River channel grid laying area construction quality evaluation method and system

Abstract

The application provides a river channel grid laying area construction quality evaluation method and system. It belongs to the field of river channel grid laying area construction quality evaluation. The method comprises the following steps: establishing a river channel grid three-dimensional model with real geographic coordinates based on unmanned aerial vehicle oblique photography; identifying the laying area of the river channel grid three-dimensional model based on a preset YOLOv8 model, connecting the three-dimensional geographic coordinates of the corner points or key points of the laying area to form a polygon boundary describing the three-dimensional space range of the laying area; spatially registering the point cloud data of the laying area with design drawings, setting direction constraint conditions according to the boundary parameter of the laying area in the design drawings to perform point cloud segmentation, generating a structured report according to the laying area, thickness and fender / cage size coding rules of the design drawings, and importing the design value deviation into a BIM model to visualize the over-limit area through color mapping. The method can evaluate and effectively control the construction progress and quality in the construction process.

Description

Technical Field

[0001] This application relates to the field of construction quality assessment in river grid laying areas, and more specifically, to a method and system for construction quality assessment in river grid laying areas. Background Technology

[0002] The quality control of traditional river grid engineering construction mainly relies on manual inspection, and is mainly based on random sampling. This has problems such as low efficiency, strong subjectivity, and large measurement errors, making it difficult to effectively control the construction progress and quality during the construction process. Summary of the Invention

[0003] The purpose of this application is to provide a method and system for assessing the construction quality of river grid laying areas, which can assess and effectively control the construction progress and quality during the construction process.

[0004] This application is implemented as follows: Firstly, this application provides a method for assessing the construction quality of a river grid laying area, including: Based on the acquisition of multi-directional and multi-angle geomorphic image data of river grid area by UAV oblique photography, the coordinate position information of the shooting point and the attitude information at the time of shooting are recorded to form the exterior orientation element data of the image. Based on the geomorphic image data and the exterior orientation element data, a three-dimensional model of river grid with real geographic coordinates is generated. Based on the preset YOLOv8 model, the laying area of ​​the river grid 3D model is identified, and the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area are connected to form a polygon boundary describing the 3D spatial range of the laying area. The point cloud data of the paving area is spatially registered with the design drawings. After registration, directional constraints are set according to the boundary parameters of the paving area in the design drawings to segment the point cloud, extracting the set of points in the point cloud that satisfy the plane equation. The formula is: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; Based on the segmented point cloud, the laying area, thickness, and mattress / gabion dimensions are obtained. The laying area, thickness, and mattress / gabion dimensions are then used to generate a structured report according to the coding rules of the design drawings. The design deviations of the laying area, thickness, and material unfolding length are imported into the BIM model, and the out-of-tolerance areas are visualized through color mapping.

[0005] Based on the first aspect, it also includes: training a ridge regression model based on historical deviation data to predict the material loss rate of the next paving area.

[0006] Based on the first aspect, the steps for obtaining the laying area from the segmented point cloud include: The segmented point cloud is processed by a region growing algorithm to merge the point sets of continuous surfaces, and the least squares method is used to fit the equation of the paving plane to obtain the fitted plane. The segmented point cloud is projected onto the fitting plane, and the minimum convex hull boundary is generated using the Alpha-Shape algorithm to calculate the polygon area. For non-planar tiling, differential subdivision is performed along the direction of the surface normal vector. The segmented point cloud is divided into triangular meshes. The actual surface area is obtained by summing the areas of all triangles. The calculation results are output to the reporting system and compared with the design value. Areas with deviations exceeding the limit are marked.

[0007] Based on the first aspect, the steps for obtaining the laying thickness from the segmented point cloud include: Based on the layer thickness parameters in the design drawings, a vertical profile line is generated along the direction of the normal vector of the segmented point cloud. On the profile line, the point cloud of the upper and lower surfaces is fitted using local weighted regression LOESS, and the Euclidean distance between the layers is calculated as the measured thickness value.

[0008] Based on the first aspect, the steps for obtaining the mattress / gabion dimensions from the segmented point cloud include: The height, length, and width of the entire mattress / gabion are obtained by fitting the structural boundary line and combining it with camera intrinsic and extrinsic parameters or point cloud calibration conversion calculations. The mesh size of the mattress / gabion is calculated using the following formula: ; in, The three-dimensional coordinates of the intersection points of the skeleton lines. For the number of measurements; d mesh Mesh size; The specific method for extracting the binding spacing is as follows: After arranging the identified binding point coordinates in order, the spacing between adjacent binding points is calculated using the following formula: ; in, Indicates the spacing between adjacent binding points. Indicates the first Location of each binding point; M This represents the total number of identified tying points.

[0009] Based on the first aspect, the pre-defined methods for constructing YOLOv8 models include: The target scene images containing the paving area are obtained as the training sample set, covering different viewpoints, lighting conditions and occlusion situations, and the training sample set is preprocessed. Input the prepared target sample set into the initialized YOLOv8 model to perform transfer learning training; After training, the model's performance is evaluated using an independent test sample set. Each image in the test set is input into the trained YOLOv8 model to perform inference. The model outputs prediction information for each detected tiled area, including the class confidence of its bounding box and precise pixel-level coordinates. Low-confidence predicted boxes and overlapping boxes are filtered out according to the preset confidence threshold and non-maximum suppression (NMS) threshold, retaining high-quality detection results. For images in the test set where the model successfully detected the paving area, the coordinate information of all predicted bounding boxes that meet the confidence requirements is extracted and saved. Establish a mapping relationship between image coordinates and real-world geographic coordinates.

[0010] Based on the first aspect, the steps for spatially registering the point cloud data of the laying area with the design drawings include: Feature points are extracted from the design drawings, including wall corners and pipe joints. These feature points are then matched with corresponding feature points in the point cloud. The Iterative Closest Point (ICP) algorithm is used to calculate the rotation matrix and translation vector to align the coordinate systems of the point cloud and the drawings. For areas with local deformation, the Thin Plate Spline Interpolation (TPS) algorithm is introduced to compensate for non-rigid deformation, ensuring that the geometric error between the point cloud and the drawings is less than a threshold.

[0011] Secondly, this application provides a construction quality assessment system for river grid laying areas, comprising: The 3D modeling module is configured to: collect multi-directional and multi-angle geomorphic image data of the river grid area based on UAV oblique photography, and record the coordinate position information of the shooting point and the attitude information at the time of shooting to form the exterior orientation element data of the image. Based on the geomorphic image data and the exterior orientation element data, a 3D model of the river grid with real geographic coordinates is generated. The identification module is configured to: identify the laying area of ​​the river grid 3D model based on the preset YOLOv8 model, and connect the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area to form a polygon boundary describing the 3D spatial range of the laying area. The registration and segmentation module is configured to: spatially register the point cloud data of the paving area with the design drawings; after registration, set directional constraints based on the boundary parameters of the paving area in the design drawings to segment the point cloud, extracting a set of points in the point cloud that satisfy the plane equation, using the following formula: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; The quality feedback module is configured to: obtain the laying area, thickness, and mattress / gabion dimensions from the segmented point cloud; generate a structured report of the laying area, thickness, and mattress / gabion dimensions according to the coding rules of the design drawings; import the design value deviations of the laying area, thickness, and material unfolding length into the BIM model; and visualize the out-of-tolerance areas through color mapping.

[0012] Thirdly, this application provides an electronic device, comprising: Memory, used to store one or more programs; processor; The above method is implemented when one or more programs are executed by the processor.

[0013] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0014] Secondly, this application provides a solution.

[0015] Compared with the prior art, this application has at least the following advantages or beneficial effects: This application provides a method and system for assessing the construction quality of river grid laying areas. It utilizes UAV oblique photogrammetry to generate a 3D model of the river grid with real geographic coordinates, significantly improving data acquisition capabilities in complex terrain environments. Compared to the inefficiency and difficulty in covering high-risk areas associated with traditional manual surveying, UAV oblique photogrammetry not only improves the efficiency of field data acquisition but also covers dangerous areas inaccessible to manual methods. The generated high-resolution real-scene 3D model provides millimeter-level precision data support for subsequent measurement and analysis, enhancing the reliability of engineering quality control. The laying area, thickness, and mattress / gabion dimensions are obtained from the segmented point cloud. A structured report is generated based on the laying area, thickness, and mattress / gabion dimensions according to the design drawing coding rules. The design deviations of the laying area, thickness, and material unfolding length are imported into the BIM model, and areas exceeding the tolerance are visualized through color mapping. This achieves accurate measurement of construction targets and effectively avoids and reduces engineering data deviations caused by errors in manual identification or recording. It makes up for the shortcomings of traditional manual inspections, and compared with existing smart construction site platforms, it can control the quality of construction while managing the construction progress, which is more comprehensive and accurate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for assessing the construction quality of a river grid laying area according to this application. Figure 2 This is a schematic diagram of a photograph of a gabion / shield used in a method for assessing the construction quality of a river gabion laying area according to this application. Figure 3 This is a schematic diagram illustrating the extraction of the mesh size of the mattress / gabion in the construction quality assessment method for the river gabion laying area of ​​this application; Figure 4 This is a schematic diagram illustrating the extraction of the binding spacing of the protective mattress / gabion in a method for assessing the construction quality of a river grid laying area according to this application. Figure 5 This is a structural schematic diagram of a construction quality assessment system for a river grid laying area according to this application; Figure 6 This is a schematic diagram of the structure of an electronic device according to this application.

[0018] icon: 1. 3D modeling module; 2. Recognition module; 3. Registration and segmentation module; 4. Quality reflection module; 5. Processor; 6. Memory; 7. Communication interface. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the various embodiments and features described below can be combined with each other.

[0021] Example This application provides a method for assessing the construction quality of a river grid laying area, which can evaluate and effectively control the construction progress and quality during the construction process.

[0022] Please refer to Figure 1The method for assessing the construction quality of a river grid laying area includes the following steps: S1: Based on the acquisition of multi-directional and multi-angle river grid area image data by UAV oblique photography, the coordinate position information of the shooting point and the attitude information at the time of shooting are recorded to form the exterior orientation element data of the image. Based on the image data and exterior orientation element data, a three-dimensional model of the river grid with real geographic coordinates is generated. Specifically, to achieve accurate geometric reconstruction of the river grid area, an oblique photogrammetry system was used on a drone. Through 3D flight path planning, an autonomous measurement mode with omnidirectional obstacle avoidance and obstacle detour was implemented to acquire high-resolution terrain image data. Simultaneously, the coordinates of the shooting points and the attitude information during shooting were recorded to form the exterior orientation data of the image, which was used to identify the grid laying progress. Based on the terrain image data and exterior orientation data, a 3D model of the river grid with real geographic coordinates was generated to obtain grid laying dimension parameters. After completing the oblique photogrammetry real-scene 3D model, any local problems that might exist in the initially generated 3D model were edited and refined. This step significantly improved the data acquisition capability in complex terrain environments. Compared to the low efficiency and difficulty in covering high-risk areas of traditional manual surveying, drone oblique photogrammetry data acquisition not only improves the efficiency of field data acquisition but also covers dangerous areas that cannot be reached manually. The generated high-resolution real-scene 3D model provides millimeter-level precision basic data support for subsequent metrological analysis, improving the reliability of engineering quality control. S2: Based on the preset YOLOv8 model, identify the laying area of ​​the river grid 3D model, connect the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area to form a polygon boundary describing the 3D spatial range of the laying area. This step, using a pre-defined YOLOv8 model, can process data across the entire project area instantly (within seconds), automatically identifying all "paved" areas without any manual intervention. It creates polygonal boundaries describing the three-dimensional spatial extent of the paved area; it offers measurability: the polygonal boundaries directly define the precise location and extent of the paved area in the real world. The perimeter and area of ​​this polygon can be calculated immediately, allowing the project team to understand the project progress. It is also integrable: the polygonal boundaries can be directly imported into various management platforms for overlay analysis and conflict detection with design drawings, land acquisition boundaries, and other engineering components. Furthermore, it provides traceability: each identification result can be saved as a vector layer. By comparing polygons from different dates, project progress can be automatically and quantitatively calculated (e.g., how many square meters of paved area were added today compared to yesterday). Traditional methods require manual, aimless searching for analysis points throughout the entire area to calculate paving thickness. This step precisely describes the polygonal boundaries of the paved area's three-dimensional spatial extent, requiring analysis only within these boundaries. Subsequent point cloud thickness analysis can be focused entirely on this area, avoiding calculations for useless areas (such as original river channels and mounds), significantly improving computational efficiency and accuracy. The number of gabions and the amount of wire mesh used can be quickly estimated based on this area.

[0023] As one implementation method, the pre-defined YOLOv8 model construction method includes: acquiring target scene images containing the tiled area as a training sample set, covering different viewpoints, lighting conditions, and occlusion situations, and preprocessing the training sample set; preprocessing includes size normalization (e.g., adjusting to 640x640 pixels to adapt to YOLOv8 input), color space normalization, and applying data augmentation techniques (e.g., random rotation, cropping, flipping, brightness / contrast adjustment) to improve the model's generalization ability. Simultaneously, the bounding boxes of the tiled area in the image are accurately labeled, generating label files in YOLO format (txt) or COCO format (JSON). Lightweight base models such as the officially pre-trained YOLOv8n (Nano) or YOLOv8s (Small) are selected. These models have been pre-trained on large general datasets (such as COCO) and possess powerful feature extraction capabilities and fast inference speeds, serving as the starting point for transfer learning.

[0024] The prepared target sample set is input into the initialized YOLOv8 model to perform transfer learning training. During this process, a specific training strategy is employed: typically, most layers of the pre-trained model's backbone (e.g., the first 10 layers) are frozen to preserve its ability to extract general low-level features (such as edges and textures), while only the network layers near the output (e.g., the head part) and some deep backbone layers are unfrozen for fine-tuning. Iterative training is performed on a training subset of the target sample set using an optimizer (such as SGD or AdamW) and an appropriate learning rate scheduler (such as cosine annealing). The core objective of training is to optimize the model's detection performance on the laid-out regions. The loss function comprehensively considers the accuracy of bounding box position regression (e.g., CIoU Loss), target confidence, and class prediction.

[0025] After training, the model's performance is evaluated using an independent test sample set. Each image in the test set is input into the trained YOLOv8 model to perform inference. The model outputs prediction information for each detected tiled area, including the class confidence of its bounding box and precise pixel-level coordinates. Low-confidence predicted boxes and overlapping boxes are filtered out according to the preset confidence threshold and non-maximum suppression (NMS) threshold, retaining high-quality detection results. For images in the test set where the model successfully detected the tiled region, the coordinate information of all predicted bounding boxes that met the confidence requirements was extracted and saved. This coordinate information needed to be converted from the normalized format output by the model (usually a scale value relative to the input image size) back to absolute pixel coordinates at the original input image size. Specifically, the normalized center point coordinates were multiplied by the image's width and height to obtain the absolute center point coordinates, and then combined with the normalized width and height to calculate the absolute pixel coordinates of the top-left and bottom-right corners of the bounding box. These coordinates precisely defined the two-dimensional position and extent of the tiled region in the original input image.

[0026] Establishing a mapping relationship between image coordinates and real-world geographic coordinates is crucial. To achieve the transformation from two-dimensional image coordinates to three-dimensional geographic space through 2D-3D spatial mapping calculations, it's necessary to establish this mapping relationship. This can be achieved using known camera parameters (intrinsic parameters such as focal length and principal point, extrinsic parameters such as camera pose) or georeferenced information (such as ground control points). Spatial geometric transformation methods are employed, most commonly calculating the homography matrix or directly using camera imaging models (such as pinhole models combined with distortion correction) and external positioning and orientation system (POS) data. Using these models or matrices, the two-dimensional image coordinates (x, y) of the four corner points (or contour key points) of the paving area boundary obtained in the previous step are projected or back-projected to determine the corresponding geodetic coordinates (X, Y, Z) of these points in three-dimensional geographic space (e.g., WGS84 coordinate system or local projection coordinate system). The elevation Z may need to be obtained through digital elevation model (DEM) interpolation, stereo vision matching, or LiDAR point cloud registration.

[0027] The calculated 3D geographic coordinates (X, Y, Z) of the corner points (or key points) of the paving area are connected to form a polygon boundary describing the 3D spatial extent of the paving area. Depending on the application requirements, this 3D geographic coordinate data is organized into a standard GIS data format. For vector data, GeoJSON format (lightweight and easy for web applications) or Shapefile format (industry-standard) are preferred. These files contain geometric information (vertex coordinate sequences) of polygon features and possible attribute information (such as paving area type, confidence level, timestamp, etc.). For raster data or situations requiring more precise description, registration information files (such as World Files) can also be generated. These GeoJSON or Shapefile files are the final output 3D geographic coordinate boundary data, which can be directly provided to 3D Geographic Information System (3D GIS) or Building Information Modeling (BIM) platforms for loading and visualization. They serve as a precise spatial reference for the target paving area during 3D model registration, supporting applications in planning, construction, and monitoring.

[0028] S3: Spatial registration is performed between the point cloud data of the paving area and the design drawings. After registration, directional constraints are set according to the boundary parameters of the paving area in the design drawings to segment the point cloud, extracting the set of points in the point cloud that satisfy the plane equation. The formula is: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; Traditional on-site scanning data (point cloud) and design CAD drawings exist in two completely independent coordinate systems. Like a photograph of a real object and a design sketch, although they depict the same thing, their sizes, angles, and positions don't match, making direct comparison and measurement impossible. This step, through registration, "aligns" the real-world scanned point cloud data to the coordinate system of the design drawings. The design drawings provide the ideal normal vector direction (n) and position (d) for the paving surface. This is achieved through formulas... Only retain those that are almost entirely located on the design plane (allowing for some) This method, which filters points based on their location rather than simply "selecting" them by their physical attribute of "whether they belong to the target plane," can extremely accurately remove all point clouds containing debris that are not on the plane, retaining only the points that truly constitute the grid surface, thus providing an incredibly "clean" dataset for subsequent analysis.

[0029] One implementation method involves spatially registering the point cloud data of the paving area with the design drawings. This includes: extracting feature points from the design drawings, such as wall corners and pipe joints; matching these feature points with corresponding feature points in the point cloud; and using the Iterative Nearest Point (ICP) algorithm to calculate the rotation matrix and translation vector to align the coordinate systems of the point cloud and the drawings. For areas with local deformation, the Thin Plate Spline Interpolation (TPS) algorithm is introduced to compensate for non-rigid deformation, ensuring that the geometric error between the point cloud and the drawings is less than a threshold (e.g., 2mm), thus establishing a spatial reference for subsequent measurement and analysis. In this step, after point cloud segmentation, statistical outlier filtering (SOR) can be applied to the segmented point cloud to remove noise points, retaining the effective paving area point cloud.

[0030] S4: Based on the segmented point cloud, obtain the laying area, thickness, and mattress / gabion dimensions. Generate a structured report based on the laying area, thickness, and mattress / gabion dimensions according to the design drawing coding rules. Import the design value deviations of the laying area, thickness, and material unfolding length into the BIM model and visualize the out-of-tolerance areas through color mapping.

[0031] Specifically, traditional methods for measuring the laying area rely on manual measurement with a measuring tape or estimation based on a coarse grid method, resulting in large errors and low efficiency. For measuring the laying thickness, destructive methods such as inserting probes are used for spot checks, making it impossible to achieve full-area, non-destructive testing. For already laid grids, many dimensions cannot be directly measured. This application, based on high-precision point clouds, can calculate the thickness per square meter, the accurate area of ​​the entire region, and the dimensions of each gabion, providing comprehensive measurement results. The geometric parameters derived from point cloud computing achieve accuracy down to the centimeter or even millimeter level, far exceeding the accuracy of manual operations, providing a reliable data foundation for project settlement and quality assessment. By calculating the deviation between measured and design values, this deviation data is directly imported into the BIM model. The deviation value for each area is visually presented on the BIM model using color rendering (red: negative deviation >10%; yellow: negative deviation 5%~10%; green: normal).

[0032] As one implementation method, the step of obtaining the laying area from the segmented point cloud includes: The segmented point cloud is processed using a region growing algorithm to merge continuous surface point sets, and then the least squares method is used to fit the paving plane equation to obtain a fitted plane. The region growing algorithm merges adjacent points with similar characteristics in the segmented point cloud (these points are roughly on a plane) into a larger, continuous point set. This effectively handles planes that may be broken due to noise or minor undulations. The least squares fitted plane optimally approximates this continuous point set using a mathematical plane (equation of ax + by + cz + d = 0). This fitted plane represents the "ideal average plane" of the entire paving area. The least squares method effectively eliminates random errors and minor undulations in measurement, finding the plane that best represents the overall paving trend and avoiding interference from local unevenness in the overall area calculation.

[0033] The segmented point cloud is projected onto the fitting plane, and the Alpha-Shape algorithm is used to generate the minimum convex hull boundary to calculate the polygon area. Specifically, the traditional convex hull algorithm calculates a minimum convex polygon that can enclose all points. However, if the paving area is concave (e.g., an "L" or "U" shaped river channel), the convex hull will significantly overestimate the area. This embodiment uses the Alpha-Shape algorithm to generate a concave hull, which can identify the concave parts on the boundary, thereby generating a polygon that highly matches the contour of the actual paving area, resulting in extremely accurate calculations. The area of ​​this concave hull polygon is the projected area of ​​the paving area.

[0034] For non-planar surfaces, differential subdivision is performed along the normal vector direction of the curved surface. The resulting point cloud is divided into triangular meshes, and the areas of all triangles are summed to obtain the actual surface area. The calculation result is output to the reporting system and compared with the design value, marking areas with deviations exceeding limits. Specifically, for curved surfaces, the projected area will be smaller than the actual surface area (like a crumpled piece of paper, the projected area decreases, but the surface area remains the same). The triangular mesh method can restore its true size. The calculated area value (whether projected or actual surface area) is automatically filled into a structured report (such as Excel or CSV), automatically retrieving the area value from the design drawing for that area, and calculating the deviation: Deviation rate = (Measured value - Design value) / Design value 100%. If the deviation exceeds the allowable range of the specification (e.g., >±5%), the system will specially mark the area in the report (e.g., highlight it) and visualize it in the BIM model (e.g., turn it red).

[0035] The steps for obtaining the laying thickness based on the segmented point cloud include: Based on the layer thickness parameters in the design drawings, a vertical profile line is generated along the normal vector direction of the segmented point cloud. On this profile line, a local weighted regression (LOESS) is used to fit the point clouds of the upper and lower surfaces, and the Euclidean distance between the layers is calculated as the measured thickness value. Specifically, on irregular surfaces, simply measuring thickness vertically downwards will result in errors. For example, on a slope, a vertical downward measurement will overestimate the actual thickness. In this embodiment, measuring along the normal vector direction is the shortest distance that truly passes through the material layer, i.e., the true thickness. Considering that neither the base layer of the lower surface nor the gabion on the upper surface is perfectly smooth—they are composed of stones, soil, etc.—this embodiment uses the LOESS algorithm for fitting. For each point on the profile line, a fitting value is calculated individually based on the distribution of its neighboring points, thus forming a smooth curve to represent the overall trend of the surface. After obtaining two smooth trend lines representing the "lower surface" and "upper surface" through LOESS, the Euclidean distance between these two lines in three-dimensional space is calculated at each point on the profile line.

[0036] The steps for obtaining the mattress / gabion dimensions from the segmented point cloud include: The height, length, and width of the entire mattress / gabion are obtained by fitting the structural boundary line and combining it with camera intrinsic and extrinsic parameters or point cloud calibration conversion calculations. The mesh size of the mattress / gabion is calculated using the following formula: ; in, The three-dimensional coordinates of the intersection points of the skeleton lines. For the number of measurements; d mesh This refers to the mesh size. Please refer to it. Figure 2-3, Figure 2 This is an illustration of a mattress / gabion photograph; Figure 3 A schematic diagram for extracting the mesh size of mattresses / gabion; The specific method for extracting the binding spacing is as follows: After arranging the identified binding point coordinates in order, the spacing between adjacent binding points is calculated using the following formula: ; in, Indicates the spacing between adjacent binding points. Indicates the first Location of each binding point; M This represents the total number of identified lashing points. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram illustrating the extraction of the binding spacing for mattresses / gabions.

[0037] In some embodiments of the present invention, the method further includes training a ridge regression model based on historical deviation data to predict the material loss rate of the next paving area. This setup transforms material procurement from "rough estimation" to "precise prediction," directly reducing waste at the source and achieving significant cost savings.

[0038] Please refer to Figure 5 Secondly, this application provides a construction quality assessment system for river grid laying areas, comprising: The 3D modeling module 1 is configured to: collect multi-directional and multi-angle geomorphic image data of the river grid area based on UAV oblique photography, and record the coordinate position information of the shooting point and the attitude information at the time of shooting to form the exterior orientation element data of the image, and generate a 3D model of the river grid with real geographic coordinates based on the geomorphic image data and the exterior orientation element data. The identification module 2 is configured to: identify the laying area of ​​the river grid 3D model based on the preset YOLOv8 model, and connect the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area to form a polygon boundary describing the 3D spatial range of the laying area. Registration and segmentation module 3 is configured to: spatially register the point cloud data of the paving area with the design drawings; after registration, set directional constraints based on the boundary parameters of the paving area in the design drawings to segment the point cloud, and segment out the point set that satisfies the plane equation, as shown in the formula: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; The quality feedback module 4 is configured to: obtain the laying area, thickness and mattress / gabion size from the segmented point cloud; generate a structured report of the laying area, thickness and mattress / gabion size according to the coding rules of the design drawings; import the design value deviation of the laying area, thickness and material unfolding length into the BIM model; and visualize the out-of-tolerance area through color mapping.

[0039] For a detailed implementation of a construction quality assessment system for river grid laying areas, please refer to the detailed implementation of the above-mentioned method for assessing construction quality in river grid laying areas; further details will not be provided here.

[0040] Please refer to Figure 6 Thirdly, this application provides an electronic device, comprising: Memory 6 is used to store one or more programs; Processor 5; Processor 5 and memory 6 are connected via communication interface 7; The above method is implemented when one or more programs are executed by processor 5.

[0041] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor 5, implements the above-described method.

[0042] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for evaluating the quality of construction of a river grid laying area, characterized by, include: Based on the acquisition of multi-directional and multi-angle geomorphic image data of river grid area by UAV oblique photography, the coordinate position information of the shooting point and the attitude information at the time of shooting are recorded to form the exterior orientation element data of the image. Based on the geomorphic image data and the exterior orientation element data, a three-dimensional model of river grid with real geographic coordinates is generated. Based on the preset YOLOv8 model, the laying area of ​​the river grid 3D model is identified, and the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area are connected to form a polygon boundary describing the 3D spatial range of the laying area. The point cloud data of the paving area is spatially registered with the design drawings. After registration, directional constraints are set according to the boundary parameters of the paving area in the design drawings to segment the point cloud, extracting the set of points in the point cloud that satisfy the plane equation. The formula is: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; Based on the segmented point cloud, the laying area, thickness, and mattress / gabion dimensions are obtained. The laying area, thickness, and mattress / gabion dimensions are then used to generate a structured report according to the coding rules of the design drawings. The design deviations of the laying area, thickness, and material unfolding length are imported into the BIM model, and the out-of-tolerance areas are visualized through color mapping. The steps for obtaining the laying thickness based on the segmented point cloud include: Based on the layer thickness parameters in the design drawings, a vertical profile line is generated along the direction of the normal vector of the segmented point cloud. On the profile line, the point cloud of the upper and lower surfaces is fitted using local weighted regression LOESS, and the Euclidean distance between the layers is calculated as the measured thickness value.

2. The method for assessing the construction quality of a river grid laying area according to claim 1, characterized in that, Also includes: A ridge regression model is trained based on historical bias data to predict the material loss rate of the next paving area.

3. The method for evaluating the quality of river channel grid laying area construction according to claim 1, characterized in that, The steps for obtaining the laying area from the segmented point cloud include: The segmented point cloud is processed by a region growing algorithm to merge the point sets of continuous surfaces, and the least squares method is used to fit the equation of the paving plane to obtain the fitted plane. The segmented point cloud is projected onto the fitting plane, and the minimum convex hull boundary is generated using the Alpha-Shape algorithm to calculate the polygon area. For non-planar tiling, differential subdivision is performed along the direction of the surface normal vector. The segmented point cloud is divided into triangular meshes. The actual surface area is obtained by summing the areas of all triangles. The calculation results are output to the reporting system and compared with the design value. Areas with deviations exceeding the limit are marked.

4. The river channel grid laying area construction quality evaluation method according to claim 1, characterized in that, The steps for obtaining the mattress / gabion dimensions from the segmented point cloud include: The height, length, and width of the entire mattress / gabion are obtained by fitting the structural boundary line and combining it with camera intrinsic and extrinsic parameters or point cloud calibration conversion calculations. The mesh size of the mattress / gabion is calculated using the following formula: ； in, The three-dimensional coordinates of the intersection points of the skeleton lines. For the number of measurements; d mesh Mesh size; The specific method for extracting the binding spacing is as follows: After arranging the identified binding point coordinates in order, the spacing between adjacent binding points is calculated using the following formula: ； wherein, denotes the spacing between adjacent binding points, denotes the position of the th binding point, M is the total number of identified binding points.

5. The river channel grid laying area construction quality evaluation method according to claim 1, characterized in that, The default methods for building YOLOv8 models include: The target scene images containing the paving area are obtained as the training sample set, covering different viewpoints, lighting conditions and occlusion situations, and the training sample set is preprocessed. Input the prepared target sample set into the initialized YOLOv8 model to perform transfer learning training; After training, the model's performance is evaluated using an independent test sample set. Each image in the test set is input into the trained YOLOv8 model to perform inference. The model outputs prediction information for each detected tiled area, including the class confidence of its bounding box and precise pixel-level coordinates. Low-confidence predicted boxes and overlapping boxes are filtered out according to the preset confidence threshold and non-maximum suppression (NMS) threshold, retaining high-quality detection results. For images in the test set where the model successfully detected the paving area, the coordinate information of all predicted bounding boxes that meet the confidence requirements is extracted and saved. Establish a mapping relationship between image coordinates and real-world geographic coordinates.

6. The river channel grid laying area construction quality evaluation method according to claim 3, characterized in that, The step of spatially registering the point cloud data of the laying area with the design drawings includes: Feature points are extracted from the design drawings, including wall corners and pipe joints. These feature points are then matched with corresponding feature points in the point cloud. The Iterative Closest Point (ICP) algorithm is used to calculate the rotation matrix and translation vector to align the coordinate systems of the point cloud and the drawings. For areas with local deformation, the Thin Plate Spline Interpolation (TPS) algorithm is introduced to compensate for non-rigid deformation, ensuring that the geometric error between the point cloud and the drawings is less than a threshold.

7. A construction quality assessment system for river grid laying areas, characterized in that, include: The 3D modeling module is configured to: collect multi-directional and multi-angle geomorphic image data of the river grid area based on UAV oblique photography, and record the coordinate position information of the shooting point and the attitude information at the time of shooting to form the exterior orientation element data of the image. Based on the geomorphic image data and the exterior orientation element data, a 3D model of the river grid with real geographic coordinates is generated. The identification module is configured to: identify the laying area of ​​the river grid 3D model based on the preset YOLOv8 model, and connect the 3D geographic coordinates (X, Y, Z) of the corner points or key points of the laying area to form a polygon boundary describing the 3D spatial range of the laying area. The registration and segmentation module is configured to: spatially register the point cloud data of the paving area with the design drawings; after registration, set directional constraints based on the boundary parameters of the paving area in the design drawings to segment the point cloud, extracting a set of points in the point cloud that satisfy the plane equation, using the following formula: Where: n is the unit normal vector of the paved surface in the design drawing, p is the point cloud coordinates, and d is the plane constant. For tolerance, ∥...∥ indicates taking the absolute value; The quality feedback module is configured to: obtain the laying area, thickness and mattress / gabion size from the segmented point cloud; generate a structured report of the laying area, thickness and mattress / gabion size according to the coding rules of the design drawings; import the design value deviation of the laying area, thickness and material unfolding length into the BIM model; and visualize the out-of-tolerance area through color mapping. The laying thickness is obtained from the segmented point cloud, including: Based on the layer thickness parameters in the design drawings, a vertical profile line is generated along the direction of the normal vector of the segmented point cloud. On the profile line, the point cloud of the upper and lower surfaces is fitted using local weighted regression LOESS, and the Euclidean distance between the layers is calculated as the measured thickness value.

8. An electronic device, comprising: include: Memory, used to store one or more programs; processor; When the one or more programs are executed by the processor, the method as described in any one of claims 1-6 is implemented.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images