Cofferdam water seepage area identification system based on unmanned aerial vehicle aerial photography

Abstract

The application discloses a cofferdam water seepage area identification system based on unmanned aerial vehicle aerial photography, and belongs to the technical field of unmanned aerial vehicle remote sensing data processing and cofferdam water seepage detection. The system comprises an image acquisition module, which acquires a multispectral image sequence, acquisition time and pose data of unmanned aerial vehicle aerial photography; a material identification module, which extracts spectral reflectance curves of each pixel point and identifies the material type of ground objects; a point cloud generation module, which performs three-dimensional space registration to generate three-dimensional spectral point clouds of a cofferdam surface; a main body extraction module, which removes non-cofferdam material point clouds in the three-dimensional spectral point clouds to obtain cofferdam main body point clouds; a water content analysis module, which determines water content characteristic values of each point according to near-infrared band reflection peak value offsets, and constructs a water content characteristic distribution grid; and a water seepage identification module, which identifies a water seepage abnormal area with a linear increasing trend of water content characteristic values along the thickness direction of the cofferdam according to the water content characteristic value difference and spatial gradient direction of adjacent grid nodes.

Description

Technical Field

[0001] This invention relates to the fields of UAV remote sensing data processing and cofferdam seepage detection technology, specifically a cofferdam seepage area identification system based on UAV aerial photography. Background Technology

[0002] As temporary water-retaining structures in water conservancy projects, the rapid and accurate identification of seepage conditions in cofferdams is directly related to project safety. Current cofferdam seepage detection technologies mainly rely on contact methods such as manual visual inspection, pre-embedded piezometers or temperature sensors, and non-contact remote sensing methods based on visible light or thermal infrared imagery. Manual inspections have limited coverage, low efficiency, and rely on personnel experience, making it difficult to detect potential seepage channels within the cofferdam surface. Pre-embedded sensors are costly to deploy and can only acquire discrete point information, failing to fully reflect the spatial distribution of seepage on the cofferdam surface. Thermal infrared imagery infers seepage areas by detecting temperature anomalies, but surface watermarks, damp vegetation, and actual seepage areas have similar thermal characteristics, resulting in a high false alarm rate.

[0003] Multispectral UAV aerial photography can simultaneously acquire reflectance information of the cofferdam surface in multiple narrow bands, providing a data foundation for identifying seepage from the perspective of material spectral characteristics. Conventional seepage detection methods based on multispectral imagery mostly directly calculate the normalized water index or construct a water content inversion model. Without removing interfering features such as vegetation, silt, and temporary water accumulation on the cofferdam surface, seepage judgment is easily misreported as seepage anomalies from water sources outside the main cofferdam structure. Furthermore, existing methods typically judge seepage in isolation based on whether the absolute water content of each pixel or region exceeds a certain threshold, ignoring the hydraulic gradient characteristics of seepage gradually developing from the water-facing side to the back side of the cofferdam. This fails to effectively distinguish between uniform water content changes caused by surface wetting and residual rainfall, and seepage paths that truly penetrate the thickness of the cofferdam.

[0004] To address the above issues, it is necessary to solve how to accurately isolate non-cofferdam structures from the 3D point cloud constructed from multispectral images, retaining only the main body of the cofferdam for water content analysis, and how to utilize the spatial continuous variation trend of water content characteristics on the cofferdam surface to identify seepage anomaly areas with hydraulically increasing patterns in the thickness direction. Summary of the Invention

[0005] The purpose of this invention is to provide a cofferdam seepage area identification system based on UAV aerial photography, in order to solve the problems in the prior art where the extraction of the cofferdam main point cloud is mixed with non-cofferdam features, resulting in insufficient accuracy of water content analysis, and the failure of seepage identification to capture the hydraulic increase trend in the thickness direction of the cofferdam, leading to missed detection or misjudgment of abnormal seepage areas.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a cofferdam seepage area identification system based on UAV aerial photography, including an image acquisition module, a material identification module, a point cloud generation module, a subject extraction module, a water content analysis module, and a seepage identification module.

[0007] The image acquisition module is used to acquire multispectral image sequences taken by the UAV, as well as the acquisition time and UAV pose data for each frame.

[0008] The material identification module performs radiometric calibration on each frame of the multispectral image sequence to obtain the radiance value of each pixel in each band. Based on the radiance value of each band and the solar irradiance calibration coefficient of the corresponding band, it calculates the directional reflectance of each pixel in each band and arranges the directional reflectance of the same pixel in each band in band order to generate the spectral reflectance curve of that pixel. The spectral absorption peak position and spectral absorption peak depth of each pixel are extracted from the spectral reflectance curve and matched with the standard absorption peak parameters of each material type in the preset cofferdam material spectral feature library. The material type with the highest matching degree is determined as the ground feature material type corresponding to that pixel, thereby accurately distinguishing cofferdam materials such as concrete, rockfill, and clay core walls from non-cofferdam materials such as water, vegetation, and silt.

[0009] The point cloud generation module extracts the camera projection center coordinates and camera attitude angles corresponding to each frame of image from the UAV pose data based on the acquisition time of each frame of image. It transforms the image coordinates of each pixel in each frame of image to the same world coordinate system to obtain the initial three-dimensional coordinates. With the goal of minimizing the three-dimensional coordinate error of the same pixel in the overlapping area between adjacent frames of image, it iteratively adjusts the camera attitude angle of each frame of image to generate a high-precision three-dimensional spectral point cloud.

[0010] The main extraction module obtains the ground material type of each pixel at each point cloud location in the three-dimensional spectral point cloud. Point cloud locations where the ground material type is determined to be water, vegetation, or silt are marked as non-cofferdam points and deleted. Point cloud data where the ground material type is determined to be concrete, riprap, or clay core wall are retained to obtain the main point cloud of the cofferdam, effectively eliminating the interference of non-cofferdam objects on subsequent water content analysis.

[0011] The water content analysis module removes the envelope of the spectral reflectance curves at each point location in the main point cloud of the cofferdam. Within the envelope-removed spectral curves, it locates the maximum reflectance value in the near-infrared band and its corresponding peak wavelength. It calculates the wavelength offset of this peak wavelength relative to the standard peak wavelength of the dry cofferdam material and uses this wavelength offset as the water content characteristic value at that point location. Based on the three-dimensional coordinates of all point locations in the main point cloud of the cofferdam, a triangular irregular network is established on the cofferdam surface. Using the water content characteristic value of each vertex as the attribute value, the triangular irregular network is resampled into a regular grid. The water content characteristic value at each resampled grid node is determined by linear interpolation of the water content characteristic values ​​of the three vertices of its triangular facet, forming a water content characteristic distribution grid reflecting the fine distribution of water content on the cofferdam surface.

[0012] The seepage identification module selects the intermediate layer grid nodes located between the upstream and downstream sides of the cofferdam as reference nodes within the water-bearing characteristic distribution grid. It calculates the difference in water-bearing characteristic values ​​between each reference node and its adjacent inner and outer grid nodes along the cofferdam thickness direction, obtaining the water-bearing increment in the thickness direction. Starting from the top boundary grid node of the cofferdam, the difference in water-bearing characteristic values ​​between adjacent grid nodes is calculated layer by layer along the cofferdam thickness direction, obtaining a sequence of water-bearing increments at each depth. The module determines whether the ratio between two adjacent water-bearing increments in the sequence is within a preset ratio range. Grid nodes corresponding to continuous depth layers that satisfy the preset ratio range and whose water-bearing increments are all greater than a preset increment threshold are grouped into a continuous grid node set. The area covered by this continuous grid node set is the seepage anomaly region where the water-bearing characteristic value increases linearly along the cofferdam thickness direction. By capturing the arithmetic progression of water-bearing increments along the thickness direction, the system can effectively distinguish the gradual changes in internal water content caused by seepage, significantly improving identification reliability.

[0013] As a technical solution of the present invention, the seepage identification module also obtains the extension length and extension direction of the continuous grid node set in the horizontal direction of the cofferdam, generates the boundary contour line of the seepage anomaly area, and calculates the comprehensive seepage intensity index of the seepage anomaly area based on the area enclosed by the boundary contour line and the average water content characteristic value of each grid node in the continuous grid node set, providing a direct quantitative basis for the assessment of leakage hazards.

[0014] The system of this invention combines the three-dimensional spatial reconstruction of UAV multispectral images with the spectral characteristics of cofferdam materials and near-infrared moisture-sensitive features. Based on the accurate removal of irrelevant ground features, it constructs a water-bearing feature grid along the thickness direction of the cofferdam and captures linearly increasing seepage features. This not only achieves automated and accurate identification of seepage areas, but also outputs the spatial distribution of seepage intensity, providing efficient and reliable technical support for cofferdam seepage safety monitoring.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: The spectral reflectance curves of each pixel in the multispectral image sequence are matched with a pre-defined spectral feature library of cofferdam materials. Pixel-by-pixel identification is used to classify cofferdam materials such as concrete, riprap, and clay core walls, as well as non-cofferdam features such as water, vegetation, and silt. Then, the main body is extracted using the 3D spectral point cloud generated after 3D spatial registration. Point cloud data belonging to non-cofferdam material types are directly discarded, retaining only the main cofferdam point cloud. Using this method, interfering features such as moist vegetation, temporary surface water, and silt attached to the cofferdam surface are excluded from the water content analysis. The subsequently constructed water content feature distribution grid only reflects the water content of the cofferdam structure itself, avoiding interference from non-cofferdam water sources in seepage identification, thus improving the accuracy of identifying abnormal seepage areas.

[0016] For each location in the point cloud of the cofferdam body, the water content characteristic value at that point is quantified by using the offset of the near-infrared peak wavelength of the spectral curve after envelope removal processing relative to the standard peak wavelength of the dry cofferdam material. Based on this, a water content characteristic distribution grid of the cofferdam surface is established. By selecting the intermediate layer grid nodes between the upstream and downstream sides of the cofferdam as the benchmark, the difference in water content characteristic values ​​between adjacent grid nodes along the thickness direction of the cofferdam is calculated layer by layer to obtain the water content increment in the thickness direction. A continuous set of grid nodes with water content increments greater than a preset increment threshold and with water content increments increasing arithmetically with depth is selected, and the area covered by this set is identified as an abnormal seepage area. By utilizing the inherent hydraulic seepage law that water content characteristics increase linearly along the thickness of the cofferdam, it is possible to distinguish between the uniform surface water content increase caused by rainfall, splashing, etc., and the gradual internal water content gradient change from the water-facing side to the back side caused by seepage. This allows for the reliable detection of early linear seepage channels. At the same time, the obtained contour of the seepage anomaly area and the mean water content characteristics within the continuous node set can be directly used to calculate the comprehensive seepage intensity, providing structured information for assessing the seepage risk level. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a schematic diagram of a cofferdam seepage area identification system based on drone aerial photography; Figure 2 This is a flowchart of the method for generating and registering 3D point clouds of multispectral images; Figure 3This is a flowchart of the method for identifying and assessing the strength of abnormal seepage areas in a cofferdam. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1 This invention provides a system for identifying seepage areas in a cofferdam based on UAV aerial photography, comprising an image acquisition module, a material identification module, a point cloud generation module, a subject extraction module, a water content analysis module, and a seepage identification module. The image acquisition module acquires a multispectral image sequence from a UAV, along with the acquisition time and UAV pose data for each frame. The material identification module extracts the spectral reflectance curve of each pixel from the multispectral image sequence and identifies the corresponding ground material type based on the spectral reflectance curve. The point cloud generation module performs three-dimensional spatial registration on the multispectral image sequence based on the acquisition time and UAV pose data for each frame, generating a three-dimensional spectral point cloud of the cofferdam surface. The subject extraction module removes point cloud data belonging to non-cofferdam material types from the three-dimensional spectral point cloud based on the ground material type corresponding to each pixel, obtaining the cofferdam's main point cloud. The water content analysis module determines the water content characteristic value at each point cloud location in the main body of the cofferdam based on the shift of the reflection peak in the near-infrared band of the spectral reflectance curve at each point cloud location. It then constructs a water content characteristic distribution grid on the cofferdam surface based on the three-dimensional coordinates and water content characteristic values ​​of each point cloud location. The seepage identification module identifies anomalous seepage areas where the water content characteristic value linearly increases along the cofferdam thickness direction based on the differences in water content characteristic values ​​between adjacent grid nodes and the spatial gradient direction between adjacent grid nodes.

[0021] Example 1:

[0022] In practice, the multispectral image sequence is acquired by a multispectral camera mounted on a UAV over the cofferdam area along a predetermined flight path. Each frame contains image data from multiple bands. The radiometric calibration process for each frame in the multispectral image sequence is as follows: The radiometric calibration coefficients corresponding to each band of the multispectral camera are obtained. These coefficients include gain and offset coefficients, which are obtained by calibrating the multispectral camera's band response using a standard light source in a laboratory integrating sphere environment. For each pixel in each frame, the original digital count value in each band is multiplied by the gain coefficient of the corresponding band and then added to the offset coefficient of the corresponding band to obtain the radiance value of the pixel in the corresponding band.

[0023] In some implementations, the process of calculating the directional reflectance of each pixel in each band involves: acquiring the solar zenith angle and solar azimuth angle at the time of multispectral image acquisition, as well as the observation zenith angle and observation azimuth angle of the multispectral camera mounted on the UAV at the time of acquisition. The solar zenith angle and solar azimuth angle are calculated using astronomical algorithms based on the image acquisition time and the geographic latitude and longitude coordinates of the cofferdam area. The observation zenith angle and observation azimuth angle are calculated based on the attitude angle information of the multispectral camera in the UAV pose data of the corresponding frame image. Based on the solar zenith angle, solar azimuth angle, observation zenith angle, and observation azimuth angle, the relative geometric relationship between the solar incidence direction and the camera observation direction is calculated to obtain the solar-target-sensor geometric parameters. The solar-target-sensor geometric parameters include the phase angle between the incidence direction and the observation direction.

[0024] When calculating directional reflectance, the solar irradiance calibration coefficient for the corresponding band is obtained. This calibration coefficient is obtained by deploying a standard diffuse reflection reference plate with known reflectance on the ground, and then using a multispectral camera to vertically observe the reference plate under the same illumination conditions. The coefficient is retrieved from the known absolute reflectance of the reference plate in the corresponding band and its radiance value in the multispectral image. For the directional reflectance of a pixel in a band, the radiance value of the pixel in the corresponding band is multiplied by pi (a constant representing pi), and then divided by the product of the solar irradiance calibration coefficient for the corresponding band and the cosine of the solar zenith angle.

[0025] After obtaining the directional reflectance of each pixel in each band, the directional reflectance of a pixel in all bands is arranged in ascending order of wavelength to form a discrete spectral sequence. In the discrete spectral sequence, the horizontal axis represents the center wavelength value of the band, and the vertical axis represents the directional reflectance value. The discrete spectral sequence constitutes the spectral reflectance curve of the pixel.

[0026] Optionally, for the spectral reflectance curve of a single pixel, the position and depth of the spectral absorption peak are extracted by finding local minima. The first derivative of the spectral reflectance curve in the wavelength dimension is calculated, and the zero-crossing point where the first derivative changes from negative to positive is located. The wavelength value corresponding to the zero-crossing point is the position of the spectral absorption peak. For each spectral absorption peak position, two adjacent local maxima are obtained on the spectral reflectance curve. The line connecting the reflectance values ​​of the two local maxima is used as the local baseline of the absorption peak position. The difference between the reflectance value of the local baseline at the absorption peak position and the actual reflectance value at the absorption peak position is taken as the spectral absorption peak depth.

[0027] In practice, a pre-set spectral feature library of cofferdam materials stores standard absorption peak parameters for various common land cover material types in the cofferdam area. The material types in this library include concrete, riprap, clay core walls, water, vegetation, and silt. The standard absorption peak parameters for each material type include several standard absorption peak positions and corresponding standard absorption peak depths. These parameters are obtained through statistical analysis of multispectral images and spectral feature extraction of various known material areas during the cofferdam construction phase.

[0028] The process of matching the spectral absorption peak position and depth of a pixel with the standard absorption peak parameters of each material type in the spectral feature library of cofferdam materials is as follows: First, each material type in the spectral feature library is taken as a candidate material type. For each candidate material type, several standard absorption peak positions are obtained from the standard absorption peak parameters. The absolute value of the positional deviation between each spectral absorption peak position of the pixel and all the standard absorption peak positions of the candidate material type is calculated. Spectral absorption peaks with an absolute value of positional deviation less than a preset positional tolerance threshold are identified as matching absorption peaks of the candidate material type. The preset positional tolerance threshold is determined based on the spectral sampling interval of the multispectral camera and is set to half of the spectral sampling interval of the multispectral camera.

[0029] For each matching absorption peak in the candidate material type, obtain the value of the matching absorption peak in the pixel's spectral absorption peak depth and the value in the candidate material type's standard absorption peak depth, and calculate the absolute value of the depth deviation between the pixel's spectral absorption peak depth and the standard absorption peak depth. Average the absolute values ​​of the depth deviations of all matching absorption peaks for the candidate material type according to the number of matching absorption peaks to obtain the average depth deviation of the candidate material type. The average depth deviation of the candidate material type is determined as the matching degree of the candidate material type. After traversing all candidate material types in the cofferdam material spectral feature library, the candidate material type with the smallest matching degree value is determined as the ground feature material type corresponding to that pixel.

[0030] In some implementations, when there are no candidate material types in the cofferdam material spectral feature library that match the spectral absorption peak position and depth of a pixel with a value less than a preset matching threshold, the corresponding land cover material type is marked as an unknown type. The preset matching threshold is set based on the minimum difference between the standard absorption peak depths of various material types in the cofferdam material spectral feature library, and the preset matching threshold is taken as half of the minimum difference value.

[0031] Example 2:

[0032] In specific implementation, please refer to Figure 2 The UAV pose data is recorded in real time by the UAV flight control system during aerial photography. This data includes the timestamp of each multispectral image acquisition frame, UAV positioning data, and UAV attitude data. The UAV positioning data consists of the UAV's longitude, latitude, and geodetic height coordinates from the Global Navigation Satellite System (GNSS). The UAV attitude data includes the UAV's pitch, roll, and yaw angles. Based on the acquisition time of each multispectral image frame, positioning and attitude data matching the timestamp of the acquisition time are extracted from the UAV pose data. When the timestamp of the acquisition time does not perfectly match the timestamp recorded in the UAV pose data, a linear interpolation method is used to calculate the positioning and attitude data corresponding to the acquisition time using the positioning and attitude data corresponding to two adjacent timestamps.

[0033] The longitude, latitude, and geodetic height coordinates corresponding to the acquisition time are converted into camera projection center coordinates in the world coordinate system. The world coordinate system adopted is the Northeast-Sky coordinate system with the center point of the cofferdam area as the origin. The conversion process is as follows: first, the longitude, latitude, and geodetic height coordinates are converted to the geocentric rectangular coordinate system, and then the coordinates in the geocentric rectangular coordinate system are converted to the Northeast-Sky world coordinate system. The conversion parameters for the longitude, latitude, and geodetic height coordinates to the geocentric rectangular coordinate system adopt the WGS-84 ellipsoid parameters. The origin longitude coordinates, origin latitude coordinates, and origin geodetic height coordinates of the Northeast-Sky world coordinate system are preset according to the location of the center point of the cofferdam area.

[0034] The pitch, roll, and yaw angles corresponding to the acquisition time are used as the camera attitude angles. Based on the camera projection center coordinates and the camera attitude angles, the image coordinates of each pixel in a frame of multispectral imagery are transformed to the Northeast-Northeast UTF-8 coordinate system to obtain the initial three-dimensional coordinates of each pixel. The transformation relationship is as follows:

[0035] in, , , These are the north, east, and elevation components of the initial 3D coordinates of the pixel in the world coordinate system, respectively. , , These are the north, east, and elevation components of the camera projection center coordinates, respectively. is the projection scaling factor, which is a positive real scalar to be solved; Let be the rotation matrix formed by the camera attitude angles. From pitch angle Roll angle and yaw angle The yaw angle, pitch angle, and roll angle are combined in a rotating order. It is a 3x3 matrix; and These are the image column coordinates and image row coordinates of the pixel, respectively; and These are the column coordinates and row coordinates of the principal points of the multispectral camera, provided in the camera calibration report. This is the focal length value of the multispectral camera, and the unit is consistent with the image coordinate unit.

[0036] When solving for the initial 3D coordinates, an initial reference elevation value is set. Initial reference elevation value Take the design elevation value of the top of the cofferdam. Let... Substitute the values ​​into the third row of the transformation equation to solve for the scaling factor. The numerical value; then the calculated scaling factor. Substitute the numerical values ​​into the first and second rows of equations in the transformation relation, and solve them respectively. and The numerical values ​​are obtained by performing the above transformation process on each pixel in each frame of multispectral image to obtain the initial three-dimensional coordinates of all pixels, forming an initial three-dimensional spectral point cloud. Each point in the initial three-dimensional spectral point cloud contains the north component coordinates, the east component coordinates, the elevation component coordinates, and the multi-band spectral reflectance information inherited from the corresponding pixel in the multispectral image.

[0037] In some implementations, the corresponding pixels are obtained as follows: for two adjacent multispectral images, the grayscale images of each image are extracted, and the grayscale images are synthesized from the visible light band data in the multispectral images; on the grayscale images of the two adjacent images, feature points are extracted using a scale-invariant feature transform algorithm, and descriptor matching is performed on the feature points to obtain a set of matching point pairs in the overlapping area between the two adjacent images, and each pair of matching points in the set of matching point pairs is the corresponding pixel between the two adjacent images.

[0038] After obtaining the corresponding pixels between adjacent frames, the camera pose angle of each frame is iteratively adjusted with the goal of minimizing the 3D coordinate error of the corresponding pixels in the overlapping area between adjacent frames. A total error function is constructed, and its expression is:

[0039] in, This represents the total error function value. This is the set of camera attitude angle parameters for all frames, including frames 1 to 2. The frame's pitch, roll, and yaw angles; This represents the total number of frames in the multispectral image sequence. and Frame sequence number index; For the first Frame image and the first The total number of matching pixel pairs with the same name between frames; The index of the pair of pixels with the same name; For the first For pixels with the same name, those belonging to the first... The pixels of the frame image, using the first Camera pose angle of frame image , , The three-dimensional coordinate vector is calculated using the above transformation relationship; For the first For pixels with the same name, those belonging to the first... The pixels of the frame image, using the first Camera pose angle of frame image , , The three-dimensional coordinate vector is calculated using the above transformation relationship; The Euclidean norm of a vector is denoted by .

[0040] In each iteration, the camera projection center coordinates of all frames are kept constant, and the total error function is used. Minimizing the value of is the optimization objective. The Levenberg-Marquardt algorithm is used to solve for the correction of the camera pose angles in each frame. The solved correction is added to the camera pose angle parameters of the current iteration step to update the camera pose angles of each frame. Using the updated camera pose angles, the 3D coordinates of all pixels are recalculated according to the transformation relationship, and the corresponding pixels between adjacent frames are re-extracted. The total error function value is then recalculated. This iterative process is repeated until the relative change in the total error function value is less than a preset convergence threshold, which is set to . After the iteration ends, the final registered camera attitude angles and the corresponding 3D spectral point cloud are obtained. The coordinates of each point in the final 3D spectral point cloud have been uniformly registered to the Northeast-Northeast UTC coordinate system, forming the registered 3D spectral point cloud.

[0041] Example 3:

[0042] In practice, each point cloud location in the 3D spectral point cloud corresponds to a pixel in the multispectral image. The point cloud location inherits the row and column indices of the corresponding pixel in the multispectral image during the point cloud generation process. The method for obtaining the land cover material type of the pixel corresponding to each point cloud location in the 3D spectral point cloud is as follows: Read the multispectral image frame number associated with the point cloud location, as well as the pixel column and row coordinates in that frame. Locate the corresponding frame image in the multispectral image sequence based on the frame number. In the land cover material type identification result image of the corresponding frame image, find the pixel indicated by the pixel column and row coordinates, extract the identified land cover material type label for that pixel, and assign the extracted land cover material type label to the point cloud location.

[0043] The land cover material type identification result image is generated by the material identification module after identifying the land cover material type corresponding to each pixel. The image is a single-channel raster image of the same size as the multispectral image, where each pixel stores a land cover material type code value. The encoding rules for the land cover material type code values ​​are as follows: concrete is assigned a value of 1, riprap is assigned a value of 2, clay core wall is assigned a value of 3, water body is assigned a value of 4, vegetation is assigned a value of 5, and silt is assigned a value of 6. A corresponding land cover material type identification result image is generated for each frame in the multispectral image sequence.

[0044] The material type label of each point location in the 3D spectral point cloud is compared with a preset set of non-cofferdam material types. The non-cofferdam material type set includes three material types: water, vegetation, and silt. When the material type label of a point location corresponds to any one of water, vegetation, or silt, that point location is marked as a non-cofferdam point. The method for marking non-cofferdam points is as follows: in the point cloud data structure of the 3D spectral point cloud, a Boolean removal flag is set for the point cloud record corresponding to the non-cofferdam point, and the removal flag is set to true.

[0045] In some implementations, a unique identifier is assigned to each point cloud location, and an index array is constructed. The length of the index array is equal to the total number of points in the 3D spectral point cloud. Each element in the index array stores the removal flag status of the corresponding point cloud location. All point cloud locations are traversed, and the land cover material type label at each location is determined sequentially. For point cloud locations where the land cover material type code value corresponding to the label is equal to 4, 5, or 6, it indicates that the land cover material type is determined to be water, vegetation, or silt, and the removal flag of the element corresponding to the unique identifier in the index array is set to true. For point cloud locations where the land cover material type code value corresponding to the label is equal to 1, 2, or 3, it indicates that the land cover material type is determined to be concrete, riprap, or clay core wall, and the removal flag of the element corresponding to the unique identifier in the index array is set to false.

[0046] After marking all point cloud locations, the method for deleting point cloud data corresponding to non-cofferdam points from the 3D spectral point cloud is as follows: traverse all point cloud records in the 3D spectral point cloud, remove point cloud records with a true removal flag from the point cloud dataset, and retain point cloud records with a false removal flag. After the removal operation is completed, renumber the remaining point cloud records consecutively according to their unique point cloud identifier.

[0047] The material types corresponding to the feature material type labels in the retained point cloud records include concrete, riprap, and clay core wall, constituting the main point cloud of the cofferdam. Each point cloud location in the main point cloud of the cofferdam contains three-dimensional spatial coordinates and corresponding spectral reflectance information. The three-dimensional spatial coordinates are determined by the point cloud generation module during the three-dimensional spatial registration process, and the spectral reflectance information includes the directional reflectance values ​​of each multispectral band.

[0048] In some implementations, the horizontal coverage of the cofferdam's main point cloud is defined by the water-facing and backwater-facing boundaries of the cofferdam. The water-facing side is the surface of the cofferdam facing the reservoir water, and the backwater side is the surface of the cofferdam facing away from the reservoir water. The top boundary of the cofferdam's main point cloud is determined by the design elevation of the cofferdam's top, and the bottom boundary is determined based on the boundary line between the cofferdam and the original terrain. The spatial distribution of the point cloud data in the cofferdam's main point cloud can reflect the geometric shape and surface continuity of the cofferdam structure in three-dimensional space.

[0049] Example 4:

[0050] In practice, the spectral reflectance curve at each point location in the cofferdam's main point cloud undergoes envelope removal processing. The envelope removal process is as follows: The spectral reflectance curve at a point location is acquired. This curve consists of several discrete band-reflectance data pairs, where bands are represented by wavelength values ​​and reflectance by directional reflectance values. Within the wavelength range of the spectral reflectance curve, all local maxima are identified. Local maxima are bands whose reflectance values ​​are greater than those of their adjacent bands. All identified local maxima are connected sequentially with straight line segments in ascending order of wavelength, forming a continuous broken line segment, which is the envelope of the spectral reflectance curve. The envelope reflectance value at each band location is calculated through linear interpolation of the envelope at the corresponding wavelength. The original reflectance value of the spectral reflectance curve at each band is divided by the corresponding envelope reflectance value to obtain the spectral reflectance value after envelope removal. Arrange the spectral reflectance values ​​after removing the envelopes of all bands in wavelength order to generate the spectral curves after envelope removal.

[0051] In some implementations, the near-infrared band range is set to 700 nm to 1300 nm. In the spectral curve after envelope removal, all bands within the near-infrared band range are used as the search range. The spectral reflectance values ​​of all bands after envelope removal are traversed within the search range. The maximum value among the spectral reflectance values ​​after envelope removal is found and recorded as the maximum reflectance value. The wavelength value of the band corresponding to the maximum reflectance value is recorded as the peak wavelength position.

[0052] The standard peak wavelength position of the dry cofferdam material was obtained as follows: During the cofferdam construction phase, multispectral images were acquired of the area of ​​the cofferdam material known to be in a dry state. The dry state was determined by the cofferdam material not being in contact with liquid water for 72 consecutive hours before the acquisition and the relative humidity of the surrounding environment being below 40%. The spectral reflectance curve of the multispectral image of the dry cofferdam material area was extracted and envelope removal was performed. The wavelength position corresponding to the maximum reflectance value in the near-infrared band of the spectral curve after envelope removal was located and used as the standard peak wavelength position of the dry cofferdam material. The standard peak wavelength positions of the dry cofferdam material were determined separately for different cofferdam material types: concrete material corresponds to a standard peak wavelength position for dry concrete, rockfill material corresponds to a standard peak wavelength position for dry rockfill, and clay core wall material corresponds to a standard peak wavelength position for dry clay core wall.

[0053] When determining the water content characteristic value at a point cloud location, the corresponding feature material type label is first obtained. Then, the dry standard peak wavelength position (DBS) for the corresponding cofferdam material type is selected based on the feature material type label. The feature material type label is read from the point cloud attributes of the main cofferdam point cloud, and its value is one of concrete, rockfill, or clay core wall. When the feature material type label at the point cloud location is concrete, the dry standard peak wavelength position for concrete is selected; when the feature material type label is rockfill, the dry standard peak wavelength position for rockfill is selected; and when the feature material type label is clay core wall, the dry standard peak wavelength position for clay core wall is selected.

[0054] Calculate the wavelength offset of the peak wavelength position relative to the standard peak wavelength position of the selected dry cofferdam material. The wavelength offset is calculated using the following formula:

[0055] in, This refers to the index number of the point cloud location within the point cloud of the main body of the cofferdam. The value of is a positive integer. The total number of points increasing from 1 to the center point of the main body of the cofferdam; For the first The wavelength offset at each point cloud location, in nanometers; For the first The peak wavelength position at each point cloud location, in nanometers; For the first The standard peak wavelength position of the dry cofferdam material selected at each point cloud location based on its land cover material type label, in nanometers. Wavelength offset. Determined as the number The water content characteristic value is a value at a point cloud location, with the unit being nanometers. The larger the value of the water content characteristic value, the more significant the change in water content at the point cloud location relative to the dry state.

[0056] After obtaining the water content characteristics of all point cloud locations in the main point cloud of the cofferdam, a triangular irregular network of the cofferdam surface is established based on the three-dimensional coordinates of all point cloud locations in the main point cloud of the cofferdam. The three-dimensional coordinates are represented by the north component coordinates, east component coordinates, and elevation component coordinates in the Northeast-Northeast Celestial coordinate system. The triangular irregular network is established as follows: the three-dimensional coordinates of all point cloud locations in the main point cloud of the cofferdam are used as a scatter set, and the Delaunay triangulation algorithm is used to spatially triangulate the scatter set. The Delaunay triangulation algorithm is executed on a two-dimensional plane, taking the north component coordinates and east component coordinates of each point cloud location in the scatter set as plane coordinate inputs. The elevation component coordinates are retained during the triangulation process but do not participate in the triangulation decision. The triangulation algorithm generates several triangular patches, each of which has three point cloud locations as vertices, and no other point cloud locations are located inside the circumcircle of the triangular patch between the three vertices. The triangular irregular network is composed of all generated triangular patches.

[0057] The water content feature value at each vertex of the triangular irregular network is used as the attribute value of the corresponding vertex. The water content feature value of the point cloud position corresponding to each vertex in the triangular irregular network has been calculated in the previous steps, and this water content feature value is directly assigned to the vertex as an attribute value for storage.

[0058] The process of resampling a triangular irregular network into a regular grid involves defining the planar extent and grid resolution of the regular grid. The planar extent of the regular grid is determined by the minimum and maximum values ​​of the north and east component coordinates of all vertices in the triangular irregular network. The planar extent extends from the minimum to the maximum value of the north component coordinates in the north direction, and from the minimum to the maximum value in the east direction. The grid resolution of the regular grid is set based on the average point spacing of the 3D spectral point cloud. The grid resolution in both the north and east directions is set to half the average Euclidean distance between adjacent points in the 3D spectral point cloud. The regular grid consists of several rectangular grid cells, each with a length defined by the grid resolution in both the north and east directions.

[0059] For each grid node in a regular grid, determine the triangular facet containing the grid node. The planar position of the grid node is represented by its north and east component coordinates. Traverse all triangular faces in the irregular triangular network. For each triangular facet, obtain the north and east component coordinates of its three vertices, forming a planar triangular region. Determine whether the planar position of the grid node is located inside or on the boundary of the planar triangular region. The determination method is: calculate the sum of the areas of the three sub-triangles formed by the grid node and the three vertices of the planar triangular region. If the sum of the areas of the three sub-triangles equals the area of ​​the planar triangular region, then the grid node is located inside or on the boundary of the planar triangular region, and the triangular facet containing the grid node is the corresponding triangular facet.

[0060] Obtain the water content characteristic values ​​of the three vertices of the triangular facet containing the mesh node. The water content characteristic values ​​of the three vertices are denoted as follows: , and The north and east component coordinates of the three vertices on the plane are denoted as follows: , and The coordinates of the north and east components of the grid nodes are denoted as follows: Based on the relative positions of the mesh nodes and the three vertices of the triangular facets, the barycentric coordinate coefficients from the mesh nodes to the three vertices are calculated. The barycentric coordinate coefficients are obtained by solving the following system of linear equations: the coordinates of the mesh nodes in the north direction satisfy... The coordinates of the grid nodes in the east direction satisfy... And the centroid coordinate coefficients satisfy the constraints ,in , and These are the centroid coordinate coefficients of the mesh nodes relative to the first, second, and third vertices, respectively. Solving for them yields... , and After obtaining the numerical values, the water content characteristic value at each grid node is determined by linear interpolation using the water content characteristic values ​​of the three vertices of the triangular facet containing that node. The linear interpolation expression is as follows: ,in The value represents the water content characteristic at each grid node. The above interpolation calculation is performed on all grid nodes in the regular grid to obtain the water content characteristic value at each grid node. All grid nodes and their corresponding water content characteristic values ​​constitute the water content characteristic distribution grid.

[0061] Example 5:

[0062] In specific implementation, please refer to Figure 3The water-bearing characteristic distribution grid is a two-dimensional regular grid. The two-dimensional planar coordinates of the grid nodes correspond to the planar unfolded positions of the cofferdam surface, and the attribute values ​​of the grid nodes are water-bearing characteristic values. The cofferdam thickness direction is defined as the direction from the upstream face of the cofferdam to the downstream face. The upstream face is the side of the cofferdam facing the reservoir water, and the downstream face is the side of the cofferdam facing away from the reservoir water. The boundaries of the upstream and downstream faces of the cofferdam are pre-marked in the water-bearing characteristic distribution grid. The marking method is as follows: based on the planar projection coordinate range of the upstream and downstream faces in the cofferdam design drawings, the set of grid nodes for the upstream face boundary and the set of grid nodes for the downstream face boundary are delineated in the water-bearing characteristic distribution grid.

[0063] In the water-bearing characteristic distribution grid, intermediate layer grid nodes located between the upstream and downstream faces of the cofferdam are selected as reference nodes. The selection method for intermediate layer grid nodes is as follows: starting from each upstream grid node in the upstream boundary grid node set, proceeding node by node along the cofferdam thickness towards the downstream face, determining all grid nodes between the upstream grid node and the corresponding downstream grid node in the downstream boundary grid node set, forming a thickness-direction grid node chain. In the thickness-direction grid node chain, the total number of grid nodes in the chain is calculated, denoted as [the number of nodes is missing in the original text]. , the first in the grid node chain One grid node is determined as the reference node, among which This indicates a floor operation. For each mesh node on the water-facing side corresponding to a mesh node chain in the thickness direction, a reference node is determined, and all reference nodes constitute a reference node set.

[0064] The difference in water-bearing characteristic values ​​between each reference node and its adjacent inner and outer grid nodes along the thickness direction of the cofferdam is calculated to obtain the water-bearing increment in the thickness direction. For each reference node in the reference node set, the inner direction is the direction from the reference node to the water-facing side of the cofferdam, and the outer direction is the direction from the reference node to the water-repelling side of the cofferdam. An adjacent grid node along the inner direction is denoted as the inner grid node, and an adjacent grid node along the outer direction is denoted as the outer grid node. Water-bearing characteristic values ​​of the reference nodes, inner grid nodes, and outer grid nodes are extracted from the water-bearing characteristic distribution grid. The absolute value of the difference between the water-bearing characteristic value of the inner grid node and the water-bearing characteristic value of the reference node is calculated as the inner water-bearing increment; the absolute value of the difference between the water-bearing characteristic value of the outer grid node and the water-bearing characteristic value of the reference node is calculated as the outer water-bearing increment. The average value of the inner and outer water-bearing increments is determined as the water-bearing increment in the thickness direction.

[0065] In some implementations, the process of selecting a continuous set of grid nodes whose thickness-direction water cut increments are greater than a preset threshold and whose thickness-direction water cut increments increase arithmetically with depth is as follows: Starting from the top boundary grid node of the cofferdam, the difference in water cut characteristic values ​​between adjacent grid nodes is calculated layer by layer along the cofferdam thickness direction to obtain the water cut increment sequence for each depth layer. The top boundary grid node of the cofferdam is a grid node in the water cut characteristic distribution grid whose elevation component coordinates are equal to the design elevation value of the top of the cofferdam. The layering method in the thickness direction of the cofferdam is as follows: Starting from the top boundary grid node of the cofferdam, along the direction of decrease in cofferdam elevation, the grid nodes are divided into multiple depth layers according to the grid resolution step size of the water cut characteristic distribution grid, and the elevation difference between adjacent depth layers is equal to the grid resolution. For each depth layer, the difference in water cut characteristic values ​​between all grid nodes in that depth layer and the corresponding grid nodes in the next depth layer adjacent to it along the cofferdam thickness direction is calculated, and the absolute value of the difference is taken as the water cut increment at the corresponding position. The average value of the water cut increments at all positions in the same depth layer is taken to obtain the layer water cut increment corresponding to the depth layer. The layers at different depths are arranged in descending order of elevation to form a water-bearing increment sequence, denoted as [missing information]. ,in The depth layer number. The depth layer corresponding to the grid node closest to the top boundary of the cofferdam. The increasing depths correspond to layers with gradually decreasing elevations.

[0066] Determine whether the ratio between two adjacent water cut increments in the water cut increment sequence falls within a preset ratio range. The preset ratio range is... ,in This is the ratio tolerance coefficient. The value is 0.15. The basis for setting it to 0.15 is as follows: when the difference between the water-bearing increments of two adjacent layers does not exceed 15%, the relationship between the water-bearing increment and depth can be approximated as an arithmetic progression. This tolerance range can accommodate local fluctuations in the spatial distribution of water-bearing characteristics without affecting the overall trend judgment. The ratio of two adjacent water-bearing increments is calculated as follows: for depth layer number... and Calculate the ratio . judge Does it meet the requirements? .

[0067] The preset incremental threshold value is determined based on the background fluctuation range of the peak wavelength shift of the dry cofferdam material in the near-infrared band. During the cofferdam construction phase, multispectral images are repeatedly acquired for the area of ​​the cofferdam material in a known dry state, and the standard deviation of the wavelength shift under dry conditions is calculated. The preset incremental threshold is set to This setup can eliminate minor fluctuations in the water content characteristics of the dry cofferdam material caused by surface roughness or observation angle.

[0068] Grid nodes corresponding to consecutive depth layers that satisfy a preset ratio range and whose water cut increments are all greater than a preset increment threshold are added to a consecutive grid node set. Starting from the initial depth layer of the water cut increment sequence, the process iterates through the layers. When the first depth layer with a water cut increment greater than the preset increment threshold is found, this depth layer is recorded as the initial anomalous layer. From the initial anomalous layer, each subsequent depth layer is checked to see if its water cut increment is greater than the preset increment threshold, and the ratio of the current depth layer's water cut increment to the previous depth layer is checked to see if it is within a preset ratio range. If multiple consecutive depth layers simultaneously satisfy both conditions, all grid nodes of the consecutive depth layers are added to the consecutive grid node set. If a depth layer does not satisfy the condition that its water cut increment is greater than the preset increment threshold or its ratio is not within the preset ratio range, the extension of the current consecutive depth layer is stopped, and the addition of one consecutive grid node set is completed. After traversing all depth layers, several consecutive grid node sets are obtained. The area covered by each consecutive grid node set is identified as a seepage anomaly area.

[0069] After assigning the grid nodes corresponding to the continuous depth layers that meet the conditions to a continuous grid node set, the extension length and direction of the continuous grid node set in the horizontal direction of the cofferdam are obtained. The horizontal direction of the cofferdam is the direction perpendicular to the thickness direction of the cofferdam and extends along the axis of the cofferdam. For all grid nodes assigned to the continuous grid node set, the north and east component coordinates of the grid nodes are extracted. The minimum and maximum values ​​of the north and east component coordinates of all grid nodes in the continuous grid node set are calculated to determine the circumscribed rectangle of the continuous grid node set in the horizontal plane. The direction of the long side of the circumscribed rectangle is determined as the extension direction, and the length of the long side of the circumscribed rectangle is determined as the extension length.

[0070] Based on the extension length and direction, the boundary contour of the seepage anomaly region is generated. Using the circumscribed rectangle of the continuous mesh node set as the initial contour, the region is extended outwards from both ends of the circumscribed rectangle along the extension direction by a distance equal to one mesh resolution. Then, along the width direction perpendicular to the extension direction, the region is extended outwards from both sides of the circumscribed rectangle by a distance equal to half a mesh resolution, resulting in an extended rectangular region. A convex hull calculation is performed on the continuous mesh node set within the extended rectangular region using the Graham scan algorithm. The convex hull polygon serves as the boundary contour of the seepage anomaly region. The boundary contour is formed by connecting several vertices sequentially, with the coordinates of each vertex determined by the north and east component coordinates of the mesh node.

[0071] Based on the area enclosed by the boundary contour line and the average water-bearing characteristic value of each grid node in the continuous grid node set, the comprehensive seepage intensity index of the seepage anomaly region is calculated. The calculation formula is:

[0072] in, The comprehensive permeability intensity index for areas with abnormal permeability is expressed in nanometers per square meter. The area enclosed by the boundary contour line is obtained by integrating the area using the shoelace formula over the coordinates of the convex hull polygon vertices, and the unit is square meters. This represents the average water-bearing characteristic value of all grid nodes in a continuous grid set. The average water-bearing characteristic value is calculated by summing the water-bearing characteristic values ​​of all grid nodes in the continuous grid set and then dividing by the total number of grid nodes in the continuous grid set. The unit is nanometers. (Comprehensive permeability intensity index) The larger the value, the more significant the overall performance of the abnormal seepage area in terms of spatial range and degree of water content abnormality.

[0073] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A system for identifying seepage areas in cofferdams based on UAV aerial photography, characterized in that, include: The image acquisition module acquires multispectral image sequences taken by the UAV, as well as the acquisition time and UAV pose data for each frame of the image. The material identification module extracts the spectral reflectance curve of each pixel based on the multispectral image sequence, and identifies the type of ground material corresponding to each pixel based on the spectral reflectance curve. The point cloud generation module performs three-dimensional spatial registration on the multispectral image sequence based on the acquisition time of each frame of image and the pose data of the UAV, and generates a three-dimensional spectral point cloud on the surface of the cofferdam. The main body extraction module removes point cloud data belonging to non-cofferdam material types from the three-dimensional spectral point cloud based on the ground material type corresponding to each pixel, thus obtaining the main body point cloud of the cofferdam. The water content analysis module determines the water content characteristic value at each point cloud location based on the shift of the reflection peak of the spectral reflectance curve at each point cloud location in the near-infrared band, and constructs a water content characteristic distribution grid on the surface of the cofferdam based on the three-dimensional coordinates and water content characteristic values ​​of each point cloud location in the cofferdam main body point cloud. The seepage identification module identifies seepage anomaly areas where the water-bearing characteristic values ​​increase linearly along the cofferdam thickness direction, based on the differences in water-bearing characteristic values ​​between adjacent grid nodes in the water-bearing characteristic distribution grid and the spatial gradient direction between adjacent grid nodes.

2. The cofferdam seepage area identification system based on UAV aerial photography according to claim 1, characterized in that, Based on the multispectral image sequence, the spectral reflectance curve of each pixel is extracted, including: Radiometric calibration is performed on each frame of the multispectral image sequence to obtain the radiance value of each pixel in each band. Based on the irradiance value of each band and the solar irradiance calibration coefficient of the corresponding band, the directional reflectance of each pixel in each band is calculated. Arrange the directional reflectance of the same pixel in each band according to the band order to generate the spectral reflectance curve of that pixel.

3. The cofferdam seepage area identification system based on UAV aerial photography according to claim 2, characterized in that, Based on the spectral reflectance curve, the type of ground material corresponding to each pixel is identified, including: Extract the spectral absorption peak position and spectral absorption peak depth of each pixel from the spectral reflectance curve; Based on the position and depth of the spectral absorption peak, the parameters are matched with the standard absorption peak parameters of each material type in the preset spectral feature library of cofferdam materials. The material type that best matches the standard absorption peak parameters is determined as the ground feature material type corresponding to that pixel.

4. The seepage area identification system for cofferdams based on UAV aerial photography according to claim 1, characterized in that, Based on the acquisition time of each image frame and the UAV pose data, the multispectral image sequence is registered in three-dimensional space, including: Based on the acquisition time of each frame of image, the camera projection center coordinates and camera attitude angles corresponding to each frame of image are extracted from the UAV pose data. Based on the camera projection center coordinates and the camera attitude angle, the image coordinates of each pixel in each frame are transformed to the same world coordinate system to obtain the initial three-dimensional coordinates of each pixel. With the goal of minimizing the 3D coordinate error of the same-named pixels in the overlapping area between adjacent frames, the camera attitude angle of each frame is iteratively adjusted to obtain the registered 3D spectral point cloud.

5. The cofferdam seepage area identification system based on UAV aerial photography according to claim 4, characterized in that, Based on the land feature material type corresponding to each pixel, point cloud data belonging to non-cofferdam material types are removed from the three-dimensional spectral point cloud, including: Obtain the ground material type of the corresponding pixel at each point cloud location in the three-dimensional spectral point cloud; Point cloud locations whose ground features are classified as water bodies, vegetation, or silt are marked as non-cofferdam points. Delete the point cloud data corresponding to the non-cofferdam points from the three-dimensional spectral point cloud, and retain the point cloud data of the ground material type determined to be concrete, riprap, or clay core wall, as the main point cloud of the cofferdam.

6. The seepage area identification system for cofferdams based on UAV aerial photography according to claim 1, characterized in that, Based on the shift of the reflectance peak in the near-infrared band of the spectral reflectance curves at each point cloud location, the water content characteristic values ​​at each point cloud location are determined, including: The envelope of the spectral reflectance curves at each point cloud location is removed to obtain the spectral curves after envelope removal. In the spectral curve after the envelope is removed, locate the maximum reflectance value and its corresponding peak wavelength position in the near-infrared band. Calculate the wavelength offset of the peak wavelength position relative to the standard peak wavelength position of the dry cofferdam material, and determine the wavelength offset as the water content characteristic value at that point cloud position.

7. The seepage area identification system for cofferdams based on UAV aerial photography according to claim 1, characterized in that, Based on the three-dimensional coordinates and water-bearing characteristic values ​​of each point cloud position in the main body of the cofferdam, a water-bearing characteristic distribution grid is constructed on the surface of the cofferdam, including: Based on the three-dimensional coordinates of all point cloud positions in the main point cloud of the cofferdam, a triangular irregular network is established on the surface of the cofferdam. The water content characteristic value at each vertex of the triangular irregular network is used as the attribute value of that vertex; The triangular irregular network is resampled into a regular grid. The water content feature value at each resampled grid node is determined by linear interpolation using the water content feature values ​​of the three vertices of the triangular facet where it is located, thus obtaining the water content feature distribution grid.

8. The cofferdam seepage area identification system based on UAV aerial photography according to claim 7, characterized in that, Based on the differences in water-bearing characteristic values ​​between adjacent grid nodes in the water-bearing characteristic distribution grid, and the spatial gradient direction between adjacent grid nodes, seepage anomaly regions where the water-bearing characteristic values ​​show a linear increasing trend along the cofferdam thickness direction are identified, including: In the water-bearing characteristic distribution grid, the intermediate layer grid node located between the upstream and downstream sides of the cofferdam is selected as the reference node; The difference in water content characteristic values ​​between each reference node and its adjacent inner and outer grid nodes along the thickness direction of the cofferdam is calculated to obtain the water content increment in the thickness direction. A continuous set of grid nodes is selected where the increase in water content in the thickness direction is greater than a preset threshold and the increase in water content in the thickness direction increases arithmetically with depth. The area covered by the continuous set of grid nodes is identified as an abnormal seepage area.

9. The cofferdam seepage area identification system based on UAV aerial photography according to claim 8, characterized in that, The set of continuous grid nodes that are selected based on the water content increment in the thickness direction being greater than a preset threshold and the water content increment in the thickness direction increasing arithmetically with depth includes: Starting from the top boundary grid node of the cofferdam, the difference in water-bearing characteristic value between adjacent grid nodes is calculated layer by layer along the thickness direction of the cofferdam to obtain the water-bearing increment sequence of each depth layer. Determine whether the ratio between two adjacent water content increments in the water content increment sequence is within a preset ratio range; The grid nodes corresponding to continuous depth layers that satisfy the preset ratio range and whose water content increments in each layer are greater than the preset increment threshold are included in the continuous grid node set.

10. The cofferdam seepage area identification system based on UAV aerial photography according to claim 9, characterized in that, After classifying the grid nodes corresponding to continuous depth layers that satisfy the preset ratio range and whose water content increments in each layer are greater than the preset increment threshold into the continuous grid node set, the following further includes: Obtain the extension length and extension direction of the continuous grid node set in the horizontal direction of the cofferdam; Based on the extension length and the extension direction, a boundary contour line of the abnormal seepage area is generated; Based on the area enclosed by the boundary contour line and the average water content characteristic value of each grid node in the continuous grid node set, the comprehensive seepage intensity index of the seepage anomaly area is calculated.

Images