High-definition image rapid reconstruction method for concrete plastic hinge area identification

Abstract

The invention provides a high-definition image rapid reconstruction method for concrete plastic hinge area identification, and the method is especially used for carrying out damage evaluation on a key part, namely a plastic hinge area, of a beam-column joint in a reinforced concrete frame structure after an earthquake. Fusing the texture features of the dust deposition covering area, and recovering covered cross oblique crack details to obtain a complete crack form image set; and optimizing reconstruction of a three-dimensional space and generating a final damage model through point cloud data obtained through oblique photography of the unmanned aerial vehicle in combination with the updated stripping depth damage model, outputting visual three-dimensional reconstruction results of spatial distribution of cross oblique cracks, concrete stripping depth and a steel bar buckling form, and generating a plastic hinge region damage state image. The method provides powerful technical support for accurately evaluating the safety of the post-earthquake structure.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a high-definition image rapid reconstruction method for concrete plastic hinge area identification. Background Art

[0002] Safety assessment of reinforced concrete frame structures after earthquakes is a crucial topic in civil engineering, directly impacting the efficiency of post-disaster rescue and reconstruction. Beam-column joints are critical components of structural seismic resistance, and the damage morphology of their plastic hinge regions significantly impacts the overall load-bearing capacity. Accurately reconstructing high-definition images of joint damage is crucial for improving the reliability of assessments. However, existing image reconstruction methods perform poorly in complex post-earthquake environments. Image sequences acquired by traditional drones or ground-based equipment often suffer from obstructions such as falling objects and dust deposition, making it difficult to identify microcracks and crushing spalling boundaries. Furthermore, these methods often rely on a single data source and lack comprehensive analysis of the reinforcement morphology and spatial distribution of cracks within the joints, making them difficult to meet the requirements of high-precision visualization. The complexity of post-earthquake scenarios further exacerbates technical challenges. Irregular obstructions in the core joint area due to falling objects obscure key damage features; dust deposition obscures early microcracks, complicating damage identification. Furthermore, single-view image sequences struggle to capture the full three-dimensional distribution of reinforcement buckling and cracks within the joints. These factors combine to blur the boundaries of damage and obstructions, making it difficult to quantify concrete spalling depth and reinforcement deformation morphology. How to effectively integrate drone oblique photography with incomplete image sequences from ground-based handheld devices, fully utilize node design reinforcement drawings as prior knowledge, accurately distinguish between obstructions and true damage boundaries, and achieve visual reconstruction of the spalling depth of the crushed zone, the buckling morphology of the steel bars, and the three-dimensional distribution of the main cracks, has become a key issue that needs to be solved urgently. Summary of the Invention

[0003] The present invention provides a high-definition image rapid reconstruction method for concrete plastic hinge area identification, which mainly includes:

[0004] Extract an image set of intersecting oblique cracks and concrete crush areas related to the plastic hinge properties of beam-column joints from multi-view image sequences acquired through drone oblique photography and ground-based handheld devices. If the image is incomplete due to obstruction by falling objects or dust, the concrete damage area is determined and a set of images without obstructions is generated.

[0005] Based on the image set with occluded objects removed, the texture features of the dust deposition masked area are integrated to restore the details of the masked cross-slant cracks and obtain a complete crack morphology image set;

[0006] The spatial distribution characteristics of cross-slant cracks and main cracks are extracted from the complete crack morphology image set. Combined with the steel bar layout information of the node reinforcement drawings, a three-dimensional crack model is constructed to obtain the spatial distribution parameters of the cracks.

[0007] Based on the spatial distribution parameters of cracks, the grayscale gradient characteristics of the concrete crush area are introduced to calculate the spalling depth of the crushed area under the influence of occlusion and dust. The spalling depth damage model is generated. The clarity and continuity of the three-dimensional contours of the steel bars in the spalling depth damage model are analyzed to determine whether the buckling morphology of the steel bars in the spalling depth damage model is clear.

[0008] If the buckling morphology of the steel bars in the spalling depth damage model is unclear, the buckling morphology parameters are derived by combining the steel bar diameter and spacing constraint information from the node reinforcement drawings and the morphology of the exposed steel bars observed in the occluded image set to obtain an updated spalling depth damage model.

[0009] If the buckling morphology of the steel bars in the spalling depth damage model is clear, the buckling morphology parameters of the steel bars in the spalling depth damage model are used to perform damage assessment or structural response analysis;

[0010] The final damage model was generated by optimizing the 3D space reconstruction using point cloud data from drone oblique photography combined with the updated spalling depth damage model.

[0011] Based on the final damage model, the spatial distribution of cross-diagonal cracks, concrete spalling depth, and reinforcement buckling morphology are visualized and reconstructed in three dimensions, generating an image of the damage state in the plastic hinge area.

[0012] Furthermore, a set of images of intersecting oblique cracks and concrete crush areas associated with the plastic hinge properties of beam-column joints is extracted from multi-view image sequences acquired via drone oblique photography and ground-based handheld devices. If the images are incomplete due to obstruction by falling objects or dust, the concrete damage area is determined and a set of images excluding the obstructions is generated. This process involves extracting feature points from the multi-view image sequences acquired by the drone and handheld device based on photogrammetric adjustment principles, obtaining feature descriptors using a scale-invariant feature transformation algorithm, obtaining a set of corresponding points between images using a nearest neighbor matching algorithm, and calculating the spatial transformation matrix of the multi-view image sequence. The multi-view image sequences are then registered based on the spatial transformation matrix, and a three-dimensional point cloud model of the joint surface is generated using a structured light stereo reconstruction algorithm. Missing areas in the point cloud are identified using a density threshold segmentation method, and the outlines of obstructions and dust obstruction ranges are marked. The joint reinforcement drawings are vectorized, and the rebar position coordinates and diameter information are extracted. A three-dimensional reinforcement geometry model is constructed, and the reinforcement geometry model is aligned with the point cloud model using a rigid body transformation algorithm to obtain the actual rebar distribution at the joint. A mesh was created on the surface of the 3D point cloud model to extract the crack edge contours, calculate the crack width and orientation angle, and determine the extent of the plastic hinge region based on the crack distribution density and width threshold. The concrete crush depth distribution was calculated based on the local curvature changes of the point cloud model. Combined with the actual distribution of steel bars, a damage feature vector was constructed that included both crack and crush characteristics. A random forest regressor was used to predict the extent of damage. Density clustering analysis was performed on the damage feature vector to delineate the boundaries of the damaged region. A convolutional neural network was then used to classify the clustered damaged regions, outputting the location of the plastic hinge damage, the damage extent, and the extent of the concrete crush.

[0013] Furthermore, based on the image set after removing obstructions, the texture features of the dust-obscured areas are integrated to restore the details of the obscured cross-slant cracks, resulting in a complete set of crack morphology images. This involves calculating three characteristic parameters: local texture contrast, uniformity, and entropy, based on the image grayscale distribution. The boundaries of the dust-obscured areas are identified using the maximum entropy threshold segmentation method to obtain an initial crack-obscured area map. A local texture feature matrix is ​​extracted from the initial crack-obscured area map, and the texture block grayscale co-occurrence matrix is ​​calculated. Three texture description parameters, angular second moment, correlation, and energy, are obtained to construct a texture feature vector. A bilateral filtering algorithm is used to enhance the texture feature vector. The optimal enhancement coefficient is determined through grayscale gradient consistency evaluation, and adaptive texture enhancement is performed on the dust-obscured areas. A crack edge response map is calculated based on the enhanced texture features. The locations of crack intersections are extracted using the Harris corner detection algorithm, and a crack direction field description is established. The crack edge response map is then decomposed at multiple scales using a wavelet transform to extract high-frequency coefficients in the horizontal, vertical, and diagonal directions, reconstructing the crack edge contour. The reconstructed crack edge contour is morphologically refined, and the crack connectivity is judged using a random forest classifier. Multi-scale features are fused to obtain a complete crack morphology image set.

[0014] Furthermore, the spatial distribution features of cross-diagonal cracks and main cracks were extracted from the complete crack morphology image set. Combined with the reinforcement layout information from the node reinforcement drawings, a three-dimensional crack model was constructed to obtain the spatial distribution parameters of the cracks. This process involved extracting the crack edge contours using the Sobel edge detection operator, calculating the numerical parameters of the crack length and width, and obtaining the crack centerline using a morphological thinning algorithm. This led to the establishment of a dataset of edge features for cross-diagonal cracks and main cracks. The spatial layout of the reinforcement was extracted from the node reinforcement drawings, and a rectangular coordinate matrix of longitudinal and stirrup positions was established. The reinforcement geometric parameters were calculated, and a three-dimensional spatial reference frame for the reinforcement was generated. Directional field analysis was performed on the crack edge feature dataset to extract the crack strike angle. The spatial position of the cracks was calibrated using the three-dimensional reinforcement reference frame, and the crack depth distribution was calculated using a random forest regressor. The crack edge contours were spatially mapped using polar coordinate transformation, and the crack projection trajectory in three-dimensional space was calculated to construct a crack depth distribution surface. A three-dimensional mesh was constructed based on the crack depth distribution surface. The spatial coordinates of the cracks were calculated using a spatial interpolation algorithm, and the crack inclination parameters were obtained using the least squares fitting method. The three-dimensional spatial reference frame of steel bars is used to correct the spatial distribution of cracks, and the spatial parameters of cross-oblique cracks and main cracks are integrated to generate a complete three-dimensional crack spatial distribution model.

[0015] Furthermore, the spatial distribution parameters of cracks are analyzed by incorporating grayscale gradient features of the concrete crush region. The spalling depth of the crush region under occlusion and dust influences is calculated, generating a spalling depth damage model. The clarity and continuity of the three-dimensional contours of the steel bars in the model are analyzed to determine whether the buckling morphology of the steel bars in the spalling depth damage model is clear. This involves calculating the gradient amplitude distribution of the crush region based on the grayscale image of the concrete crush region, segmenting the dust-occluded region using a regional adaptive thresholding method, and highlighting the edge features of the crush region using a local contrast enhancement algorithm to obtain an initial crush region contour map. Edge detection is performed on the initial crush region contour map, and the gradient direction field and edge strength values ​​are calculated. A dual-threshold segmentation method is used to construct an edge connectivity map of the crush region, marking the complete boundary of the crush region. A random forest regressor is used to calculate the crush region depth distribution. Input features include gradient amplitude, direction field, and edge strength values. A crush depth prediction function is established to generate a three-dimensional morphological map of the crush region. The exposed steel bars are extracted from the three-dimensional morphological map, the grayscale gradient of the steel bar edges is calculated, and the spatial coordinates of the steel bar axes are fitted using the least squares method to obtain the spatial contour of the steel bar. The curvature of the rebar's spatial profile is calculated, and a threshold criterion for rebar deformation curvature is established. Connected domain analysis is used to assess rebar profile integrity and identify discontinuities in rebar continuity. A rebar buckling deformation field is constructed based on these discontinuities, and a morphological reconstruction algorithm is used to complete missing sections of the rebar profile, generating a complete spalling depth damage model.

[0016] Furthermore, if the buckling morphology of the steel bars in the spalling depth damage model is unclear, the buckling morphology parameters are derived by combining the steel bar diameter and spacing constraint information from the node reinforcement drawings and using the morphology of the exposed steel bar area observed in the image set after removing obstructions. This results in an updated spalling depth damage model. This includes extracting the spatial layout information of the steel bars from the node reinforcement drawings, establishing a three-dimensional coordinate reference system, obtaining the diameter and spacing parameters of the longitudinal bars and stirrups through contour vectorization, and constructing a standard steel bar layout spatial position diagram. A three-dimensional point cloud dataset is established for the exposed steel bar area. The surface point set of the steel bars is marked using a region growing algorithm. The steel bar centerline is fitted using the least squares method to obtain the axis equation of the exposed steel bar area. The curvature distribution of the exposed area is calculated based on the axis equation of the steel bar. The curvature threshold segmentation method is used to identify the key points of steel bar deformation and obtain the steel bar deformation curvature and inclination parameters. Buckling deformation characteristics are extracted based on the key points of steel bar deformation. The buckling wavelength and amplitude are calculated through Fourier series decomposition to establish a function describing the steel bar buckling deformation. The deformation of the obscured area is predicted using a rebar buckling function, and the rebar buckling extension trajectory is calculated using a random forest regressor. Constrained optimization of the buckling extension trajectory is performed using a standard rebar layout spatial position diagram. A complete rebar spatial curve is generated using a three-dimensional surface interpolation algorithm, and the spalling depth distribution data is integrated to update the damage model.

[0017] Furthermore, if the buckling morphology of the reinforcement in the spalling depth damage model is clear, the reinforcement buckling morphology parameters in the spalling depth damage model are used to perform damage assessment or structural response analysis. This involves obtaining the reinforcement spatial curve equation from the spalling depth damage model, calculating the coordinate sequence of discrete points on the curve, calculating the reinforcement deformation field distribution through curvature calculation, and establishing a strain distribution function based on the material constitutive relationship. The reinforcement yield strain ratio is calculated based on the strain distribution function. The node stiffness matrix is ​​constructed using a hierarchical numerical integration method, and the material strength loss distribution is determined using the strain energy density function. Based on the material strength loss distribution, a node damage variable field is established, and a random forest algorithm is used to calculate the damage severity index and determine the spatial distribution of damage in the node region. Based on the spatial distribution of damage in the node region, a nonlinear equilibrium equation is constructed. The internal force redistribution process is solved using the Newton iteration method, and the node deformation energy is calculated based on the virtual displacement principle. A generalized stiffness loss function is established using the node deformation energy, and the bearing capacity attenuation curve is calculated through piecewise linear interpolation to obtain the node bearing capacity assessment index. Based on the bearing capacity assessment index, a neural network prediction model is constructed, the structural response evolution parameters are calculated, and the damaged region is spatially partitioned using a hierarchical clustering method to generate a structural damage distribution map.

[0018] Furthermore, using point cloud data from drone oblique photography and an updated spalling depth damage model, the reconstruction of three-dimensional space was optimized to generate the final damage model. This process involved establishing a spatial reference system based on the point cloud data acquired from drone oblique photography, resampling the point cloud using voxel meshing, and extracting the point cloud surface normal and curvature distribution using a local feature description algorithm. Boundaries were extracted based on the point cloud surface features, and a mesh reconstruction algorithm was used to generate the initial damaged region surface. The volume and surface area parameters of the crushed region were calculated using discrete surface integrals. The damaged region surface was spatially layered and sampled to establish a surface feature vector field. A random forest regressor was used to predict crack propagation paths and construct a crack spatial topology map. The coordinates of the damaged region's centroid were calculated based on the crack spatial topology map. The connection between cross-oblique cracks and the main crack was determined using a region growing algorithm, generating a crack network structure map. An adaptive mesh subdivision algorithm was used to spatially segment the damaged region, and a neural network segmenter was used to identify the buckling morphology of the steel bars and the crushed concrete boundary. A minimum spanning tree algorithm was used to establish the internal connectivity of the damaged region, and a topological distance matrix between nodes was calculated to construct a hierarchical optimized damage model.

[0019] Furthermore, based on the final damage model, a visual 3D reconstruction of the spatial distribution of intersecting diagonal cracks, concrete spalling depth, and rebar buckling morphology is output to generate a damage state image of the plastic hinge region. This process involves extracting the spatial coordinate sequence of the intersecting diagonal cracks based on the optimized damage model, generating a crack mesh model using a 3D mesh reconstruction algorithm, and calculating a continuous crack surface using cubic spline interpolation. The continuous crack surface is meshed, the depth of the concrete spalling region is extracted, and a color mapping rule is established based on the depth values ​​to generate a spalling region surface with depth information. The plastic hinge damage boundary is constructed based on the spalling region surface, and the boundary transition region is optimized using a surface smoothing algorithm. A texture fusion method is used to generate a realistic texture of the damage region. A 3D skeleton is established based on the rebar buckling deformation, and a random forest regressor is used to fit the rebar centerline. A solid rebar model is generated through cylindrical scanning. The solid rebar model and the damage region are spatially combined, and a layer overlay relationship is established. Regions of varying depths are marked with gradient colors. Material properties are constructed based on the layer overlay relationship, and ambient lighting parameters are set. Finally, a ray tracing algorithm is used to generate a damage state image of the plastic hinge region.

[0020] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0021] The present invention discloses a high-definition image rapid reconstruction method for concrete plastic hinge area identification. The method acquires multi-view images through oblique photography by unmanned aerial vehicles and ground handheld devices, and extracts information on cross-oblique cracks and concrete crush areas in combination with node reinforcement drawings. In order to eliminate image incompleteness caused by falling objects or dust, three-dimensional reconstruction and deep learning segmentation algorithms are used to eliminate interference. The details of the masked cracks are restored through image enhancement, and a three-dimensional crack model is constructed. The grayscale gradient characteristics of the concrete crush area are introduced to calculate the spalling depth and generate a damage model. The damage model is optimized in combination with the analysis of the steel bar buckling morphology. Finally, the visualized three-dimensional reconstruction results of the spatial distribution of cross-oblique cracks, the concrete spalling depth and the steel bar buckling morphology are output, providing an effective method for evaluating the true damage state of the plastic hinge area of ​​the beam-column node. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a high-definition image rapid reconstruction method for concrete plastic hinge area identification according to the present invention. DETAILED DESCRIPTION

[0023] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0024] like Figure 1 A high-definition image fast reconstruction method for concrete plastic hinge area identification includes:

[0025] S101. Based on the multi-view image sequences collected by drone oblique photography and ground-based handheld devices, a set of damage feature images related to the plastic hinge of the beam-column node is extracted. Combined with the geometric constraints of the node reinforcement drawings, the damage area data after interference removal is generated.

[0026] In an embodiment of the present invention, the reconstruction function can be integrated into dedicated image processing software or an embedded system, and triggered by the user through interface operation or remote command. The specific triggering method can be determined according to the actual scenario, such as by clicking a software interface icon or receiving a command from an external device.

[0027] S1011. Extract feature descriptors based on the multi-view image sequence, obtain corresponding point sets through a matching algorithm and calculate the spatial transformation matrix, align the image sequence to generate a point cloud model, and identify missing areas in the point cloud.

[0028] In an embodiment of the present invention, a scale-invariant feature transformation algorithm is used to extract feature descriptors from a multi-view image sequence. A nearest neighbor matching algorithm is used to filter the corresponding point sets between the images to generate a spatial transformation matrix containing rotation, translation, and scale parameters. Based on this matrix, the image sequence is registered, and a structured light stereo reconstruction algorithm is used to generate a three-dimensional point cloud model of the node surface. The point cloud model is analyzed using a density threshold segmentation method. Areas with a marker point density below a preset threshold are identified as missing areas. The threshold is typically set at 10 points per square centimeter to identify areas obstructed by falling objects or affected by dust.

[0029] S1012. Vectorize the node reinforcement drawings, build a steel bar geometry model and align it with the point cloud model, obtain the actual steel bar distribution data, and extract the crack contour and crush characteristics by combining the point cloud surface meshing.

[0030] In this embodiment of the present invention, a vectorization algorithm is used to analyze the reinforcement drawings, extract the position coordinates, diameter, and spacing information of the steel bars, and construct a three-dimensional steel bar geometric model. A rigid body transformation algorithm is used to align the steel bar geometric model with the point cloud model, ensuring that the steel bar distribution is consistent with the actual node structure. A grid is created on the surface of the point cloud model with a grid size of 2 mm x 2 mm. A region growing algorithm is used to extract crack contours and calculate crack width and inclination characteristics. The concrete crush depth is calculated based on the local curvature changes of the point cloud, generating a damage feature vector containing crack and crush characteristics.

[0031] S1013. If the image sequence is incomplete due to interference from falling objects or dust, a deep learning segmentation algorithm is used to separate the interference factors, generate a set of damaged images after removing occlusions, and classify the damaged areas through a convolutional neural network.

[0032] In this embodiment of the present invention, a deep learning-based segmentation algorithm is used to identify the outline of the falling object and the dust-covered area in case of image incompleteness. Pixel-level classification is then used to eliminate interference and generate a clear set of damaged images. Density clustering is performed based on the damage feature vectors to delineate the boundaries of the damaged areas. A convolutional neural network is then used to classify the damaged areas, outputting data on the location and severity of damage in the plastic hinge area, for example, determining whether the damage is mild, moderate, or severe.

[0033] In this embodiment of the present invention, by fusing multi-view image sequences with reinforcement drawings and incorporating deep learning segmentation technology, interference from falling objects and dust can be effectively eliminated, allowing for precise extraction of crack and crush characteristics. Aligning the point cloud model with the reinforcement geometry ensures geometric consistency in the reconstructed results, providing a reliable foundation for subsequent crack and spalling depth modeling. The resulting damage region data clearly demonstrates the damage distribution within the plastic hinge region, significantly improving reconstruction accuracy and assessment reliability.

[0034] It is understood that the above embodiments are only specific implementations of the present invention, and those skilled in the art may appropriately expand or adjust the present invention within the scope of the technical solution. The scope of protection of the present invention shall be subject to the claims.

[0035] S102. Based on the damaged image set after removing the obstructions, feature extraction is performed based on the texture characteristics of the dust deposition area, and image enhancement technology is used to restore the details of the obscured cross-slant cracks to generate a complete crack morphology image set.

[0036] In the embodiment of the present invention, for the damaged image set after removing the obstructions, the focus is on dealing with the problem of dust deposition masking the crack features. Through texture characteristic analysis and enhancement algorithm, the detailed information of tiny cracks is restored to ensure the integrity of the crack morphology.

[0037] S1021. Calculate the texture characteristic parameters of the local area based on the damage image set, use the maximum entropy threshold segmentation method to identify the dust deposition area, and generate an initial crack occlusion area map.

[0038] In an embodiment of the present invention, the grayscale value distribution of a local area is extracted from a set of damaged images, and characteristic parameters such as texture contrast, uniformity, and entropy are calculated to quantify the surface damage state. For example, for a local area of ​​16×16 pixels, through grayscale histogram analysis, when the contrast is greater than 0.75, the uniformity is less than 0.45, or the entropy value is greater than 5.0, it is determined that the area may have crack features. The maximum entropy threshold segmentation algorithm is adopted to determine the optimal segmentation threshold based on the entropy value distribution of the image grayscale histogram to identify the boundary of the dust deposition area. Generally, when the dust coverage thickness is about 0.4 mm, the grayscale value range is between 130 and 170. The segmentation threshold can be automatically determined by maximizing the entropy value to generate an initial crack occlusion area map, which provides a basis for subsequent texture enhancement.

[0039] S1022. Extract the gray-level co-occurrence matrix from the initial crack occlusion area map, construct a texture feature vector, and perform adaptive enhancement through a bilateral filtering algorithm to obtain enhanced texture characteristics.

[0040] In an embodiment of the present invention, for the initial crack occlusion area map, the grayscale co-occurrence matrix of the local area is calculated to capture the spatial correlation between pixels. For an 8-bit grayscale image, a 256×256-dimensional co-occurrence matrix is ​​generated, and descriptive parameters such as the angular second moment, correlation, and inverse moment are extracted to construct a texture feature vector. For example, when the angular second moment value is between 0.35 and 0.65 and the correlation is between 0.55 and 0.85, it indicates that the texture feature has good discrimination. The texture feature vector is enhanced using a bilateral filtering algorithm, and the filter window is set to 7×7 pixels, the spatial domain standard deviation is 2.5, and the grayscale value domain standard deviation is 20, so as to smooth the dust area while retaining the crack edge information. The enhancement coefficient is determined by calculating the mean square error of the grayscale gradient before and after filtering. When the error is less than 0.9, the enhancement effect meets the requirements and the enhanced texture characteristic data is generated.

[0041] S1023. Generate a crack edge response map based on the enhanced texture characteristics, use corner detection and wavelet transform technology to extract crack intersections and edge contours, and reconstruct a complete crack morphology image.

[0042] In an embodiment of the present invention, the enhanced texture characteristics are used to calculate the crack edge response map, and the Sobel operator is used to extract the grayscale gradient direction to generate a preliminary edge intensity distribution. Based on the Harris corner detection algorithm, the response threshold is set to 0.015, the non-maximum suppression window is 5×5 pixels, and the crack intersection position is extracted. The intersection angle is usually between 45 degrees and 135 degrees, and the corner response value is about 30% higher than that of a single crack area. The db3 wavelet basis function is further used to perform a 4-layer wavelet transform to decompose the high-frequency coefficients in the horizontal, vertical and diagonal directions, and retain the coefficients with an amplitude greater than 45 to highlight the crack edge features. The decomposed edge response map is subjected to an inverse wavelet transform to reconstruct the crack contour.

[0043] S1024. Optimize the crack contour through morphological processing and classification algorithm, and fuse multi-scale features to generate a complete crack morphology image set.

[0044] In this embodiment of the present invention, the reconstructed crack contours are subjected to morphological refinement using an 8-neighborhood connectivity analysis with a maximum of 15 iterations to ensure contour continuity. A support vector machine classifier is used to determine crack connectivity and classify cracks based on width, length, and orientation, eliminating interference from non-structural cracks. By integrating the multi-scale features of a wavelet transform with information on crack intersections and edge contours, a complete set of crack morphological images is generated. The resulting crack images achieve a width accuracy of 0.08 mm and an angular error of less than 4 degrees, significantly improving crack identification accuracy.

[0045] In this embodiment of the present invention, by combining texture feature extraction with an enhancement algorithm, crack details obscured by dust deposition are effectively restored, resolving the difficulty in identifying tiny cracks in post-earthquake environments. Maximum entropy threshold segmentation and gray-level co-occurrence matrix analysis ensure accurate positioning of dust areas, while a bilateral filtering algorithm optimizes texture characteristics while preserving edge information. Wavelet transforms and morphological processing further enhance the clarity and connectivity of crack contours. The resulting set of crack morphology images provides high-quality data support for subsequent 3D modeling, significantly improving the reliability and visualization of damage reconstruction.

[0046] S103. Extract the spatial distribution characteristics of the cross-slant cracks and main cracks based on the complete crack morphology image set, and construct a three-dimensional crack model based on the steel bar layout information of the node reinforcement drawing to generate the spatial distribution parameters of the cracks.

[0047] In an embodiment of the present invention, for a complete set of crack morphology images, the spatial distribution characteristics of the cracks are extracted through edge detection and direction analysis. Combined with the steel bar geometry information provided by the reinforcement drawings, a high-precision three-dimensional crack model is constructed to provide reliable geometric parameters for damage assessment.

[0048] S1031. Use edge detection algorithm to extract edge contours of the crack image, calculate crack length and width parameters, and generate crack centerline data through morphological processing.

[0049] In an embodiment of the present invention, the Canny edge detection algorithm is used to process a complete set of crack morphology images, with high and low thresholds set to 100 and 50, respectively, to extract the edge contour lines of the cracks, ensuring that crack boundaries with significant grayscale gradient changes are accurately captured. For main cracks, the edge grayscale gradient value is usually between 60 and 90, while the gradient value of cross-oblique cracks is between 40 and 70. The length and width of the cracks are calculated using a contour tracking algorithm, with a length accuracy of 1 mm and a width accuracy of 0.1 mm. The edge contour is skeletonized using a morphological refinement algorithm, iterated until the contour width is a single pixel, and the crack centerline data is generated, preserving the topological structure of the cracks and providing a basis for subsequent spatial analysis.

[0050] S1032. Extract the reinforcement spatial arrangement information based on the node reinforcement drawings and construct a three-dimensional reinforcement reference frame for calibrating the spatial position of the cracks.

[0051] In an embodiment of the present invention, the geometric parameters of the steel bars, including the coordinates, diameters, and spacings of the longitudinal bars and stirrups, are parsed from the node reinforcement drawings. A rectangular coordinate system is established with the node center as the origin, and a three-dimensional steel bar position matrix is ​​generated. For example, the longitudinal bar diameter of a typical beam-column node is 22 mm, the stirrup diameter is 10 mm, and the stirrup spacing is 140 mm, forming a regular grid structure. The Euclidean transformation algorithm is used to map the steel bar position matrix to three-dimensional space, and a steel bar reference frame is constructed to ensure that the steel bar distribution is consistent with the actual structure. This framework provides a geometric benchmark for the calibration of the spatial position of the crack, reducing positioning errors.

[0052] S1033. Extract the crack strike angle through directional field analysis, calibrate the crack coordinates based on the steel bar reference frame, and use the regression model to predict the crack depth.

[0053] In an embodiment of the present invention, a directional field analysis is performed on the crack centerline data, and a Gabor filter is used to extract the local directional characteristics of the crack and calculate the crack strike angle. The strike angle of the main crack is usually 40 to 50 degrees with the force direction of the node, and the angle of the cross-oblique crack ranges from 50 to 130 degrees. Combined with the steel bar reference frame, the spatial position of the crack is calibrated by coordinate transformation to generate a three-dimensional coordinate set of the crack with a positioning accuracy of 0.8 mm. The gradient boosting tree regression model is used to predict the crack depth. The input features include crack width, strike angle and buried depth of adjacent steel bars. The depth value of the training sample ranges from 15 mm to 90 mm, and the prediction error is controlled within 2 mm.

[0054] S1034. Use spatial mapping and interpolation algorithms to construct a three-dimensional distribution model of cracks and generate spatial distribution parameters including coordinates, depth, and inclination.

[0055] In an embodiment of the present invention, the center line of the crack is mapped to three-dimensional space through cylindrical coordinate transformation to generate a projection trajectory of the crack to describe the curvature change of the crack. For cracks with a curvature radius of less than 450 mm, it is shown that there are significant bending features. Based on the projection trajectory, the Kriging interpolation algorithm is used to calculate the spatial coordinates and depth distribution of the crack to generate a three-dimensional grid model with a grid unit size of 4 mm × 4 mm. The weighted least squares method is used to fit the crack inclination, with an inclination range of 25 degrees to 65 degrees and a fitting error of less than 2.5 degrees. The spatial parameters of the cross-oblique cracks and the main cracks are integrated to generate a complete three-dimensional crack distribution model, which clearly presents the relative position of the cracks and the steel bars.

[0056] In this embodiment of the present invention, a combination of edge detection, directional analysis, and spatial interpolation techniques enables high-precision extraction of the spatial distribution characteristics of cracks. The introduction of a steel reference frame ensures geometric consistency in crack location, and the regression model effectively predicts crack depth. The resulting three-dimensional crack model achieves millimeter-level accuracy in coordinate accuracy, depth measurement, and inclination angle calculation. This provides high-quality data support for subsequent spalling depth analysis and structural assessment, improving the reliability and practicality of post-earthquake node damage reconstruction.

[0057] S104. Based on the spatial distribution parameters of cracks and the grayscale gradient characteristics of the concrete crush area, the spalling depth of the crushed area affected by occlusion and dust is calculated, and a spalling depth damage model that includes the buckling morphology of the steel bars is constructed.

[0058] In an embodiment of the present invention, the grayscale gradient characteristics of the crushed area are introduced based on the spatial distribution parameters of the cracks. Through edge detection, depth prediction and steel bar contour analysis, a high-precision spalling depth damage model is generated, which clearly displays the three-dimensional morphology of the crushed area and the deformation characteristics of the steel bars, providing a reliable basis for post-earthquake structural assessment.

[0059] S1041. Extract the gradient amplitude distribution of the crushed area from the grayscale image, and use adaptive threshold segmentation and contrast enhancement technology to generate an initial crushed area contour map.

[0060] In an embodiment of the present invention, the gradient amplitude distribution of the grayscale image is calculated by the Prewitt operator to quantify the degree of damage on the concrete surface. The gradient amplitude of the intact area is usually between 15 and 35, while the gradient amplitude of the crushed area can reach more than 90. In order to cope with the contrast drop caused by dust occlusion, an adaptive threshold segmentation algorithm is adopted to dynamically determine the segmentation threshold based on the grayscale mean and variance of the 8×8 pixel local window. When the local contrast is lower than 0.35, a contrast enhancement factor of 1.4 is applied to highlight the edge features of the crushed area and generate an initial contour map to ensure that the boundary clarity is improved by about 20%.

[0061] S1042: Perform edge detection on the initial crush area contour map, calculate the gradient direction field and edge strength, and construct a boundary marker map of the crush area.

[0062] In an embodiment of the present invention, the Canny edge detection algorithm is applied to the initial contour map, and the high threshold is set to the 80% quantile of the gradient amplitude, and the low threshold is set to 35% of the high threshold to extract the edge contour of the crushed area. The gradient direction field is calculated by the Sobel operator to evaluate the gradient direction consistency of adjacent pixels. When the direction difference is less than 25 degrees, it is determined to be the same edge. The edge intensity is based on the normalization of the gradient amplitude, and the area with an intensity value greater than 0.7 is marked as the crushed boundary. The edge tracking algorithm is used to generate a connectivity graph, and the areas with edge length greater than 40 pixels and curvature change less than 0.1 are screened to construct a complete boundary marking map to improve the robustness of boundary detection.

[0063] S1043. Use the random forest regression model to predict the depth of the crushed area, generate a three-dimensional morphological map, and extract the contour information of the exposed steel bar area.

[0064] In an embodiment of the present invention, a random forest regression model is constructed, and the input features include gradient amplitude, gradient direction consistency and edge strength to predict the spalling depth of the crushed area. The training data set covers samples with slight crushing of 5 to 12 mm, moderate crushing of 12 to 25 mm and severe crushing of more than 25 mm. The root mean square error of cross-validation is controlled within 2.5 mm. A three-dimensional morphological map is generated based on the predicted depth, using triangular mesh division with a mesh size of 4 mm and a depth accuracy of 1.8 mm. The exposed area of ​​the steel bar is extracted from the three-dimensional morphological map, and the grayscale gradient difference is used to locate the edge of the steel bar. The gradient value of the steel bar edge is usually between 110 and 160, which is significantly higher than the surrounding concrete area.

[0065] S1044. Fit the spatial contour of the steel bar using the least squares method, analyze the curvature and connectivity of the steel bar, complete the missing parts of the contour, and generate a complete damage model.

[0066] In an embodiment of the present invention, the least squares method is used to fit the edge points of the exposed area of ​​the steel bar to generate the spatial coordinates of the steel bar axis, and the fitting error is controlled within 1.5 mm. The local curvature of the steel bar axis is calculated. When the curvature exceeds 0.015 / mm, it is marked as a buckling point, reflecting the deformation characteristics of the steel bar. The integrity of the steel bar contour is evaluated by connected domain analysis, and the connected length threshold is set to 4 times the steel bar diameter. If the interval exceeds 1.2 times the steel bar diameter, it is determined to be a continuity interruption. For the interruption point, a morphological reconstruction algorithm is used, and the structural operator radius is set to 1.5 times the steel bar diameter. The number of iterations is controlled within 8 times to complete the missing contour. The final generated spalling depth damage model uses a triangular grid to represent the crushed area, the grid sampling spacing is 6 mm, the steel bar deformation is described by a spatial curve, the sampling point spacing is 8 mm, and the displacement accuracy reaches 0.9 mm, which fully presents the spatial relationship between crush damage and steel bar buckling.

[0067] In this embodiment of the present invention, the contours of the crushed area affected by dust were accurately extracted through a combination of grayscale gradient analysis and edge detection. A random forest regression model significantly improved the accuracy of depth prediction. The rebar contour fitting and morphological reconstruction effectively restored the buckling morphology. The resulting damage model not only reflects the three-dimensional characteristics of the crushed area but also clearly demonstrates the spatial distribution and deformation characteristics of the rebar, providing a highly accurate geometric basis for subsequent structural safety assessments.

[0068] S105. If the buckling morphology of the steel bars in the spalling depth damage model is unclear, the buckling morphology parameters are derived and the spalling depth damage model is updated by using the steel bar diameter and spacing constraints provided by the node reinforcement drawing and combining the steel bar exposed area features in the occluded image set.

[0069] In an embodiment of the present invention, for the case where the buckling morphology of the steel bars is unclear, the buckling parameters are derived through the geometric constraints of the reinforcement drawings and the morphological analysis of the exposed areas, and a complete description of the spatial deformation of the steel bars is generated to optimize the accuracy of the damage model and provide a reliable basis for the post-earthquake beam-column node assessment.

[0070] S1051. Extract the spatial layout information of steel bars based on the node reinforcement drawings, build a standard steel bar position model, and obtain the geometric parameters of the longitudinal bars and stirrups.

[0071] In an embodiment of the present invention, the geometric information of the steel bars is parsed from the node reinforcement drawings, including the diameter, spacing, and spatial coordinates of the longitudinal bars and stirrups. A three-dimensional orthogonal coordinate system is established with the node center as the origin, and a standard steel bar layout position model is generated. For example, the longitudinal bar diameter of a typical beam-column node is 20 mm, the stirrup diameter is 10 mm, and the stirrup spacing is 160 mm, forming a regular three-dimensional grid. An edge detection algorithm is used to vectorize the reinforcement drawings and extract the steel bar contours with an accuracy of 0.08 mm, ensuring that the steel bar position model is highly consistent with the actual structure and providing a geometric benchmark for subsequent buckling analysis.

[0072] S1052. Generate three-dimensional point cloud data from the exposed area, mark the steel bar surface point set and fit the steel bar center line, and calculate the curvature distribution of the exposed area.

[0073] In an embodiment of the present invention, based on an image set with occluded objects removed, three-dimensional point cloud data of the exposed area of ​​the steel bars is generated, and the point cloud density is required to be at least 40 points per square centimeter to ensure fitting accuracy. The regional growing algorithm is used to mark the surface point set of the steel bars, and the grayscale similarity threshold is set to 8, the spatial distance threshold is set to 1.5 mm, and the coverage of the point set is gradually expanded. The center line of the steel bar is fitted by the least squares method to generate the axis equation, and the residual is controlled within 0.9 mm. The local curvature of the center line is calculated using the sliding window method, and the window length is 4 times the diameter of the steel bar. When the curvature value exceeds 0.018 / mm, it is marked as a potential deformation point, reflecting the local deformation characteristics of the steel bar.

[0074] S1053. Identify the key points of steel bar deformation through curvature threshold segmentation, extract buckling parameters and construct a deformation description function.

[0075] In an embodiment of the present invention, based on the curvature distribution of the center line of the steel bar, a threshold segmentation method is used to identify the key points of deformation, and the curvature threshold is set to 0.015 / mm. The curvature of the steel bar that has not buckled is usually less than 0.004 / mm, while the buckling area can reach 0.04 / mm. The inclination changes between the key points are calculated. The inclination fluctuation of normal steel bars is less than 4 degrees, and the buckling area can reach 15 degrees. The periodic characteristics of the buckling are extracted using Fourier series decomposition, and the first 5 harmonic components are taken to calculate the buckling wavelength and amplitude. The wavelength is usually 12 to 20 times the diameter of the steel bar, and the amplitude is 1.5 to 3.5 times the diameter. Based on these parameters, a function describing the buckling deformation of the steel bar is constructed, and the deformation law is expressed by the Fourier coefficient matrix to provide a mathematical model for the prediction of the occluded area.

[0076] S1054. Use the regression model to predict the buckling trajectory of the steel bars in the obscured area, combine it with constraint optimization to generate the complete steel bar spatial curve, and update the damage model.

[0077] In an embodiment of the present invention, a support vector regression model is used to predict the buckling trajectory of the steel bars in the obscured area. The input features include the curvature, inclination, wavelength and amplitude of the exposed area. The training data covers a variety of buckling modes, and the displacement prediction error is controlled within 1.8 mm. Combined with the standard steel bar layout position model, material strength constraints are imposed, and the maximum allowable strain is set to 0.018. The cubic spline interpolation algorithm is used to generate a complete steel bar space curve. The interpolation node spacing is 1.8 times the steel bar diameter to ensure a smooth and continuous curve. The spalling depth distribution data is integrated, and the spalling depth damage model is updated. The sampling point spacing of the steel bar space curve is 4 mm to fully record the buckling deformation state.

[0078] In this embodiment of the present invention, the buckling parameters of the reinforcement bars are precisely derived by combining the geometric constraints of the reinforcement drawings with point cloud analysis of the exposed areas. The region growing algorithm and the least squares method ensure the accuracy of centerline fitting, and Fourier decomposition effectively extracts the periodic characteristics of the buckling. The application of regression prediction and spline interpolation techniques enables reliable extension of deformation in obscured areas. The updated damage model clearly demonstrates the relationship between the spatial deformation of the reinforcement bars and the spalling depth, supporting the high-precision requirements of structural damage assessment.

[0079] S106. If the buckling morphology of the steel bars in the spalling depth damage model is clear, the steel bar buckling morphology parameters are used to perform damage assessment and structural response analysis to generate a node damage distribution map.

[0080] In an embodiment of the present invention, based on the clear steel bar buckling morphology in the spalling depth damage model, spatial curve characteristics are extracted, strain distribution and stiffness loss are calculated, and machine learning and clustering algorithms are combined to generate a high-precision damage distribution map, providing a comprehensive basis for the safety assessment of beam-column nodes after earthquakes.

[0081] S1061. Extract the reinforcement spatial curve from the spalling depth damage model, calculate the deformation field distribution and construct the strain distribution function to generate the node stiffness matrix.

[0082] In an embodiment of the present invention, the spatial curve equation of the steel bar is obtained from the spalling depth damage model, and a discrete point coordinate sequence is generated by uniform sampling. The sampling interval is set to 4 mm to ensure the accuracy of the deformation field calculation. The curvature distribution is calculated by the difference method. The curvature value in the buckling area is usually above 0.04 / mm, which is much higher than 0.003 / mm in the normal area. Based on the constitutive relationship of the steel bar material, a strain distribution function is constructed to reflect the mechanical state of the steel bar. When the local strain exceeds 0.018, the steel bar enters the plastic stage, indicating that the damage is significant. The node stiffness matrix is ​​constructed by the layered numerical integration method. The number of integration layers is set to 18 layers, and the thickness of each layer is 0.2 times the diameter of the steel bar. The node stiffness is calculated by matrix assembly, and the accuracy is improved by about 15% compared with the traditional method.

[0083] S1062. Calculate material strength loss based on the stiffness matrix, construct a damage variable field, and use the random forest algorithm to evaluate the damage degree.

[0084] In an embodiment of the present invention, the strain energy density distribution is calculated using the node stiffness matrix to quantify the material strength loss. The strain energy density in the buckling area can reach 4.5 times that of the undamaged state, and the strength loss rate is between 35% and 45%. Based on the strength loss distribution, a node damage variable field is constructed to reflect the spatial heterogeneity of the damage. The random forest algorithm is used to calculate the damage degree index. The input features include strain distribution, curvature change and strength loss rate. The output damage level is divided into mild, moderate and severe, and the corresponding index values ​​are 0.15, 0.45 and 0.75 respectively. The algorithm is integrated through 100 decision trees, and the cross-validation error is controlled within 5% to ensure the robustness of the evaluation results.

[0085] S1063. Analyze the internal force redistribution through nonlinear equilibrium equations, calculate the node deformation energy and bearing capacity attenuation, and construct a generalized stiffness loss function.

[0086] In an embodiment of the present invention, a nonlinear equilibrium equation is constructed based on the damage variable field, and the Newton-Raphson iteration method is used to solve the internal force redistribution, and the convergence error is controlled within 0.08%. The bearing capacity of the steel bars in the buckling area decreases, resulting in the transfer of stress to the surrounding undamaged areas, and the amplitude of internal force redistribution can reach 30% of the initial state. The node deformation energy is calculated by the virtual displacement principle. When the deformation energy exceeds the critical value, it indicates that the structure is close to instability. Based on the deformation energy distribution, a generalized stiffness loss function is constructed using piecewise linear interpolation. The interpolation node interval is 0.08 of the damage index to generate a bearing capacity attenuation curve. When the damage index reaches 0.7, the node bearing capacity drops to 65% of the initial value, which clearly reflects the degradation of structural performance.

[0087] In an embodiment of the present invention, a neural network prediction model is used to analyze the evolution of structural response, and the damage area is divided by hierarchical clustering. A three-layer neural network is constructed with 72 hidden neurons. The training data includes damage indicators, strain distribution and stiffness loss, and the prediction accuracy reaches 92%. An adaptive hierarchical clustering method is used to dynamically determine the number of clusters according to the degree of damage, and usually 3 to 5 damage areas are generated. The final structural damage distribution map is presented in color coding, with red marking severe damage areas, yellow indicating moderate damage, and green indicating slight damage. It intuitively displays the damage concentration characteristics of the plastic hinge area, providing a high-precision and visual basis for structural safety assessment.

[0088] S107. Utilize the point cloud data obtained from drone oblique photography and combine it with the updated spalling depth damage model to optimize the three-dimensional spatial reconstruction, generate the final damage model including cross-oblique cracks, main crack distribution, spalling depth, and reinforcement buckling morphology, and refine the internal topological relationship.

[0089] In an embodiment of the present invention, by fusing high-density point cloud data with the spalling depth damage model, the three-dimensional reconstruction accuracy is optimized, the geometric and topological features are extracted, and a comprehensive damage model is generated, providing a high-precision and visual basis for the safety assessment of beam-column nodes after an earthquake.

[0090] S1071. Resample the point cloud data through voxel grid division, extract the surface normal vector and curvature distribution, and generate an initial damage area surface model.

[0091] In an embodiment of the present invention, based on the point cloud data obtained by oblique photography of a drone, a local coordinate system with the node center as the origin is established, and the point cloud data is aligned by coordinate transformation. The voxel grid division method is used for resampling, and the voxel size is set to 4 mm to ensure that the point cloud density is uniform and the computing efficiency is improved by about 20%. The principal component analysis algorithm is used to calculate the surface normal vector of the local point cloud. For the edge of the damaged area, the normal vector change gradient can reach more than 0.15. The Gaussian curvature estimation method is used to calculate the point cloud curvature distribution. The curvature value of the damaged area is usually above 0.08, which is significantly higher than 0.01 in the flat area. Based on the normal vector and curvature data, the Poisson surface reconstruction algorithm is applied to generate the initial damaged area surface model, which is represented by a triangular mesh with a mesh size of 1.5 mm. The volume and surface area of ​​the crushed area are calculated by discrete surface integral. The typical crushed area volume is between 150 and 450 cubic centimeters, and the surface area is between 350 and 750 square centimeters, with an accuracy of 1 cubic millimeter.

[0092] S1072. Perform spatial layered sampling on the initial surface model, construct a surface feature vector field, use a regression model to predict the crack extension path, and generate a crack space topology map.

[0093] In an embodiment of the present invention, the initial surface model is spatially layered sampled with a sampling interval of 2 mm to generate a multi-layer depth profile. Based on a local window of 16×16 pixels, the curvature, normal vector and spalling depth features are extracted to construct a high-dimensional surface feature vector field with a feature dimension of 96 dimensions. A gradient boosting tree regression model is used to predict the crack extension path. The input features include surface feature vectors and geometric parameters of adjacent cracks. The training samples cover a variety of crack morphologies, and the prediction error is controlled within 0.7 mm. Based on the predicted path, a crack space topology map is constructed and represented by a weighted adjacency matrix. The matrix edge weights are determined by the crack width and intersection angle. The nodes correspond to crack intersections, and the average number of nodes is 50 to 100. The topology map clearly reflects the connectivity and spatial distribution of the cracks, providing a basis for the subsequent division of the damage area.

[0094] S1073. Identify the crack connection relationship through the region growing algorithm, construct the crack network structure, and calculate the centroid coordinates of the damaged area.

[0095] In an embodiment of the present invention, a regional growing algorithm is used to analyze the spatial topology of the cracks, and the growth threshold is set to 1.2 times the crack width. When the distance between the endpoints of two cracks is less than the threshold, it is determined to be connected, and the recognition accuracy rate reaches 96%. Based on the connectivity relationship, a crack network structure diagram is generated to distinguish between cross-oblique cracks and main cracks. The cross angle is usually between 50 and 130 degrees, and the main crack extension length is between 180 and 480 mm. The center of mass coordinates of the damaged area are calculated by the weighted integral method. The center of mass position reflects the spatial distribution characteristics of the crushed area. For irregular crushed areas, the center of mass offset can reach more than 10 mm from the node center. The network structure and center of mass data jointly optimize the geometric expression ability of the damage model.

[0096] In an embodiment of the present invention, adaptive mesh subdivision and neural network segmentation technology are used to further optimize the topological structure of the damaged area. An adaptive mesh subdivision algorithm is applied to the crack network structure diagram, and the mesh density is dynamically adjusted according to the local curvature. The mesh size in the area with a curvature greater than 0.1 is refined to 0.4 mm, and the flat area is kept at 4 mm. A deep convolutional neural network segmenter is used to input the depth map, normal vector map and curvature map to identify the buckling morphology of the steel bar and the concrete crush boundary, with a segmentation accuracy of 93%. The internal connection relationship of the damaged area is constructed by the Kruskal minimum spanning tree algorithm, and the edge weight integrates the Euclidean distance and the damage degree to generate a hierarchical topological distance matrix. The matrix eigenvalue analysis reveals the main damage mode, and the maximum eigenvalue corresponds to the concentrated damage in the plastic hinge area. The final damage model describes the buckling morphology of the steel bar with a piecewise cubic spline function, a sampling spacing of 8 mm, a fitting accuracy better than 0.4 mm, and a spatial accuracy of 0.8 mm. It fully presents the spatial relationship between crack distribution, spalling depth and steel bar deformation, providing a high-precision basis for structural damage assessment.

[0097] S108. Based on the optimized final damage model, a three-dimensional visualization reconstruction result including the spatial distribution of cross-oblique cracks, concrete spalling depth, and steel bar buckling morphology is generated, and an image reflecting the actual damage state of the plastic hinge area of ​​the beam-column node is output.

[0098] In an embodiment of the present invention, the optimized final damage model is used to generate intuitive three-dimensional damage presentation images through high-precision mesh reconstruction, depth mapping and ray tracing technology, clearly showing the spatial characteristics of cracks, spalling depth and steel deformation, providing high-precision visualization support for post-earthquake structural assessment.

[0099] S1081. Extract the spatial coordinate sequence of the cross-oblique cracks, use mesh reconstruction and spline interpolation to generate a continuous surface of the cracks, and construct the surface of the spalling area.

[0100] In an embodiment of the present invention, the spatial coordinate sequence of the cross-oblique cracks is extracted from the final damage model, and the sampling point spacing is set to 1.5 mm to ensure contour accuracy. The crack mesh model is generated by the Delaunay triangulation algorithm, and the mesh unit size is 0.8 mm to ensure the integrity of the geometric details. The continuous surface of the crack is calculated using cubic B-spline interpolation, and the fitting error is controlled within 0.4 mm to form a smooth crack surface. The surface is adaptively meshed and encrypted, and the depth value is extracted for the spalling area with a depth range of 0 to 45 mm. A color mapping rule is constructed based on the depth value, using the HSV color space. Shallow spalling of 0 to 15 mm is mapped to blue, the middle layer of 15 to 30 mm is green, and the deep layer of 30 mm or more is red. The color transition is achieved through quadratic interpolation to achieve smooth changes, generating a spalling area surface with depth information, which improves the visualization intuitiveness by about 25%.

[0101] S1082. Construct a plastic hinge damage boundary based on the surface of the peeling area, optimize the boundary texture and generate a realistic surface effect.

[0102] In an embodiment of the present invention, the surface data of the spalling area is used to construct the plastic hinge damage boundary using a surface segmentation algorithm. The segmentation threshold is based on the depth gradient, and the area with a gradient greater than 0.2 is marked as the boundary. The Laplace smoothing algorithm is applied to optimize the boundary transition, and the kernel radius is set to 4 mm to eliminate the visual discontinuity caused by the depth mutation. Multi-layer texture fusion technology is used to generate a realistic surface: the bottom layer is a high-resolution concrete texture with a pixel density of 20 pixels per square millimeter; the middle layer simulates the concave and convex effects of cracks and spalling through normal vector perturbation; the top layer adds ambient light occlusion to enhance the three-dimensional effect, and the occlusion strength is 0.3. The fused texture significantly improves the visual realism of the damaged area, and the detail expression is improved by about 30%.

[0103] In this embodiment of the present invention, a 3D skeleton line is generated for the buckling deformation of the rebar and a solid model is constructed to complete the final rendering of the damage image. The 3D skeleton line of the rebar buckling is extracted from the final damage model, and the centerline is fitted using a support vector regression model. Input features include local curvature, displacement, and material strain, and the fitting error is controlled within 0.3 mm. A solid rebar model is generated using a cylindrical scanning algorithm. The cross-sectional diameter is consistent with the actual rebar, and the number of surface subdivision layers is 20 to ensure a smooth transition. A spatial layer overlay is performed on the concrete matrix, spalling area, crack network, and rebar entity, with transparency set to 0.9, 0.7, 0.8, and 1.0, respectively. Material properties include diffuse reflectance of 0.6, specular reflectance of 0.2, and roughness of 0.4. A four-point lighting scheme is used for ambient lighting configuration. The main light source is located directly above with an intensity of 800 lumens, and three auxiliary light sources are distributed on the front and sides with an intensity of 200 lumens. The light source ratio is 4:1:1:1. The image was rendered using a ray tracing algorithm, employing Monte Carlo path tracing, sampling 512 times per pixel at a resolution of 3840 × 2160 pixels, with each pixel corresponding to an actual size of 0.15 mm. The resulting damage image clearly displays the crack orientation, the gradient distribution of spalling depth, and the spatial pattern of rebar buckling. The damage characteristics in the plastic hinge region are intuitively discernible, supporting the precision requirements of quantitative assessment and structural analysis.

[0104] The above embodiments are intended to illustrate the technical solutions of the present invention and are not intended to limit the present invention. The present invention is described in detail with reference to the preferred embodiments only. It should be understood by those skilled in the art that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or equivalents should be included within the scope of the claims of the present invention.

Claims

1. A high-definition image fast reconstruction method for concrete plastic hinge area identification, characterized by: The method comprises: Extract an image set of intersecting oblique cracks and concrete crush areas related to the plastic hinge properties of beam-column joints from multi-view image sequences acquired through drone oblique photography and ground-based handheld devices. If the image is incomplete due to obstruction by falling objects or dust, generate an image set excluding the obstruction. Based on the image set with occluded objects removed, the texture features of the dust deposition masked area are integrated to restore the details of the masked cross-slant cracks and obtain a complete crack morphology image set; The spatial distribution characteristics of cross-slant cracks and main cracks are extracted from the complete crack morphology image set. Combined with the steel bar layout information of the node reinforcement drawings, a three-dimensional crack model is constructed to obtain the spatial distribution parameters of the cracks. The damage state image of the plastic hinge area is generated based on the spatial distribution parameters of the crack.

2. The method according to claim 1, characterized in that The method extracts a set of images of cross-oblique cracks and concrete crushing areas related to the plastic hinge properties of beam-column joints from a multi-view image sequence acquired through oblique photography by a drone and a ground-based handheld device. If the image is incomplete due to obstruction by falling objects or dust, a set of images excluding the obstructions is generated, including: Acquire feature descriptors based on a multi-view image sequence, obtain a set of image corresponding points using a nearest neighbor matching algorithm, and use the corresponding point set to calculate a spatial transformation matrix; Using the spatial transformation matrix to register the multi-view image sequence, using a stereo reconstruction algorithm to generate a point cloud model, and identifying missing areas in the point cloud model based on a density threshold segmentation method; Vectorizing the node reinforcement drawings to obtain a steel bar geometric model, aligning the steel bar geometric model with the point cloud model using a rigid body transformation algorithm to obtain actual steel bar distribution data of the node; Based on the surface mesh division results of the point cloud model, the region growing algorithm is used to extract the crack contour line. The crack contour line is calculated to obtain the damage feature vector. The damaged area is classified and judged through a convolutional neural network to determine the concrete damage area and generate a set of images without occlusions.

3. The method according to claim 1, characterized in that The method of removing the occluded image set and fusing the texture features of the dust deposit masked area to restore the masked cross-slant crack details and obtain a complete crack morphology image set includes: The local area texture contrast and uniformity parameters are calculated based on the image gray value distribution, and the initial crack occlusion area map is obtained by the maximum entropy threshold segmentation method; Calculating a gray level co-occurrence matrix for the initial crack occlusion area map, obtaining angular second-order moments and correlation parameters from the gray level co-occurrence matrix, and constructing a texture feature vector; Performing enhancement processing on the texture feature vector, obtaining an enhancement coefficient through grayscale gradient consistency evaluation, and obtaining enhanced texture features; The crack edge response map is calculated based on the enhanced texture features, the crack intersection positions are extracted, and the complete crack morphology image is obtained by wavelet transformation reconstruction.

4. The method according to claim 1, wherein The spatial distribution characteristics of the cross-slant cracks and main cracks are extracted from the complete crack morphology image set, and the three-dimensional crack model is constructed in combination with the steel bar layout information of the node reinforcement drawing to obtain the spatial distribution parameters of the cracks, including: The edge contour line in the crack image is extracted by using the Sobel edge detection operator, the crack length and width parameters are calculated based on the edge contour line, and the crack centerline data is obtained by using the morphological thinning algorithm; Extract the reinforcement spatial layout information based on the node reinforcement drawings, establish the longitudinal reinforcement and stirrup position matrix, and calculate the reinforcement three-dimensional spatial reference frame based on the position matrix; Directional field analysis is performed on the crack edge feature dataset. The crack strike angle is extracted based on the results of the direction field analysis, and the spatial coordinates of the crack are calibrated using the steel bar three-dimensional spatial reference frame. The crack edge contour is spatially mapped, and the three-dimensional projection trajectory of the crack is calculated based on the spatial mapping result. The complete spatial distribution model of the crack is obtained through the spatial interpolation algorithm, and the spatial distribution parameters of the crack are obtained.

5. The method according to claim 1, characterized in that Generating a damage state image of the plastic hinge region according to the spatial distribution parameters of the cracks includes: Based on the spatial distribution parameters of cracks, the grayscale gradient characteristics of the concrete crush area are introduced to calculate the spalling depth of the crushed area under the influence of occlusion and dust. The spalling depth damage model is generated. The clarity and continuity of the three-dimensional contours of the steel bars in the spalling depth damage model are analyzed to determine whether the buckling morphology of the steel bars in the spalling depth damage model is clear. If the buckling morphology of the steel bars in the spalling depth damage model is unclear, the buckling morphology parameters are derived by combining the steel bar diameter and spacing constraint information from the node reinforcement drawings and the morphology of the exposed steel bars observed in the occluded image set to obtain an updated spalling depth damage model. If the buckling morphology of the steel bars in the spalling depth damage model is clear, the buckling morphology parameters of the steel bars in the spalling depth damage model are used to perform damage assessment or structural response analysis; The final damage model was generated by optimizing the 3D space reconstruction using point cloud data from drone oblique photography combined with the updated spalling depth damage model. Based on the final damage model, the spatial distribution of cross-diagonal cracks, concrete spalling depth, and reinforcement buckling morphology are visualized and reconstructed in three dimensions, generating an image of the damage state in the plastic hinge area.

6. The method according to claim 5, characterized in that The method uses the grayscale gradient characteristics of the concrete crush area in response to the spatial distribution parameters of the cracks to calculate the spalling depth of the crush area under the influence of occlusion and dust, and generates a spalling depth damage model. The method then determines whether the buckling morphology of the steel bars in the spalling depth damage model is clear by analyzing the clarity and continuity of the three-dimensional contours of the steel bars in the model, including: The gradient amplitude distribution data of the crushed area is obtained based on the grayscale image, and the contour map of the crushed area is obtained by the regional adaptive threshold segmentation algorithm; Performing edge detection operation on the crushed area contour map, using a gradient operator to calculate the gradient direction field and edge intensity value, and obtaining a crushed area boundary marker map through a double threshold segmentation method; A random forest regression model is established based on the crushed area boundary marker map, and the gradient amplitude, gradient direction field and edge strength value are used as input features to obtain a three-dimensional morphological map of the crushed area; Extracting the exposed area of ​​the steel bar from the three-dimensional morphological map of the crushed area, calculating the gray gradient value of the steel bar edge, and obtaining the steel bar spatial contour by least squares fitting; The curvature of the steel bar spatial profile is calculated, and a threshold criterion for the steel bar deformation curvature is established. The integrity of the steel bar profile is evaluated using the connected domain analysis method to determine whether the steel bar buckling morphology in the spalling depth damage model is clear.

7. The method according to claim 5, characterized in that If the buckling morphology of the steel bars in the spalling depth damage model is unclear, the buckling morphology parameters are derived by combining the steel bar diameter and spacing constraint information from the node reinforcement drawings and the steel bar exposed area morphology observed in the occluded image set to obtain an updated spalling depth damage model, including: Extract the reinforcement spatial layout information based on the node reinforcement drawings, establish a three-dimensional coordinate reference system, obtain the diameter and spacing parameters of the longitudinal bars and stirrups through contour vectorization, and construct a standard reinforcement layout spatial position diagram; According to the standard steel bar layout spatial position diagram, a region growing algorithm is used to mark the steel bar surface point set, and a steel bar centerline equation is obtained by least squares fitting; According to the steel bar centerline equation, a curvature threshold segmentation method is used to identify the key points of steel bar deformation, and obtain the steel bar deformation curvature parameters and inclination parameters; A steel bar buckling deformation description function is established according to the steel bar deformation curvature parameter and the inclination parameter. The steel bar buckling deformation description function is used to predict the deformation of the blocked area, and the steel bar buckling deformation extension trajectory is obtained to obtain an updated spalling depth damage model.

8. The method according to claim 5, characterized in that If the buckling morphology of the steel bars in the spalling depth damage model is clear, the steel bar buckling morphology parameters in the spalling depth damage model are used to perform damage assessment or structural response analysis, including: Obtaining a steel bar spatial curve equation according to the spalling depth damage model, calculating a curve discrete point coordinate sequence using the steel bar spatial curve equation, and calculating a steel bar deformation field distribution using the curve discrete point coordinate sequence; Establishing a strain distribution function for the steel bar deformation field distribution, calculating the steel bar yield strain ratio using the strain distribution function, and constructing a node stiffness matrix according to the steel bar yield strain ratio; receiving the node stiffness matrix to calculate the material strength loss distribution, establishing a node damage variable field according to the material strength loss distribution, and calculating a damage degree index; The damage degree index is obtained to construct a neural network prediction model. If the neural network prediction model converges, a hierarchical clustering method is used to spatially divide the damage area.

9. The method according to claim 5, characterized in that The point cloud data obtained through drone oblique photography is combined with the updated spalling depth damage model to optimize the reconstruction of the three-dimensional space and generate the final damage model, including: The voxel grid division method is used to resample the point cloud data of the UAV oblique photography. The surface normal vector and curvature distribution data of the point cloud are obtained through the local feature description algorithm based on the resampled point cloud. Boundary features are extracted based on the surface normal vector and curvature distribution data of the point cloud, and an initial surface model of the damaged area is generated through a mesh reconstruction algorithm, wherein the initial surface model of the damaged area includes volume and surface area parameters of the damaged area; Performing spatial stratified sampling on the initial surface model of the damage area, and obtaining crack extension path data by establishing a surface feature vector field and a random forest regressor, wherein the crack extension path data is used to construct a crack space topology map; The crack space topology map is spatially segmented using an adaptive grid subdivision algorithm, and the buckling morphology of the steel bars and the concrete crush boundary data in the damage area are identified through a neural network segmenter to generate a final damage model.

10. The method according to claim 5, characterized in that According to the final damage model, the spatial distribution of cross-slant cracks, the depth of concrete spalling, and the visual three-dimensional reconstruction results of the steel bar buckling morphology are output to generate a damage state image of the plastic hinge area, including: A three-dimensional mesh reconstruction algorithm is used to generate a fracture mesh model based on the fracture spatial coordinate sequence, and a continuous fracture surface is obtained through cubic spline interpolation. Performing mesh encryption processing on the continuous surface of the crack to obtain the depth value of the spalling area, and establishing a color mapping rule according to the depth value to generate the surface of the spalling area; The surface of the spalled area is used to construct a plastic hinge damage boundary, and the damage boundary is processed by a surface smoothing algorithm to obtain a true texture of the damage area; The three-dimensional skeleton line of the steel bar buckling deformation is obtained for the damaged area, and the random forest regressor is used to fit the skeleton line to obtain the steel bar centerline. The solid steel bar model is generated through cylindrical scanning, and the damage state image of the plastic hinge area is generated.

Images