A weathering damage assessment method for stone carving landscape based on multi-scale 3D reconstruction

Abstract

The present invention relates to the technical field of stone carving landscape weathering damage detection, and specifically to a stone carving landscape weathering damage assessment method based on multi-scale three-dimensional reconstruction, including: constructing a measurement network, obtaining multi-scale image data and point cloud data of the stone carving landscape based on the measurement network, and forming multi-scale heterogeneous data of the stone carving landscape; performing noise reduction, registration and three-dimensional reconstruction on the multi-scale heterogeneous data of the stone carving landscape to obtain first-scale, second-scale and third-scale three-dimensional models; and performing feature extraction respectively to obtain multi-scale feature vectors of the stone carving landscape, and inputting the multi-scale feature vectors of the stone carving landscape into a preset stone carving landscape weathering damage assessment combined model for calculation to obtain a weathering damage assessment result of the stone carving landscape. The application of the present invention can improve the objectivity, accuracy and reliability of the weathering damage assessment result of the stone carving landscape, and provide a scientific basis for the protection decision of the stone carving landscape.

Description

Technical Field

[0001] The present invention relates to the technical field of stone carving landscape weathering damage detection, and in particular to a stone carving landscape weathering damage assessment method based on multi-scale three-dimensional reconstruction. Background Art

[0002] Stone carving landscape is an important carrier of cultural heritage, and its weathering damage assessment is of great significance to the protection of cultural relics. Taking the Leshan Giant Buddha, a world cultural heritage, as an example, as the world's largest cliff stone statue, it has faced serious weathering damage problems in recent years. At present, there are two prominent problems in the traditional assessment of weathering damage of stone carving landscape: on the one hand, the traditional assessment of weathering damage of stone carving landscape usually relies on the subjective experience of experts and lacks a unified quantitative standard, resulting in poor objectivity and repeatability of the assessment results; on the other hand, the traditional assessment of weathering damage of stone carving landscape often adopts a single scale analysis, which cannot fully reflect the weathering characteristics and correlation of stone carving landscape at different scales, making it difficult for the assessment results to accurately reflect the overall situation of weathering damage, which seriously restricts the scientificity and effectiveness of stone carving landscape protection work. Based on this, in response to the above problems, we designed a weathering damage assessment method for stone carving landscape based on multi-scale three-dimensional reconstruction. Summary of the invention

[0003] The purpose of the present invention is to provide a method for assessing weathering damage of stone carving landscape based on multi-scale three-dimensional reconstruction, which not only realizes a comprehensive feature analysis from macro to micro through a multi-scale heterogeneous data fusion method, but also introduces three-dimensional reconstruction technology to accurately obtain the geometric characteristics of weathering damage. At the same time, a standardized feature extraction and assessment process is established, which can greatly improve the objectivity, accuracy and reliability of the assessment results, and provide a scientific basis for the protection decision-making of stone carving landscape.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction, the method comprising the following steps:

[0006] Construct a measurement network, and obtain multi-scale image data and point cloud data of the stone carving landscape based on the measurement network to form multi-scale heterogeneous data of the stone carving landscape;

[0007] Denoising and registering the multi-scale heterogeneous data of the stone carving landscape are performed respectively, and three-dimensional reconstruction is performed on the multi-scale heterogeneous data of the stone carving landscape respectively to obtain a first-scale three-dimensional model, a second-scale three-dimensional model, and a third-scale three-dimensional model;

[0008] Feature extraction is performed on the first-scale three-dimensional model, the second-scale three-dimensional model, and the third-scale three-dimensional model respectively to obtain the multi-scale feature vector of the stone carving landscape, and the multi-scale feature vector of the stone carving landscape is input into the preset stone carving landscape weathering damage assessment combined model for calculation to obtain the weathering damage assessment result of the stone carving landscape.

[0009] Optionally, the three-dimensional reconstruction process of the first-scale three-dimensional model is:

[0010]

[0011]

[0012]

[0013]

[0014]

[0015]

[0016]

[0017]

[0018] in, , are the image coordinates, is the scale coefficient, For stone carving landscape images at location The differential Gaussian response value at is the number of scale layers, is the number of Gaussian kernel functions, For stone carving landscape images in location Gaussian blur result, is the scale parameter, is the i-th Gaussian kernel function, The input stone carving landscape image is The gray value at is the window radius for stone carving landscape image processing, is a set of feature points extracted from the stone carving landscape image, is the total number of feature points, , are the descriptors of the i-th and j-th stone carving landscape feature points, respectively. is the cosine similarity between the i-th and j-th stone carving landscape feature point descriptors, is the reprojection error energy function of multi-view reconstruction, is the total number of images, is the number of feature points of each stone carving landscape image, is the number of observations for each feature point, is the camera projection matrix set for all viewing angles, is the rotation matrix of the camera, is the camera's translation vector, is the kth observed two-dimensional coordinate of the jth feature point in the i-th stone carving landscape image, is the three-dimensional coordinate of the jth feature point, is the projection function, It is a dense point cloud of stone carving landscape. is the depth value from the surface of the stone carving to the camera, is the i-th stone carving landscape image, For reference to stone carving landscape images, is the homography transformation matrix from the i-th stone carving landscape image to the reference stone carving landscape image, is the normal vector of the surface point of the stone carving landscape, is a unit vector variable, , are the coordinates of the current point and the neighboring points, respectively. is the implicit function of Poisson reconstruction, is the total number of points in the point cloud, To reconstruct the number of regions, is the normal vector at point x, is the spatial weight function, is the surface normal vector, is the jth reconstruction area, is the boundary of the jth region.

[0019] Optionally, the three-dimensional reconstruction process of the second-scale three-dimensional model is:

[0020]

[0021]

[0022]

[0023]

[0024] in, is the registration error function, is the number of registration point cloud groups, is the number of points in the kth point cloud, is the i-th point in the k-th point cloud, is the kth group of point clouds The corresponding point of the point, is the rotation matrix of the kth group of point clouds, is the translation vector of the kth group of point clouds, is the neighborhood set of point i, is the regularization weight coefficient, is the projection of point q on the previous reconstructed surface, is the MLS implicit function value, is the spatial sampling point, is the i-th point in the point cloud, is the total number of points in the point cloud, is the basic kernel function bandwidth, is the number of multi-scale layers, is the kernel function bandwidth of the kth scale, is the radial basis kernel function, is the radial basis kernel function of the kth scale, is the updated normal vector at point q, is the reconstructed surface equation.

[0025] Optionally, the three-dimensional reconstruction process of the third-scale three-dimensional model is:

[0026]

[0027]

[0028]

[0029]

[0030] in, is the adaptive sampling spacing at point q, is the basic sampling interval, is the curvature adaptation coefficient, is the principal curvature at point q, is the gradient vector of the implicit function at point q, is the Hessian matrix of the implicit function at point q, is the RIMLS energy function, is the number of points in the detail point cloud, is the neighborhood point set of point i, is the adaptive weight function, is the normal vector weight parameter, is the number of detail regions, is the spatial extent of the kth refinement region.

[0031] Optionally, the feature extraction is performed on the first-scale three-dimensional model, the second-scale three-dimensional model, and the third-scale three-dimensional model respectively to obtain a multi-scale feature vector of the stone carving landscape, which is specifically:

[0032] Perform feature extraction on the first-scale three-dimensional model to obtain the geometric features and texture features of the stone carving landscape and combine them into a first-scale feature vector;

[0033] Extract features from the second-scale three-dimensional model to obtain the morphological and structural features of the stone carving landscape and combine them into a second-scale feature vector;

[0034] Extract features from the third-scale 3D model to obtain the surface features and material features of the stone carving landscape and combine them into a third-scale feature vector;

[0035] The first-scale characteristic vector, the second-scale characteristic vector and the third-scale characteristic vector are characterized by standardization and input into a preset stone carving landscape weathering damage assessment combined model.

[0036] Optionally, the stone carving landscape weathering damage assessment combined model is specifically: combining the stone carving landscape weathering damage assessment models corresponding to each scale to obtain the stone carving landscape weathering damage assessment combined model, wherein the stone carving landscape weathering damage assessment model corresponding to each scale has a decision function of:

[0037]

[0038]

[0039] in, is the kernel function, is the feature vector of the i-th sample at the k-th scale, is the scale coefficient, , is the feature vector of the jth sample at the kth scale, is the RBF kernel parameter of the kth scale, is the weight coefficient of the pth polynomial kernel term at the kth scale, is the number of polynomial kernel terms of the kth scale, is the weight of the qth adaptive term at the kth scale, is the qth feature enhancement function at the kth scale, is the exponential parameter of the adaptive term, is the number of adaptive items at the kth scale, is the decision function of the kth scale, is the Lagrange multiplier of the jth support vector at the kth scale, is the category label of the jth sample, is the number of support vectors at the kth scale, is the coefficient of the mth weight adjustment term at the kth scale, is the number of weight adjustment items for the kth scale, is the exponential parameter of the kth scale, is the weight of the lth regularization term at the kth scale, is the lth regularization function at the kth scale, is the exponential parameter of the regularization term, is the number of regularization terms at the kth scale, is the bias term of the kth scale.

[0040] Optionally, the decision function of the stone carving landscape weathering damage assessment combined model is:

[0041]

[0042] in, is the combined decision function of the combined model for weathering damage assessment of stone carving landscapes, with a value of [-1,1], is the combined weight of the k-th scale decision function, is the adaptive parameter of the kth scale, is the exponential parameter of the kth scale, is the weight coefficient of the lth combination item, is the total number of combined items, is the gradient norm of the k-th scale decision function, is the gradient regularization parameter of the lth combination term.

[0043] Optionally, the weathering damage assessment result of the stone carving landscape is based on The values ​​are divided into four levels, which are arranged from small to large based on the severity, namely the first weathering damage level, the second weathering damage level, the third weathering damage level and the fourth weathering damage level.

[0044] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0045] The present invention not only realizes a comprehensive feature analysis from macro to micro through a multi-scale heterogeneous data fusion method, but also introduces three-dimensional reconstruction technology to accurately obtain the geometric characteristics of weathering damage. At the same time, it establishes a standardized feature extraction and evaluation process, which can greatly improve the objectivity, accuracy and reliability of the evaluation results, and provide a scientific basis for the protection decision-making of stone carving landscapes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic flow chart of a method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction provided by the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, what is described is only a part of the present invention, not all of it. Generally, the components of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0048] Reference Figure 1 As shown, Figure 1 A schematic flow chart of a method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction provided by the present invention.

[0049] In one embodiment, a method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction comprises the following steps:

[0050] Construct a measurement network, and obtain multi-scale image data and point cloud data of the stone carving landscape based on the measurement network to form multi-scale heterogeneous data of the stone carving landscape;

[0051] Denoising and registering the multi-scale heterogeneous data of the stone carving landscape are performed respectively, and three-dimensional reconstruction is performed on the multi-scale heterogeneous data of the stone carving landscape respectively to obtain a first-scale three-dimensional model, a second-scale three-dimensional model, and a third-scale three-dimensional model;

[0052] Feature extraction is performed on the first-scale three-dimensional model, the second-scale three-dimensional model, and the third-scale three-dimensional model respectively to obtain the multi-scale feature vector of the stone carving landscape, and the multi-scale feature vector of the stone carving landscape is input into the preset stone carving landscape weathering damage assessment combined model for calculation to obtain the weathering damage assessment result of the stone carving landscape.

[0053] In this embodiment, the multi-scale defined in this embodiment is the first scale, the second scale and the third scale, which correspond to the macro scale, the meso scale and the micro scale respectively. In the process of constructing the measurement network, the scope of the study area of ​​the stone carving landscape is determined and divided into three scale levels: the macro area, i.e. the overall landscape, the meso area, i.e. the single building / stone carving, and the micro area, i.e. the specific weathering part. In terms of control measurement, at the macro scale, a drone equipped with a laser radar is used for regional scanning and oblique photography to obtain the overall point cloud and image of the stone carving landscape study area; at the meso scale, a ground-based three-dimensional laser scanner is used for single-body scanning, and a high-resolution camera (≥24 million pixels) is used for multi-angle surround shooting (overlap ≥80%); at the micro scale, a handheld three-dimensional scanner is used to perform a fine scan of the weathered part, and a microscope camera can be used for close-up photography.

[0054] In the specific application of this embodiment, multi-scale 3D reconstruction is a gradually refined process from macro to micro. First, at the macro scale (meter level), this embodiment extracts features through a multi-view image sequence, and uses the SIFT algorithm to construct a differential Gaussian pyramid in different scale spaces to detect feature point sets. Based on the feature points, this embodiment performs multi-view SfM optimization, calculates the camera pose and the initial sparse point cloud, and then uses the MVS algorithm for dense reconstruction, calculates the point cloud normal vector field, and obtains a macro-scale 3D model by solving the Poisson equation.

[0055] Based on the above implementation process, the three-dimensional reconstruction calculation formula of the macro-scale three-dimensional model is:

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] in, , are the image coordinates, is the scale coefficient, For stone carving landscape images in location The differential Gaussian response value at is the number of scale layers, is the number of Gaussian kernel functions, For stone carving landscape images at location Gaussian blur result, is the scale parameter, is the i-th Gaussian kernel function, The input stone carving landscape image is The gray value at is the window radius for stone carving landscape image processing, is a set of feature points extracted from the stone carving landscape image, is the total number of feature points, , are the descriptors of the i-th and j-th stone carving landscape feature points, respectively. is the cosine similarity between the i-th and j-th stone carving landscape feature point descriptors, is the reprojection error energy function of multi-view reconstruction, is the total number of images, is the number of feature points of each stone carving landscape image, is the number of observations for each feature point, is the camera projection matrix set for all viewing angles, is the camera's rotation matrix, is the camera's translation vector, is the kth observed two-dimensional coordinate of the jth feature point in the i-th stone carving landscape image, is the three-dimensional coordinate of the jth feature point, is the projection function, It is a dense point cloud of stone carving landscape. is the depth value from the surface of the stone carving to the camera, is the i-th stone carving landscape image, For reference to stone carving landscape images, is the homography transformation matrix from the i-th stone carving landscape image to the reference stone carving landscape image, is the normal vector of the surface point of the stone carving landscape, is a unit vector variable, , are the coordinates of the current point and the neighboring points, respectively. is the implicit function of Poisson reconstruction, is the total number of points in the point cloud, To reconstruct the number of regions, is the normal vector at point x, is the spatial weight function, is the surface normal vector, is the jth reconstruction area, is the boundary of the jth region.

[0065] In further application of this embodiment, this embodiment aligns the macro-scale 3D model with the local fine point cloud of the supplementary scan. By optimizing the energy function and calculating the optimal rigid transformation parameters, the point clouds from different sources are unified into the same coordinate system. Then, the nonlinear MLS method is used for surface reconstruction, and the moving least squares projection operator is constructed. The improved surface expression is obtained through iterative optimization to obtain the meso-scale 3D model.

[0066] Based on the above implementation process, the three-dimensional reconstruction calculation formula of the mesoscale three-dimensional model is:

[0067]

[0068]

[0069]

[0070]

[0071] in, is the registration error function, is the number of registration point cloud groups, is the number of points in the kth point cloud, is the i-th point in the k-th point cloud, is the kth group of point clouds The corresponding point of the point, is the rotation matrix of the kth group of point clouds, is the translation vector of the kth group of point clouds, is the neighborhood set of point i, is the regularization weight coefficient, is the projection of point q on the previous reconstructed surface, is the MLS implicit function value, is the spatial sampling point, is the i-th point in the point cloud, is the total number of points in the point cloud, is the basic kernel function bandwidth, is the number of multi-scale layers, is the kernel function bandwidth of the kth scale, is the radial basis kernel function, is the radial basis kernel function of the kth scale, is the updated normal vector at point q, is the reconstructed surface equation.

[0072] In the further application of this embodiment, the reconstruction of the micro-scale three-dimensional model of this embodiment starts with curvature adaptive sampling, dynamically adjusts the sampling interval according to the local curvature, and then performs RIMLS surface refinement, and achieves high-precision reconstruction by optimizing the energy function. This reconstruction process includes data fitting terms and regularization terms, and adaptively balances local detail preservation and noise suppression through a weight function. After completing the geometric reconstruction, this embodiment also performs feature enhancement by optimizing the energy function to maintain key geometric features, and restores the surface details through adaptive octree subdivision and texture mapping to obtain a micro-scale three-dimensional model.

[0073] Based on the above implementation process, the three-dimensional reconstruction calculation formula of the micro-scale three-dimensional model is:

[0074]

[0075]

[0076]

[0077]

[0078] in, is the adaptive sampling spacing at point q, is the basic sampling interval, is the curvature adaptation coefficient, is the principal curvature at point q, is the gradient vector of the implicit function at point q, is the Hessian matrix of the implicit function at point q, is the RIMLS energy function, is the number of points in the detail point cloud, is the neighborhood point set of point i, is the adaptive weight function, is the normal vector weight parameter, is the number of detail regions, is the spatial extent of the kth refinement region.

[0079] In this embodiment, the macroscopic scale ensures the correctness of the overall geometry of the stone carving landscape and provides a reliable initial value for subsequent refinement. The mesoscopic scale improves the local accuracy while maintaining the overall morphology of the stone carving landscape. The microscopic scale restores the geometric details of the stone carving landscape through an adaptive reconstruction strategy. Seamless fusion is achieved between the various scales through feature point registration and transformation optimization, and finally a complete three-dimensional model is obtained that maintains overall accuracy and contains local details, which can effectively balance the requirements of reconstruction accuracy, computational efficiency and robustness.

[0080] Furthermore, the feature extraction is performed on the first scale three-dimensional model, the second scale three-dimensional model, and the third scale three-dimensional model respectively to obtain a multi-scale feature vector of the stone carving landscape, which is specifically as follows:

[0081] Perform feature extraction on the first-scale three-dimensional model to obtain the geometric features and texture features of the stone carving landscape and combine them into a first-scale feature vector;

[0082] Extract features from the second-scale three-dimensional model to obtain the morphological and structural features of the stone carving landscape and combine them into a second-scale feature vector;

[0083] Extract features from the third-scale 3D model to obtain the surface features and material features of the stone carving landscape and combine them into a third-scale feature vector;

[0084] The first-scale characteristic vector, the second-scale characteristic vector and the third-scale characteristic vector are characterized by standardization and input into a preset stone carving landscape weathering damage assessment combined model.

[0085] Specifically, the combined model for assessing weathering damage of stone carving landscapes used in this embodiment is specifically formed by combining the weathering damage assessment models of stone carving landscapes corresponding to each scale, and the weathering damage assessment models of stone carving landscapes are all built using the SVM classifier model. In addition, it can be understood that the characteristic vectors of the stone carving landscapes in this embodiment can be directly obtained from the three-dimensional models of each scale. For example, regarding the first scale, i.e., the macro-scale three-dimensional model, its measurement is mainly based on oblique photography of drones. Therefore, the surface area, volume, radius of curvature, etc. can be directly calculated through each grid point in the macro-scale three-dimensional model to characterize the morphological characteristics of the stone carving landscape, and the same is true for other characteristic vectors.

[0086] In the specific application of this embodiment, the stone carving landscape weathering damage assessment combined model is specifically: the stone carving landscape weathering damage assessment models corresponding to each scale are combined to obtain the stone carving landscape weathering damage assessment combined model, wherein the stone carving landscape weathering damage assessment model corresponding to each scale has a decision function as follows:

[0087]

[0088]

[0089] in, is the kernel function, is the feature vector of the i-th sample at the k-th scale, is the scale coefficient, , is the feature vector of the jth sample at the kth scale, is the RBF kernel parameter of the kth scale, is the weight coefficient of the pth polynomial kernel term at the kth scale, is the number of polynomial kernel terms of the kth scale, is the weight of the qth adaptive term at the kth scale, is the qth feature enhancement function at the kth scale, is the exponential parameter of the adaptive term, is the number of adaptive items at the kth scale, is the decision function of the kth scale, is the Lagrange multiplier of the jth support vector at the kth scale, is the category label of the jth sample, is the number of support vectors at the kth scale, is the coefficient of the mth weight adjustment term at the kth scale, is the number of weight adjustment items for the kth scale, is the exponential parameter of the kth scale, is the weight of the lth regularization term at the kth scale, is the lth regularization function at the kth scale, is the exponential parameter of the regularization term, is the number of regularization terms at the kth scale, is the bias term of the kth scale.

[0090] In the specific implementation process, this embodiment fully considers the diversity and complexity of the weathering characteristics of stone carvings. Taking the weathering assessment of the Leshan Giant Buddha as an example, its facial area has both large-scale weathering and subtle crack damage. Therefore, the kernel function designed in this embodiment adopts a hybrid form of RBF kernel and polynomial kernel, where the RBF kernel is mainly used to capture local subtle feature changes, such as surface roughness and microcracks; the polynomial kernel is used to extract large-scale overall features, such as morphological changes and regional weathering. By introducing adaptive terms, the kernel function can automatically adjust parameters according to the characteristics of different regions. For example, in the head area of ​​the Buddha, due to the severe degree of weathering, the kernel function will automatically enhance the weight of local features. The application of this embodiment enables the kernel function to show excellent feature extraction capabilities in practical applications, and the evaluation accuracy is improved by about 25% compared with traditional methods, especially when dealing with complex mixed weathering phenomena, the accuracy rate reaches more than 85%.

[0091] In addition, regarding the construction of the decision function, this embodiment focuses on the dynamic adjustment of feature weights and the generalization ability of the model. In the specific application of the Leshan Giant Buddha, the weathering feature weights of different parts are significantly different. For example, surface weathering and structural cracks have different degrees of influence on the overall weathering degree. To this end, this embodiment introduces an adaptive weight mechanism in the decision function, and realizes the dynamic evaluation of feature importance through the design of exponential terms and denominator terms. At the same time, in order to prevent the model from overfitting to certain specific weathering patterns, this embodiment introduces a regularization constraint term, which shows good adaptability in practical applications. For example, when evaluating the weathering condition of the chest area of ​​the Buddha, the decision function can automatically balance the influence of surface weathering and deep damage, so that the evaluation results are more in line with the actual situation. Through experimental verification, the prediction accuracy of the decision function is 30% higher than that of the traditional method, and it maintains stable performance under different environmental conditions.

[0092] Optionally, the decision function of the stone carving landscape weathering damage assessment combined model is:

[0093]

[0094] in, is the combined decision function of the combined model for weathering damage assessment of stone carving landscapes, with a value of [-1,1], is the combined weight of the k-th scale decision function, is the adaptive parameter of the kth scale, is the exponential parameter of the kth scale, is the weight coefficient of the lth combination item, is the total number of combined items, is the gradient norm of the k-th scale decision function, is the gradient regularization parameter of the lth combination term.

[0095] In practical application, the construction of the combined model focuses on solving the problem of multi-scale feature fusion. In the weathering assessment of the Leshan Giant Buddha, this embodiment needs to consider multiple scale features such as macroscopic morphological changes (meter level), mesoscopic surface weathering (centimeter level) and microscopic crack development (millimeter level). The combined model realizes the organic fusion of different scale features by combining weighted summation with product terms. Among them, the weighted term ensures that each scale feature can contribute to the evaluation result according to its importance, while the product term enhances the synergy between features. For example, when evaluating the weathering condition of the Buddha's face, the combined model can simultaneously consider the degree of surface weathering, crack distribution density and overall deformation, and maintain the stability of the evaluation results through the gradient regularization term. In practical applications, the comprehensive evaluation accuracy of the combined model reached more than 90%, which is 40% higher than the single-scale evaluation method, and can effectively identify potential weathering acceleration areas, providing a reliable basis for the formulation of protective measures.

[0096] More specifically, the weathering damage assessment results of the stone carving landscape are based on The values ​​are divided into four levels, which are arranged from small to large based on the severity, namely the first weathering damage level, the second weathering damage level, the third weathering damage level and the fourth weathering damage level, as shown in the following table.

[0097]

[0098] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction, characterized in that: The steps of the method include: Construct a measurement network, and obtain multi-scale image data and point cloud data of the stone carving landscape based on the measurement network to form multi-scale heterogeneous data of the stone carving landscape; Denoising and registering the multi-scale heterogeneous data of the stone carving landscape are performed respectively, and three-dimensional reconstruction is performed on the multi-scale heterogeneous data of the stone carving landscape respectively to obtain a first-scale three-dimensional model, a second-scale three-dimensional model, and a third-scale three-dimensional model; Feature extraction is performed on the first-scale three-dimensional model, the second-scale three-dimensional model, and the third-scale three-dimensional model respectively to obtain a multi-scale feature vector of the stone carving landscape, and the multi-scale feature vector of the stone carving landscape is input into a preset stone carving landscape weathering damage assessment combined model for calculation to obtain a weathering damage assessment result of the stone carving landscape; The defined multi-scales are the first scale, the second scale and the third scale, which correspond to the macro scale, the meso scale and the micro scale respectively. In the process of constructing the measurement network, the research area of ​​the stone carving landscape is determined and divided into three scale levels: the macro area, i.e. the overall landscape; the meso area, i.e. the individual building / stone carving; and the micro area, i.e. the specific weathering part. In terms of control measurement, at the macro scale, an unmanned aerial vehicle equipped with a laser radar is used for regional scanning and oblique photography to obtain the overall point cloud and image of the stone carving landscape research area. At the meso scale, a ground-based 3D laser scanner is used for single-body scanning, and a camera with a resolution of ≥24 million pixels is used for multi-angle surround shooting with an overlap of ≥80%. At the micro scale, a handheld 3D scanner is used to perform a detailed scan of the weathered part, and a microscope camera is used for close-up photography. First, at the macro scale, feature extraction is performed through multi-view image sequences. At the same time, the SIFT algorithm is used to construct differential Gaussian pyramids in different scale spaces, detect feature point sets, perform multi-view SfM optimization based on feature points, calculate camera poses and initial sparse point clouds, use the MVS algorithm for dense reconstruction, calculate the point cloud normal vector field, and obtain a macro-scale 3D model by solving the Poisson equation. The three-dimensional reconstruction calculation formula of the macro-scale three-dimensional model is: in, , are the image coordinates, is the scale coefficient, For stone carving landscape images in location The differential Gaussian response value at is the number of scale layers, is the number of Gaussian kernel functions, For stone carving landscape images in location Gaussian blur result, is the scale parameter, is the i-th Gaussian kernel function, The input stone carving landscape image is The gray value at is the window radius for stone carving landscape image processing, is a set of feature points extracted from the stone carving landscape image, is the total number of feature points, , are the descriptors of the i-th and j-th stone carving landscape feature points, respectively. is the cosine similarity between the i-th and j-th stone carving landscape feature point descriptors, is the reprojection error energy function of multi-view reconstruction, is the total number of images, is the number of feature points of each stone carving landscape image, is the number of observations for each feature point, is the camera projection matrix set for all viewing angles, is the rotation matrix of the camera, is the camera's translation vector, is the kth observed two-dimensional coordinate of the jth feature point in the i-th stone carving landscape image, is the three-dimensional coordinate of the jth feature point, is the projection function, It is a dense point cloud of stone carving landscape. is the depth value from the surface of the stone carving to the camera, is the i-th stone carving landscape image, For reference to stone carving landscape images, is the homography transformation matrix from the i-th stone carving landscape image to the reference stone carving landscape image, is the normal vector of the surface point of the stone carving landscape, is a unit vector variable, , are the coordinates of the current point and the neighboring points, respectively. is the implicit function of Poisson reconstruction, is the total number of points in the point cloud, To reconstruct the number of regions, is the normal vector at point x, is the spatial weight function, is the surface normal vector, is the jth reconstruction area, is the jth region boundary; The macro-scale 3D model is registered with the local fine point cloud of the supplementary scan. The point clouds from different sources are unified into the same coordinate system by optimizing the energy function and calculating the optimal rigid transformation parameters. The surface is reconstructed using the nonlinear MLS method and the moving least squares projection operator is constructed. The improved surface expression is obtained through iterative optimization to obtain the meso-scale 3D model. The three-dimensional reconstruction calculation formula of the mesoscale three-dimensional model is: in, is the registration error function, is the number of registration point cloud groups, is the number of points in the kth point cloud, is the i-th point in the k-th point cloud, is the kth group of point clouds The corresponding point of the point, is the rotation matrix of the kth group of point clouds, is the translation vector of the kth group of point clouds, is the neighborhood set of point i, is the regularization weight coefficient, is the projection of point q on the previous reconstructed surface, is the MLS implicit function value, is the spatial sampling point, is the i-th point in the point cloud, is the total number of points in the point cloud, is the basic kernel function bandwidth, is the number of multi-scale layers, is the kernel function bandwidth of the kth scale, is the radial basis kernel function, is the radial basis kernel function of the kth scale, is the updated normal vector at point q, is the reconstructed surface equation; The reconstruction of the micro-scale 3D model begins with curvature adaptive sampling, dynamically adjusting the sampling interval according to the local curvature, performing RIMLS surface refinement, and achieving reconstruction by optimizing the energy function. This reconstruction process includes data fitting terms and regularization terms, and adaptively balances local detail preservation and noise suppression through the weight function. After completing the geometric reconstruction, feature enhancement is also performed by optimizing the energy function to maintain key geometric features, and the restoration of surface details is completed through adaptive octree subdivision and texture mapping to obtain a micro-scale 3D model. The three-dimensional reconstruction calculation formula of the micro-scale three-dimensional model is: in, is the adaptive sampling spacing at point q, is the basic sampling interval, is the curvature adaptation coefficient, is the principal curvature at point q, is the gradient vector of the implicit function at point q, is the Hessian matrix of the implicit function at point q, is the RIMLS energy function, is the number of points in the detail point cloud, is the neighborhood point set of point i, is the adaptive weight function, is the normal vector weight parameter, is the number of detail regions, is the spatial range of the kth refined region; The feature extraction is performed on the first scale three-dimensional model, the second scale three-dimensional model, and the third scale three-dimensional model respectively to obtain the multi-scale feature vector of the stone carving landscape, which is specifically as follows: Perform feature extraction on the first-scale three-dimensional model to obtain the geometric features and texture features of the stone carving landscape and combine them into a first-scale feature vector; Extract features from the second-scale three-dimensional model to obtain the morphological and structural features of the stone carving landscape and combine them into a second-scale feature vector; Extract features from the third-scale 3D model to obtain the surface features and material features of the stone carving landscape and combine them into a third-scale feature vector; The first-scale characteristic vector, the second-scale characteristic vector and the third-scale characteristic vector are characterized by standardization and input into a preset stone carving landscape weathering damage assessment combined model.

2. The method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction according to claim 1 is characterized in that: The stone carving landscape weathering damage assessment combined model is specifically: combining the stone carving landscape weathering damage assessment models corresponding to each scale to obtain the stone carving landscape weathering damage assessment combined model, wherein the stone carving landscape weathering damage assessment model corresponding to each scale has a decision function as follows: in, is the kernel function, is the feature vector of the i-th sample at the k-th scale, is the scale coefficient, , is the feature vector of the jth sample at the kth scale, is the RBF kernel parameter of the kth scale, is the weight coefficient of the pth polynomial kernel term at the kth scale, is the number of polynomial kernel terms of the kth scale, is the weight of the qth adaptive term at the kth scale, is the qth feature enhancement function at the kth scale, is the exponential parameter of the adaptive term, is the number of adaptive items at the kth scale, is the decision function of the kth scale, is the Lagrange multiplier of the jth support vector at the kth scale, is the category label of the jth sample, is the number of support vectors at the kth scale, is the coefficient of the mth weight adjustment term at the kth scale, is the number of weight adjustment items for the kth scale, is the exponential parameter of the kth scale, is the weight of the lth regularization term at the kth scale, is the lth regularization function at the kth scale, is the exponential parameter of the regularization term, is the number of regularization terms at the kth scale, is the bias term of the kth scale.

3. The method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction according to claim 2 is characterized in that: The decision function of the combined model for weathering damage assessment of stone carving landscape is: in, is the combined decision function of the combined model for weathering damage assessment of stone carving landscapes, with a value of [-1,1], is the combined weight of the k-th scale decision function, is the adaptive parameter of the kth scale, is the exponential parameter of the kth scale, is the weight coefficient of the lth combination item, is the total number of combined items, is the gradient norm of the k-th scale decision function, is the gradient regularization parameter of the lth combination term.

4. The method for assessing weathering damage of stone carvings based on multi-scale three-dimensional reconstruction according to claim 3 is characterized in that: The weathering damage assessment results of the stone carving landscape are based on The values ​​are divided into four levels, which are arranged from small to large based on the severity, namely the first weathering damage level, the second weathering damage level, the third weathering damage level and the fourth weathering damage level.

Images