noted that the preset analysis rules can be based on pixel grayscale thresholds or on structural region types.

[0123] Step S302: In the first feature analysis branch, based on the gray-scale distribution changes and regional continuity of the enhanced image of the embankment structure, the apparent features related to the changes in seepage wet spots are extracted, and the corresponding first abnormal feature set is formed.

[0124] In this embodiment, seepage spots typically manifest as localized grayscale changes accompanied by a certain degree of spatial continuous expansion on the surface of the embankment, which is fundamentally different from sudden changes in illumination or noise interference. Based on this, in the first feature analysis branch, the grayscale distribution within each structural constraint area is first analyzed in the enhanced image of the embankment structure to identify grayscale change segments that exhibit continuous differences from the surrounding areas. Subsequently, based on the spatial coherence of the grayscale change segments, adjacent areas with similar change trends are correlated and integrated to eliminate interference from isolated pixels or discontinuous changes. It should be noted that this process focuses on the stability of grayscale changes at the regional level, rather than single-point anomalies, thereby more accurately reflecting the apparent characteristics of seepage spots evolving over time. Through the above processing, a first set of abnormal features related to the changes in seepage spots is formed.

[0125] Step S303: In the second feature analysis branch, based on the spatial positional relationship of the structural constraint area in the enhanced image of the embankment structure, morphological change features related to the evolution of local uplift, collapse and crack are extracted, and a corresponding second set of abnormal features is formed.

[0126] In this embodiment, structural hazards such as local bulges, collapses, and cracks are mainly manifested as spatial changes in the geometric shape of the embankment. These changes often occur along the boundaries or specific directions within the structurally constrained areas, exhibiting clear structural dependence characteristics. Based on this, in the second feature analysis branch, the spatial positional relationships of each structurally constrained area under a unified coordinate reference are first retrieved from the enhanced image of the embankment structure. This relationship serves as the analytical framework for comparing and analyzing the boundary orientation and contour changes within the areas. Subsequently, by comparing the spatial morphological differences of the same structurally constrained area in images at different times, morphological change segments related to the evolution of local bulges, collapses, and cracks are identified. It should be noted that this process focuses on the relative changes between structural positions, rather than changes in a single pixel, thereby avoiding interference from surface texture disturbances on morphological judgment. Through the above processing, a second set of abnormal features reflecting changes in the structural morphology of the embankment is formed.

[0127] Step S304: The first abnormal feature set and the second abnormal feature set are fused under a unified coordinate reference. Abnormal features that have spatial overlap or temporal correlation are merged, and pixel-level change detection and connected component growth are performed on the fused abnormal region.

[0128] In this embodiment, the first set of abnormal features and the second set of abnormal features reflect the changes in the surface state and structural morphology of the dike, respectively. These two types of features often have spatial coupling or temporal correlation in the actual evolution of the emergency, so they need to be comprehensively processed under a unified spatial framework. In specific implementation, the two types of abnormal features are first aligned in position under a unified coordinate reference, and abnormal features that are in the same or adjacent structural constraint areas and show correlated changes in the time series are matched. Then, abnormal features that have spatial overlap or continuous occurrence are merged to form a fused feature describing the same abnormal area. It should be noted that pixel-level change detection is not performed in isolation, but after the abnormal feature fusion is completed, the fused abnormal area is used as the initial seed area for expansion. Specifically, under a unified coordinate reference, the pixel changes in the corresponding images of the fused abnormal area at adjacent times are compared pixel by pixel to identify the changing pixels that are consistent with the abnormal area in terms of spatial position and change trend. Then, using the changing pixels as the growth starting point, the range of the abnormal area is gradually expanded according to the spatial adjacency relationship between pixels to form an abnormal connected domain with continuous boundaries, thereby completing the pixel-level change detection and connected domain growth processing.

[0129] Step S305: Based on the range of the abnormal region after the connected component is grown and its distribution position in the structural constraint region, generate corresponding candidate hazard targets, and summarize the candidate hazard targets to form a candidate hazard target set.

[0130] The specific steps of step S305 are as follows:

[0131] Step S3051: Traverse each of the abnormal regions obtained by connected component growth, determine the spatial range of each abnormal region under a unified coordinate reference, and record the spatial inclusion relationship between each abnormal region and the corresponding structural constraint region.

[0132] In this embodiment, the abnormal regions after connected component growth have formed spatial units with clear boundaries, but they only reflect the abnormal clustering at the image level and have not yet established a correspondence with the semantics of the dike structure. Based on this, the abnormal regions are traversed one by one according to the region number, and the spatial range covered by each abnormal region is determined under a unified coordinate reference. The spatial range is then compared with the pre-divided structural constraint regions. It should be noted that by judging the spatial inclusion or overlap relationship between the abnormal region and the dike top region, dike slope region, or dike toe region, the location of the abnormal region in the dike structure can be determined. Through the above processing, each abnormal region not only has a clear spatial boundary, but also obtains its corresponding structural semantics.

[0133] Step S3052: Based on the spatial inclusion relationship, map each abnormal region to at least one structural category in the top region, slope region, or toe region of the dike, and assign a corresponding structural location identifier to each abnormal region.

[0134] In this embodiment, after obtaining the spatial inclusion relationship between the abnormal region and the structural constraint region, in order to uniformly incorporate the abnormal region into the subsequent analysis process, it is necessary to explicitly classify it into a specific dike structure category. In specific implementation, based on the overlap between the abnormal region and the dike top region, dike slope region, and dike toe region under a unified coordinate reference, the abnormal region is mapped to a structural category. When the abnormal region is completely or mainly distributed within a certain structural region, it is classified into the corresponding structural category. For abnormal regions that span multiple structural regions, they are classified according to their main coverage area or structural correlation. It should be noted that after completing the structural category mapping, a corresponding structural location identifier is assigned to each abnormal region to characterize its positional attributes in the dike cross-sectional structure.

[0135] Step S3053: Combining the spatial range, structural location identifiers, and their order of appearance in the time series of the abnormal regions, each abnormal region is converted into a candidate hazard target with an independent identifier.

[0136] In this embodiment, a single anomalous region may repeatedly appear or gradually evolve at different times. To avoid misidentifying the same hazard segment as multiple independent events, it is necessary to target the anomalous region. In specific implementation, the spatial range of the anomalous region under a unified coordinate reference, the assigned structural location identifiers, and its order of appearance in the time series are comprehensively considered. Anomalous regions that are continuous in location and consistent in structural location at adjacent times are associated and integrated, and are regarded as different stages of the same potential hazard. It should be noted that after the association and integration are completed, an independent target identifier is assigned to each group of associated anomalous regions, so that they can be tracked as independent analysis objects in subsequent processing. Through the above processing, scattered anomalous regions are transformed into candidate hazard targets with spatial continuity, structural consistency, and temporal evolution attributes.

[0137] Step S3054: The generated candidate hazard targets are uniformly numbered and archived, and summarized according to their structural location identifiers and time sequence to form a candidate hazard target set.

[0138] In this embodiment, the candidate hazard targets already possess clear spatial ranges, structural location attributes, and temporal evolution relationships. To support subsequent unified evaluation and mobilization, they need to be systematically organized. In specific implementation, each candidate hazard target is uniformly numbered according to a preset numbering rule, and the spatial range, structural location identifier, and corresponding time sequence information of each candidate hazard target are archived and recorded. It should be noted that during the archiving process, the candidate hazard targets are classified according to their structural location identifiers and arranged and summarized in chronological order within the same structural category, thereby forming a set of candidate hazard targets with clear hierarchical relationships.

[0139] Figure 2 The figure shows the candidate potential danger targets formed after completing the spatiotemporal registration, structural constraint region division and abnormal feature extraction of the dike appearance image in the dynamic monitoring method of dike danger potential. The figure is based on the dike appearance image under a unified coordinate reference, and presents three main structural regions along the dike axis: the top of the dike, the slope of the dike and the toe of the dike (distinguished by colored lines). The abnormal areas identified by analysis are marked in each structural region.

[0140] Specifically, the anomalous areas located on the top of the dike are distributed in a strip along the top of the dike, and their spatial extent is consistent with the structural constraint area of ​​the top of the dike. These anomalous areas are mainly derived from the analysis of changes in the surface state of the dike and are used to characterize the possible apparent anomalies in the top of the dike. The anomalous areas located on the slope of the dike are distributed in local blocks and are located inside the structural constraint area of ​​the slope, reflecting the anomalous characteristics corresponding to changes in the structural morphology of the slope. The anomalous areas located at the toe of the dike are adjacent to the water-shore boundary and their spatial extent overlaps with the structural constraint area of ​​the toe, and are used to characterize the possible anomalies in the toe area.

[0141] It should be noted that, Figure 2 Each anomalous region shown is a continuous region formed after pixel-level change detection and connected component growth processing, and has been mapped to a corresponding structural constraint region, thus being converted into candidate hazard targets with clear spatial range and structural location attributes. These candidate hazard targets then participate in subsequent evolutionary analysis and risk classification based on their time-series characteristics, forming the dynamic monitoring results of dike hazard risks.

[0142] Step S4: Construct a dynamic risk index based on the time series image features of candidate potential hazards, classify and determine the potential hazards, and output the dynamic monitoring results of potential hazards in the dikes.

[0143] The specific steps of step S4 are as follows:

[0144] Step S401: For each candidate hazard target in the candidate hazard target set, extract the image segments corresponding to it at multiple times according to the time identifier, and establish a time series segment set of the candidate hazard target under a unified coordinate reference.

[0145] In this embodiment, the apparent state of candidate hazardous targets at different times is an important basis for assessing their evolution trend, so it is necessary to organize them continuously in the time dimension. In specific implementation, for each candidate hazardous target in the candidate hazardous target set, firstly, based on its independent identifier and time identifier, the spatial range corresponding to the target at each time is located in the spatiotemporally registered dike appearance image, and image segments matching the spatial range are extracted from the corresponding images. It should be noted that the extracted image segments are all under a unified coordinate reference to maintain spatial consistency between different time segments. Subsequently, the image segments are organized sequentially according to the time identifier to establish a time series segment set reflecting the changes of the candidate hazardous target at multiple times.

[0146] Step S402: In the set of time series segments, based on the structural location identifiers of the candidate hazardous targets, extract time series features to characterize changes in seepage wet spots, changes in local uplift / collapse, and crack evolution, and merge the time series features into target feature sequences according to their types.

[0147] In this embodiment, the manifestations and evolution paths of different types of potential hazards in the levee structure vary significantly. Therefore, targeted processing based on structural location attributes is necessary during the time series analysis phase. Specifically, in the set of time series segments corresponding to candidate hazard targets, the analysis focus is first determined based on their structural location: targets located on the levee slope or toe area are analyzed by extracting apparent changes reflecting the expansion or contraction of seepage spots; targets located on the levee surface or near structural boundaries are analyzed by extracting structural features reflecting changes in local bulges, collapses, and crack morphology. Subsequently, corresponding features are extracted from each time segment and arranged chronologically. It should be noted that by merging features of the same type along the time dimension to form independent target feature sequences, the evolution processes of different hazard types are described separately.

[0148] Step S403: Perform time consistency processing on the target feature sequence, determine the continuous relationship, discontinuity relationship and migration relationship between adjacent time points of each type of feature, and generate an evolution trajectory description corresponding to the candidate hazard target.

[0149] In this embodiment, the target feature sequence reflects the state of candidate hazardous targets at multiple moments, but its value mainly lies in the continuity and change relationship in time. Therefore, it is necessary to organize the feature sequence in a unified time consistency manner. In specific implementation, the target feature sequences of each type are aligned moment by moment according to the time identifier, and the spatial position changes and existence status of the same type of feature in adjacent moments are compared and analyzed to determine the continuity of the feature in time. It should be noted that when a feature appears in adjacent moments and its spatial position is continuous, it is considered to have a continuous relationship; when a feature is missing at a certain moment and then reappears, it is considered to have a discontinuous relationship; when a feature shows a clear spatial shift trend in adjacent moments, it is considered to have a migration relationship. By organizing the above time relationships, the discrete time series features are organized into an evolutionary trajectory description that reflects the change process.

[0150] Step S404: Based on the evolution trajectory description, assign joint values ​​to each type of feature according to the preset risk factor set, and aggregate the joint assignment results to form a dynamic risk index corresponding to the candidate risk target.

[0151] The specific steps of step S404 are as follows:

[0152] Step S4041: For each candidate danger target, read its corresponding evolution trajectory description, and generate a feature event sequence for joint assignment according to the structural position identifier, feature type identifier and time sequence recorded in the evolution trajectory description.

[0153] In this embodiment, the evolutionary trajectory description has structurally expressed the changes of candidate hazard targets over time, but its content is still mainly descriptive information and needs to be transformed into an organizational form that can be used for risk assessment. Based on this, for each candidate hazard target, the corresponding evolutionary trajectory description is first read, and the structural position identifier, feature type identifier, and the time order of appearance of each feature recorded therein are parsed. Subsequently, according to the time sequence, the different types of feature changes are organized into a set of time-arranged feature events, and the corresponding structural position attribute and feature type attribute are retained in each feature event. It should be noted that by decomposing the evolutionary trajectory description into a continuous sequence of feature events, the change process of the candidate hazard target is expressed in the form of an event chain.

[0154] Step S4042: Call the preset risk factor set and divide the risk factor set into location risk factors, evolution risk factors and coupling risk factors according to applicable conditions. The location risk factors are used to limit the assignment entry of different structural positions, the evolution risk factors are used to limit the assignment rules corresponding to the continuity, discontinuity and migration of the feature event sequence, and the coupling risk factors are used to limit the linkage assignment rules between different feature types.

[0155] In this embodiment, different candidate hazard targets have different structural locations, evolution patterns, and types of anomalies, resulting in different risk assessment focuses. Therefore, it is necessary to introduce a risk factor system with discriminative capabilities to support subsequent joint assignment. In specific implementation, a pre-set set of risk factors is first invoked, and each risk factor is functionally classified according to its applicable conditions. Risk factors related to structural locations such as the top, slope, and toe of the dike are classified as location-based risk factors, which limit the assignment entry point for different structural locations into the risk assessment process. Risk factors related to the continuity, discontinuity, and migration status of characteristic events in the time series are classified as evolution-based risk factors, which constrain the assignment method of characteristic event sequences under different evolutionary relationships. At the same time, risk factors used to describe the simultaneous occurrence or mutual influence of different types of features in the same candidate hazard target are classified as coupling-based risk factors, and the specific values ​​are set according to the general risk perception in the safety assessment of dike projects.

[0156] Step S4043: Select a value assignment channel for the candidate hazard target based on the location-based risk factor, and perform segmented joint value assignment on the characteristic event sequence corresponding to the changes in seepage wet spots, local uplift / collapse and crack evolution based on the evolution-based risk factor within the value assignment channel. At the same time, perform cross-joint value assignment on different characteristic event sequences that have temporal overlap or spatial adjacency based on the coupling-based risk factor.

[0157] In this embodiment, the candidate hazardous targets already possess clear structural location attributes and time-organized characteristic event sequences. To achieve differentiated and correlated risk assignment, the selection and combination of assignment paths need to be completed within a unified rule framework. Specifically, firstly, based on the structural location identifier corresponding to the candidate hazardous target, location-based risk factors are invoked to determine its assignment channel for risk assessment, ensuring that candidate hazardous targets with different structural locations remain logically distinct in their assignment. Subsequently, within the selected assignment channel, according to the rules defined by evolution-based risk factors, the characteristic event sequences corresponding to changes in seepage spots, changes in local uplift / collapse, and crack evolution are segmented, and joint assignments are performed based on the temporal continuity, discontinuity, or migration status of each segment. It should be noted that, based on the completion of single characteristic type assignments, further cross-joint assignments are performed on different characteristic event sequences that overlap in time or are adjacent in space, based on coupling-based risk factors, so that the correlation between multiple types of anomalies is reflected in the assignment results. Through the above process, comprehensive assignments of candidate hazardous targets at the levels of structural location, temporal evolution, and characteristic coupling are achieved.

[0158] Step S4044: Aggregate the joint assignment results and map the aggregated assignment entries into dynamic risk indices corresponding to the candidate hazard targets in chronological order.

[0159] In this embodiment, the joint assignment result includes multiple assignment entries formed by candidate hazard targets under different time periods, different feature types, and different structural positions. If used directly for judgment, it is difficult to intuitively reflect their overall change trend. Based on this, the joint assignment result is uniformly aggregated. First, the assignment entries are sorted according to time identifiers, and the assignment entries corresponding to the same candidate hazard target within the same time node are integrated to form a comprehensive assignment record under that time node. It should be noted that the temporal order relationship between assignment entries is preserved during the aggregation process, so that the changes between different time nodes are continuously tracked. Subsequently, the comprehensive assignment record arranged in chronological order is mapped to a dynamic risk index corresponding to the candidate hazard target, so that the index reflects the changing state of the candidate hazard target over time.

[0160] Step S405: Based on the dynamic risk index and the preset classification rules, the candidate risk targets are classified and judged, and the classification results are associated with the spatial range, structural location identifier and time identifier of the candidate risk targets to form the dynamic monitoring results of dike risk hazards.

[0161] In this embodiment, the dynamic risk index comprehensively expresses the changes of candidate potential hazards over time. To achieve executable monitoring output, this index needs to be converted into a risk level with clear meaning. In specific implementation, a preset grading rule is first invoked to compare and judge the values ​​of the dynamic risk index at each time node, and the risk level of the candidate potential hazard at the corresponding time is determined accordingly. It should be noted that the grading rule is not based solely on the index value at a single moment, but rather on a comprehensive judgment based on the changing trend of the dynamic risk index over time, thereby avoiding misjudgments caused by short-term fluctuations. Subsequently, the obtained grading results are associated with the spatial range, structural location identifier, and time identifier of the candidate potential hazard, forming a dynamic monitoring result of dike potential hazards that simultaneously reflects location attributes, structural attributes, and time evolution characteristics. The preset grading rule is set as 0-3 for low risk, 3-7 for medium risk, and 7-10 for high risk.

[0162] Example 2

[0163] Please see Figure 3 Another embodiment of the present invention provides: a dynamic monitoring system for potential dangers to dikes, comprising: an image processing module, an enhancement module, a danger target generation module, and a monitoring result output module;

[0164] The image processing module is used to acquire the surface image of the dike, and establish a unified coordinate reference in combination with the dike control points, and perform sub-pixel-level spatiotemporal registration on the surface image of the dike.

[0165] The enhancement module is used to automatically generate a dike structure template based on a standard dike cross-section morphology library, perform structural constraint normalization processing on the registered image, and obtain an enhanced image of the dike structure through water surface reflection suppression and vegetation occlusion perspective reconstruction.

[0166] The hazard target generation module is used to jointly extract the hazard appearance features of the enhanced image of the levee structure using a preset bi-branch image analysis model, and to perform pixel-level change detection and connected component growth on abnormal areas to form a candidate hazard target set.

[0167] The monitoring result output module is used to construct a dynamic risk index based on the time series image features of candidate potential danger targets, classify and determine the potential danger targets, and output the dynamic monitoring results of potential dangers to the dikes.

[0168] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0169] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dynamic monitoring of potential dangers to dikes, characterized in that, include: Acquire the surface image of the dike, and establish a unified coordinate reference based on the dike control points, and perform sub-pixel-level spatiotemporal registration on the surface image of the dike; Based on the standard morphology library of dike cross sections, dike structure templates are automatically generated. The registered images are then subjected to structural constraint normalization. Finally, through water surface reflection suppression and vegetation occlusion perspective reconstruction, an enhanced image of the dike structure is obtained. Using a pre-defined bi-branch image analysis model, the apparent features of potential hazards in the enhanced image of the embankment structure are jointly extracted, and pixel-level change detection and connected component growth are performed on abnormal areas to form a set of candidate potential hazards. The apparent features of potential hazards include changes in seepage wet spots, micro-deformation of local bulges / collapses, and crack evolution. A dynamic risk index is constructed based on the time-series image features of candidate potential hazards, and the potential hazards are classified and judged, outputting the dynamic monitoring results of dike hazards.

2. The method for dynamic monitoring of potential dangers to dikes as described in claim 1, characterized in that, The process of acquiring the apparent image of the dike, establishing a unified coordinate reference based on the dike control points, and performing sub-pixel-level spatiotemporal registration on the apparent image of the dike includes: The surface images of the dikes acquired by different imaging terminals at different times are obtained, and the surface images of the dikes are classified according to the imaging terminal identifier and the time identifier to form a sequence of surface images of the dikes. In the sequence of apparent images of the embankment, a stable structural line is identified, and the embankment control points corresponding to the stable structural line are extracted. The embankment control points are then filtered to obtain a set of effective embankment control points. The stable structural line includes the embankment top, the embankment toe, and the slope boundary. Based on the set of effective dike control points, a unified coordinate reference associated with the dike axis is established, and the appearance images of each dike are mapped to the unified coordinate reference. Under the unified coordinate reference, the appearance images of each dike are divided into local regions, the feature position offset relationship within the corresponding region is determined, and sub-pixel level displacement information for image registration is obtained; Using the levee appearance image at the reference time as a reference, coordinate mapping and geometric adjustment are performed on the levee appearance images at other times in chronological order, so that the levee appearance images are spatiotemporally registered under a unified coordinate reference.

3. The method for dynamic monitoring of potential dangers to dikes as described in claim 2, characterized in that, The process of using the levee appearance image at a reference time as a guide, and performing coordinate mapping and geometric adjustment on the levee appearance images at other times in chronological order, so that all levee appearance images are spatiotemporally registered under a unified coordinate reference, includes: The images of the dike are sorted according to the time stamp, and the images of the dike within the preset reference time window are selected as the reference time images. Their spatial position under the unified coordinate reference is determined as the reference mapping state. For each levee appearance image at a non-reference time, an initial coordinate mapping relationship corresponding to the reference time image is generated based on the positional relationship of the corresponding levee control points under a unified coordinate reference. Based on the initial coordinate mapping relationship, the apparent image of the dike at the non-reference time is divided into multiple continuous sub-regions, and the local geometric adjustment parameters of each sub-region relative to the reference time image are determined respectively. Based on the local geometric adjustment parameters, the apparent image of the dike at the non-reference time is subjected to partitioned coordinate correction, and the corrected apparent image of the dike is uniformly projected onto the unified coordinate reference to complete the spatiotemporal registration of the apparent image of the dike.

4. The method for dynamic monitoring of potential dangers to dikes as described in claim 3, characterized in that, The process involves automatically generating a dike structure template based on a standard dike cross-section morphology library, performing structural constraint normalization on the registered image, and obtaining an enhanced dike structure image through water surface reflection suppression and vegetation occlusion perspective reconstruction. This includes: Call the pre-built standard morphology library of dike cross sections, automatically match the corresponding cross section morphology under a unified coordinate reference, and generate a dike structure template corresponding to the currently monitored dike section accordingly. The dike structure template is projected onto the dike appearance image that has been spatiotemporally registered. Based on the spatial distribution range of the dike top, dike slope and dike toe defined by the template, the dike appearance image is divided into structural constraint regions. Within the structural constraint area, the image grayscale distribution and spatial continuity of each region are uniformly scaled to ensure that the apparent images of the dike under different times and imaging conditions meet consistent normalization rules within the corresponding structural regions. For image content located in the area adjacent to the water surface, based on the spatial constraints of the water-shore boundary position in the structural template, the reflection interference area is identified, and the image information of the reflection interference area is reconstructed and replaced. For structurally constrained areas covered or obscured by vegetation, perspective reconstruction is performed on the obscured areas based on the cross-sectional morphology of the corresponding areas in the embankment structure template to generate an enhanced image of the embankment structure.

5. The method for dynamic monitoring of potential dangers to dikes as described in claim 4, characterized in that, The dike structure template is projected onto the spatiotemporally registered dike appearance image. Based on the spatial distribution range of the dike crest, slope, and toe defined by the template, the dike appearance image is divided into structurally constrained regions, including: Under a unified coordinate reference, the scale and orientation matching processing of the embankment structure template is performed so that its cross-sectional position corresponds to the embankment axis in the spatiotemporally registered embankment appearance image. Based on the positional relationship of the predefined dike top line, dike slope line and dike toe line in the dike structure template, the corresponding structural boundary trajectory is generated in the dike appearance image; Using the structural boundary trajectory as a constraint, the apparent image of the embankment is spatially segmented to form sets of structural constraint regions corresponding to the top region, slope region, and toe region of the embankment, respectively. The set of structural constraint regions is identified and sequentially numbered, and each structural constraint region is associated with the corresponding cross-sectional region in the embankment structural template.

6. The method for dynamic monitoring of potential dangers to dikes as described in claim 5, characterized in that, The method utilizes a pre-defined bi-branch image analysis model to jointly extract the apparent features of potential hazards from the enhanced image of the levee structure, and performs pixel-level change detection and connected component growth on abnormal areas to form a candidate hazard target set, including: The enhanced image of the embankment structure is simultaneously input into the first feature analysis branch and the second feature analysis branch according to the preset analysis rules. The first feature analysis branch is used to process the image region that reflects changes in surface state, and the second feature analysis branch is used to process the image region that reflects changes in structural morphology. In the first feature analysis branch, based on the gray-scale distribution changes and regional continuity of the enhanced image of the embankment structure, the apparent features related to the changes in seepage wet spots are extracted, and the corresponding first set of abnormal features is formed. In the second feature analysis branch, based on the spatial positional relationship of the structural constraint area in the enhanced image of the embankment structure, morphological change features related to local uplift, collapse and crack evolution are extracted, and a corresponding second set of abnormal features is formed. The first set of abnormal features and the second set of abnormal features are fused under a unified coordinate reference. Abnormal features that have spatial overlap or temporal correlation are merged, and pixel-level change detection and connected component growth are performed on the fused abnormal region. Based on the range of the abnormal region after the connected domain is grown and its distribution position in the structural constraint region, corresponding candidate hazard targets are generated, and the candidate hazard targets are summarized to form a candidate hazard target set.

7. The method for dynamic monitoring of potential dangers to dikes as described in claim 6, characterized in that, The process involves generating corresponding candidate hazard targets based on the range of the abnormal region after connected component growth and its distribution position within the structural constraint region, and then aggregating these candidate hazard targets to form a candidate hazard target set, including: The abnormal regions obtained by connected domain growth are traversed one by one to determine the spatial range of each abnormal region under a unified coordinate reference, and the spatial inclusion relationship between each abnormal region and the corresponding structural constraint region is recorded. Based on the spatial inclusion relationship, each abnormal region is mapped to at least one structural category in the top region, slope region, or toe region of the dike, and a corresponding structural location identifier is assigned to each abnormal region. By combining the spatial extent, structural location identifiers, and order of appearance of the abnormal regions in the time series, each abnormal region is converted into a candidate hazard target with an independent identifier; The generated candidate hazard targets are uniformly numbered and archived, and summarized according to their structural location identifiers and time sequence to form a candidate hazard target set.

8. The method for dynamic monitoring of potential dangers to dikes as described in claim 7, characterized in that, The dynamic risk index is constructed based on the time-series image features of candidate potential hazards, and the potential hazards are classified and judged. The dynamic monitoring results of dike hazards are output, including: For each candidate hazard target in the candidate hazard target set, extract the image segments corresponding to it at multiple times according to the time identifier, and establish a time series segment set of the candidate hazard target under a unified coordinate reference; In the set of time series segments, based on the structural location identifiers of candidate hazardous targets, time series features for characterizing changes in seepage wet spots, changes in local uplift / collapse, and crack evolution are extracted respectively, and the time series features are grouped into target feature sequences according to type; The target feature sequence is processed for temporal consistency to determine the persistence, discontinuity and migration relationships of each type of feature between adjacent time points, and an evolutionary trajectory description corresponding to the candidate hazard target is generated. Based on the evolutionary trajectory description, each type of feature is jointly assigned a value according to a preset set of risk factors, and the joint assignment results are aggregated to form a dynamic risk index corresponding to the candidate risk target. Based on the dynamic risk index and preset grading rules, candidate risk targets are graded and judged, and the grading results are associated with the spatial range, structural location and time of the candidate risk targets to form dynamic monitoring results of dike risks and hidden dangers.

9. The method for dynamic monitoring of potential dangers to dikes as described in claim 8, characterized in that, Based on the evolutionary trajectory description, each type of feature is jointly assigned a value according to a preset set of risk factors, and the joint assignment results are aggregated to form a dynamic risk index corresponding to the candidate risk target, including: For each candidate hazard target, read its corresponding evolution trajectory description, and generate a sequence of feature events for joint assignment according to the structural position identifier, feature type identifier and time order recorded in the evolution trajectory description; A preset set of risk factors is invoked, and the set of risk factors is divided into location-based risk factors, evolution-based risk factors, and coupling-based risk factors according to applicable conditions. Location-based risk factors are used to limit the assignment entry points for different structural positions, evolution-based risk factors are used to limit the assignment rules corresponding to the continuity, discontinuity, and migration of the feature event sequence, and coupling-based risk factors are used to limit the linkage assignment rules between different feature types. Based on the location-based risk factors, the candidate hazard targets are assigned values ​​through a channel selection process. Within the assigned value channel, the characteristic event sequences corresponding to changes in seepage wet spots, changes in local uplift / collapse, and crack evolution are assigned values ​​in segments based on the evolution-based risk factors. At the same time, different characteristic event sequences that have temporal overlap or spatial adjacency are assigned values ​​through cross-joint processing based on the coupling-based risk factors. The joint assignment results are aggregated, and the aggregated assignment entries are mapped in chronological order to the dynamic risk index corresponding to the candidate risk target.

10. A dynamic monitoring system for potential dangers to dikes, used to implement the dynamic monitoring method for potential dangers to dikes as described in any one of claims 1-9, characterized in that, include: Image processing module, enhancement module, hazard target generation module, and monitoring result output module; The image processing module is used to acquire the surface image of the dike, and establish a unified coordinate reference in combination with the dike control points, and perform sub-pixel-level spatiotemporal registration on the surface image of the dike. The enhancement module is used to automatically generate a dike structure template based on a standard dike cross-section morphology library, perform structural constraint normalization processing on the registered image, and obtain an enhanced image of the dike structure through water surface reflection suppression and vegetation occlusion perspective reconstruction. The hazard target generation module is used to jointly extract the hazard appearance features of the enhanced image of the levee structure using a preset bi-branch image analysis model, and to perform pixel-level change detection and connected component growth on abnormal areas to form a candidate hazard target set. The monitoring result output module is used to construct a dynamic risk index based on the time series image features of candidate potential danger targets, classify and determine the potential danger targets, and output the dynamic monitoring results of potential dangers to the dikes.

Images