A building facade maintenance method and system based on multi-information fusion

Abstract

The present application relates to the technical field of historical building facade maintenance, and particularly relates to a building facade maintenance method and system based on multi-information fusion. The method comprises the following steps: S1: collecting point cloud data and image data of historical buildings, dividing different regions of the historical building facade, and constructing a feature parameter matrix and a repair case database according to the point cloud data, image data and different regions; S2: calculating a feature tensor of different regions according to the feature parameter matrix, and obtaining a feature hash code of different regions according to the feature tensor of different regions; and constructing a structural damage feature vector of different regions, and obtaining a structural hash code of different regions through the structural damage feature vector of different regions. The present application encodes the historical buildings in the old community by hash, and obtains similar repair processes.

Description

Technical Field

[0001] This invention relates to the field of facade maintenance technology for historical buildings, and in particular to a method and system for facade maintenance based on multi-information fusion. Background Technology

[0002] As urbanization continues, cracks and peeling on the facades of old residential communities can lower the overall image of the city and affect the quality of life of residents. Facade renovation is an important part of the renovation of old communities, which can improve the quality of the city and improve the living environment. Some old communities also contain historical buildings. The facades of these buildings must preserve historical culture while ensuring the quality of life of residents, which greatly increases the maintenance costs. The facades of historical buildings are not only an important part of the building structure but also a direct carrier of historical and cultural information. With the advancement of urbanization and long-term erosion by the natural environment, the facades of historical buildings generally face the prominent problem of simultaneous structural damage and degradation of their stylistic features. Existing facade maintenance technologies have the following shortcomings: First, stylistic features lack parametric expression; existing detection methods mostly focus on structural safety indicators, making it impossible to quantify the stylistic value of historical building facades. Second, damage assessment dimensions are singular; existing technologies equate facade damage with structural damage, failing to establish a two-tiered assessment system for both structural and stylistic damage, thus hindering the effective identification and quantification of stylistic damage. Third, maintenance plan generation is inefficient and lacks stylistic guidance; existing maintenance plans are largely based on structural damage detection results, lacking a technical path to transform stylistic damage assessment results into maintenance strategies, making it difficult to meet the rapid response requirements of actual projects. Therefore, there is an urgent need to develop a building facade maintenance method and system based on multi-information fusion. Summary of the Invention

[0003] To overcome the shortcomings of damage assessment dimensions and single maintenance schemes for the facade features of historical buildings, this invention provides a method and system for building facade maintenance based on multi-information fusion.

[0004] The technical solution of this invention is: a method for maintaining building facades based on multi-information fusion, comprising the following steps: S1: Collect point cloud data and image data of historical buildings, divide the facades of historical buildings into different areas, and construct a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. S2: Calculate the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtain the landscape hash code of different regions based on the landscape feature tensor of different regions; construct the structural damage feature vector of different regions, and obtain the structural hash code of different regions through the structural damage feature vector of different regions. S3: Obtain the first set of similar cases from the renovation case database based on the appearance hash code and structure hash code. Filter the first set of similar cases using a preset similarity threshold to obtain the second set of similar cases. Construct a maintenance knowledge graph based on the renovation case database. Construct a local maintenance subgraph for the current area based on the maintenance knowledge graph and the second set of similar cases. S4: Calculate the landscape damage index for different regions, obtain the comprehensive damage assessment vector for different regions based on the landscape damage index, and filter the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. S5: Calculate the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtain the optimal maintenance plan through multi-objective optimization.

[0005] Preferably, the process of collecting point cloud data and image data of historical buildings, dividing the facades of historical buildings into different regions, and constructing a style feature parameter matrix and a restoration case database based on the point cloud data, image data, and different regions includes: acquiring point cloud data, texture data, and image data of historical buildings using scanners and multispectral imaging technology; the point cloud data being spatial coordinates and color information in a spatial coordinate system constructed with the center point of the current facade; the image data being reflectance vectors in spectral bands; and the texture data being the texture of the facade in the current spatial coordinates; dividing the facades of historical buildings into regions based on a preset number of regions; acquiring color feature parameters, line-angle ratio parameters, and material feature parameters for the corresponding facade regions based on the point cloud data, image data, texture data, and corresponding facade regions; constructing a style feature parameter matrix for this region based on the color feature parameters, line-angle ratio parameters, and material feature parameters; and constructing a restoration case database for the current historical buildings based on their restoration records. The restoration case database consists of historical restoration records of the facades of the current historical buildings.

[0006] Preferably, the step of calculating the landscape feature tensors of different regions based on the landscape feature parameter matrix and obtaining the landscape hash codes of different regions based on the landscape feature tensors of different regions includes: mapping the landscape feature parameter matrix into a unified landscape feature tensor through feature-by-feature independent nonlinear embedding, wherein the landscape feature tensor is a parameter vector that uniformly represents different landscape features of different regions; performing feature aggregation on each region based on the landscape feature tensor; obtaining the global landscape vector of different regions by max pooling, wherein the global landscape vector is the dominant semantic feature vector of facade landscape damage in the current region; and binarizing the global landscape vectors of different regions through local sensitive hashing to obtain the landscape hash codes of different regions, wherein the landscape hash codes are binary sequences of the global landscape vectors of all regions, wherein the calculation formulas for the landscape feature tensor, the global landscape vector, and the landscape hash code are as follows: ; ; ; in, In the first The first in the region The landscape feature tensor of landscape features For the first Global landscape vector of each region For the first The landscape hash code of each region For activation function, For the weight vector, For bias vectors, For the first Region 1 Measured values ​​of several landscape feature parameters For symbolic functions, For the first The normal vector of the landscape segmentation line corresponding to each hash code. For the first The intercept of the landscape segmentation line corresponding to each hash code. This is the sequence number of the hash code. This is the minimum valid hash code length.

[0007] Preferably, the construction of structural damage feature vectors for different regions, and the acquisition of structural hash codes for different regions through these feature vectors, includes: acquiring crack data for different regions of the facade using a crack width measuring instrument; acquiring hollow data for different regions of the facade using infrared imaging; wherein the crack data represents the maximum width and maximum depth of cracks in different regions, and the hollow data represents the maximum bulge height and hollow area in different regions; calculating the degree of crack damage and the degree of hollow damage for different regions based on the crack data and hollow data respectively; constructing structural damage feature vectors for different regions based on the degree of crack damage and the degree of hollow damage; wherein the structural damage feature vectors are feature vectors generated by cracks and hollows in different regions of the facade; and performing binary encoding on the structural damage feature vectors of different regions using local sensitive hash mapping to obtain structural hash codes for different regions; wherein the structural hash codes are binary sequences of the structural damage feature vectors for all regions, and the calculation formulas for the degree of crack damage, the degree of hollow damage, the structural damage feature vectors, and the structural hash codes are as follows: ; ; ; ; in, For the first The extent of crack damage in each area For the first The degree of hollow damage in each area For the first Structural damage feature vectors of each region For the first The structural hash code of each region, For the first Maximum crack width in each region This represents the critical threshold for crack width. For the first Maximum crack depth in each region This represents the critical threshold for crack depth. For the first The area of ​​hollow spots in each region This represents the critical threshold for the area of ​​voids. For the first The maximum height of the uplift in each area The critical threshold for the height of the bulge. As the first weight, As the second weight, As the third weight, As the fourth weight, For the first The normal vector of the structural dividing line corresponding to each hash code. For the first The intercept of the structural dividing line corresponding to each hash code.

[0008] Preferably, the step of obtaining a first set of similar cases from the restoration case database based on the landscape hash code and structural hash code, and filtering the first set of similar cases based on a preset similarity threshold to obtain a second set of similar cases includes: calculating the landscape hash distance and structural hash distance respectively using the Hamming distance calculation method based on the landscape hash code, structural hash code, and cases in the restoration case database of different regions; setting a preset distance threshold; searching the restoration case database based on the landscape hash distance, structural hash distance, and distance threshold; the search condition is: if both the landscape hash distance and structural hash distance are less than or equal to the distance threshold, then the current restoration case is filtered out to obtain the first set of similar cases; setting a preset similarity threshold; filtering the first set of similar cases based on the similarity threshold to obtain the second set of similar cases; if the number of cases in the first set of similar cases is less than the similarity threshold, then the distance threshold is adjusted; wherein the calculation formulas for the landscape hash distance and structural hash distance are: ; ; in, For renovation cases With the The landscape hash distance of each region For renovation cases The appearance hash code, For the first The landscape hash code of each region For renovation cases With the The structural hash distance of each region For renovation cases The structure hash code, For the first The structural hash code of each region, This is the sequence number of the hash code. This is the minimum valid hash code length.

[0009] Preferably, the step of constructing a maintenance knowledge graph based on the repair case database and constructing a local maintenance subgraph for the current region based on the maintenance knowledge graph and the second similar case set includes: constructing a maintenance knowledge graph based on the repair case database, wherein the maintenance knowledge graph is a set of nodes of the maintenance knowledge graph consisting of case nodes, feature nodes, damaged nodes, and repair nodes in different repair cases; obtaining the association relationships between nodes based on the repair case database as an edge set between the node sets; injecting cases from the second similar case set as activation nodes into the maintenance knowledge graph; setting a preset edge number threshold; and obtaining a set of nodes starting from the activation node and having an associated edge number less than or equal to the edge number threshold to construct a local maintenance subgraph, wherein the local maintenance subgraph includes case nodes, feature nodes, damaged nodes, and repair nodes that have similar damage to the current second similar case set.

[0010] Preferably, the calculation of the landscape damage index for different areas, the acquisition of a comprehensive damage assessment vector for different areas based on the landscape damage index, and the screening of local maintenance sub-maps to obtain a candidate maintenance set based on the comprehensive damage assessment vector include: constructing a landscape benchmark parameter matrix based on the original data of the historical building facades; calculating the landscape damage index for different areas based on the landscape benchmark parameter matrix; constructing a comprehensive damage assessment vector for different areas based on the landscape damage index, the degree of crack damage, and the degree of hollow damage; performing initial screening of repair nodes on the local maintenance sub-maps to obtain initial screening sub-maps for local maintenance; calculating the matching degree of different repair processes for different areas based on the comprehensive damage assessment vector and the global landscape vector; refining the initial screening sub-maps for local maintenance based on the matching degree and sorting them in descending order of matching degree to obtain a maintenance and repair set. The maintenance and repair set is the set of repair processes most suitable for the current facade area. A minimum quantity threshold is preset, and the maintenance and repair set is judged based on the minimum quantity threshold to obtain a candidate maintenance set. The formulas for calculating the landscape damage index, the comprehensive damage assessment vector, and the matching degree are as follows: ; ; ; in, For the first Area use of number The degree of matching between the repair techniques For the first The landscape damage index of the area For the first Comprehensive damage assessment vector for the region For example The Comprehensive damage assessment vector for the region This refers to the case node number. To maintain the initial screening sub-graph, the first step was to use the second step. A collection of nodes in the repair process. For bandwidth parameters, The square of the Euclidean distance. For the first Global landscape vector of the region For the first A global landscape vector of historical cases. For the first The extent of crack damage in each area For the first The degree of hollow damage in each area The first in the landscape reference parameter matrix Region 1 The baseline values ​​for each landscape feature parameter, These are the landscape feature parameters, including color feature parameters, line proportion parameters, and material feature parameters, with a value of 3; For the first Region 1 Measured values ​​of several landscape feature parameters For the first Feature weights of each landscape feature parameter This is the first protection zero constant.

[0011] Preferably, the initial screening of repair nodes in the local maintenance sub-graph to obtain the initial screening sub-graph includes: removing material nodes from the local maintenance sub-graph to obtain a first sub-graph, wherein the first sub-graph consists of case nodes, repair nodes, feature nodes, and damaged nodes in the local maintenance sub-graph; traversing all repair nodes and filtering out repair nodes that meet the filtering criteria, wherein the current repair node must be connected to both feature nodes and damaged nodes; using the case nodes corresponding to all repair nodes that meet the filtering criteria as historical case tags for the current repair node, wherein the historical case tags for the current repair node are historical cases representing the current repair process; and collecting all repair nodes with historical case tags as the initial screening sub-graph for local maintenance.

[0012] Preferably, the step of calculating the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtaining the optimal maintenance scheme through multi-objective optimization, includes: obtaining predicted residual values ​​for different areas based on different candidate maintenance schemes adopted in historical cases, wherein the predicted residual values ​​are the residual values ​​of landscape feature parameters after repair in different areas; calculating the landscape fidelity based on the predicted residual values; obtaining the maintenance costs of different candidate maintenance schemes based on different candidate maintenance schemes; and performing multi-objective optimization on the current candidate maintenance set based on the landscape fidelity and maintenance cost to obtain the optimal maintenance scheme, wherein the calculation formulas for landscape fidelity and multi-objective optimization are: ; ; in, To adopt The fidelity of the design's appearance. For the first The candidate maintenance set of the region One element, For the first Region 1 Measured values ​​of several landscape feature parameters For the renovation of the first Region 1 Predicted residual values ​​of each landscape feature parameter The second prevention is zero constant. For multi-objective optimization, the optimal candidate maintenance scheme, As the first optimization weight, As the second optimization weight, To adopt The maintenance cost of the solution The highest cost among the candidate maintenance sets, To obtain the position index of the maximum value.

[0013] Preferably, a building facade maintenance system based on multi-information fusion further includes: The data acquisition module collects point cloud data and image data of historical buildings, divides the facades of historical buildings into different areas, and constructs a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. The hash index module calculates the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtains the landscape hash code of different regions based on the landscape feature tensor of different regions; it constructs the structural damage feature vector of different regions, and obtains the structural hash code of different regions through the structural damage feature vector of different regions. The knowledge graph module obtains the first set of similar cases from the repair case database based on the appearance hash code and the structure hash code. It then filters the first set of similar cases by setting a preset similarity threshold to obtain the second set of similar cases. Based on the repair case database, it constructs and maintains a knowledge graph. Based on the maintenance knowledge graph and the second set of similar cases, it constructs a local maintenance subgraph for the current region. The candidate scheme module calculates the landscape damage index for different regions, obtains the comprehensive damage assessment vector for different regions based on the landscape damage index, and filters the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. The optimal solution module calculates the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtains the optimal maintenance solution through multi-objective optimization.

[0014] The beneficial effects of this invention are as follows: 1. This invention establishes a multi-dimensional architectural style evaluation index system, transforming the traditionally subjective judgment of architectural color, molding proportion and material texture into measurable, calculable and comparable quantitative parameters, thereby providing an objective basis for the preservation status of different buildings and different areas and providing core technical support for maintenance decisions. 2. This invention breaks through the single evaluation model of existing technology that equates damage to building facades with structural damage, and constructs a two-layer technical framework for parallel evaluation of building safety and historical and cultural value. It quantifies safety hazards at the structural level and loss of historical information at the style level, effectively reducing the problem of decision-making imbalance. 3. This invention elevates the selection of the facade maintenance scheme for historical buildings from a single dimension of technical feasibility to a multi-objective comprehensive decision that takes into account both the quality of restoration and economic costs. By quantifying the comprehensive performance of different maintenance processes in terms of both the degree of restoration of appearance and the cost of the project, it reduces the one-sided decision-making of traditional methods. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the building facade maintenance method based on multi-information fusion according to the present invention. Figure 2 This is a schematic diagram of the building facade maintenance system based on multi-information fusion according to the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0017] Example 1: A method for maintaining building facades based on multi-information fusion, such as... Figure 1 As shown, it includes the following steps: S1: Collect point cloud data and image data of historical buildings, divide the facades of historical buildings into different areas, and construct a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. S2: Calculate the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtain the landscape hash code of different regions based on the landscape feature tensor of different regions; construct the structural damage feature vector of different regions, and obtain the structural hash code of different regions through the structural damage feature vector of different regions. S3: Obtain the first set of similar cases from the renovation case database based on the appearance hash code and structure hash code. Filter the first set of similar cases using a preset similarity threshold to obtain the second set of similar cases. Construct a maintenance knowledge graph based on the renovation case database. Construct a local maintenance subgraph for the current area based on the maintenance knowledge graph and the second set of similar cases. S4: Calculate the landscape damage index for different regions, obtain the comprehensive damage assessment vector for different regions based on the landscape damage index, and filter the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. S5: Calculate the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtain the optimal maintenance plan through multi-objective optimization.

[0018] Point cloud data, texture data, and image data of historical buildings are acquired using scanners and multispectral imaging technology. The point cloud data consists of spatial coordinates and color information in a spatial coordinate system constructed with the center point of the current facade. The image data is a reflectance vector in the spectral band. The texture data is the texture of the facade in the current spatial coordinates. The facade of the historical buildings is divided into regions based on a preset number of regions. Color feature parameters, line-angle ratio parameters, and material feature parameters of the corresponding facade regions are obtained based on the point cloud data, image data, texture data, and the corresponding facade regions. A style feature parameter matrix for this region is constructed based on the color feature parameters, line-angle ratio parameters, and material feature parameters. A restoration case database of the current historical buildings is constructed based on the restoration records of the current historical buildings. The restoration case database consists of historical restoration records of the facades of the current historical buildings.

[0019] It should be explained that the specific method for obtaining the number of preset areas is as follows: based on the regional distribution of historical cases, the current building facade is evenly divided; the calculation formulas for color characteristic parameters, molding ratio parameters, and material characteristic parameters are as follows: ; ; ; in, For the first Color feature parameters of the region For the first The molding ratio parameters of the area, For the first Material characteristic parameters of the region, For the first The brightness of the area As a standard reference brightness for current historical buildings, For the first The red-green color of the area The red and green color scheme serves as a standard reference for current historical buildings. For the first The yellow-blue hue of the area The standard reference for the current historical buildings is yellow-blue. For the first The total number of decorative moldings included in the area. For the pin number, For the first The first in the region The actual width of the molding. For the first The building bay width of the area For the current historical buildings The baseline width of the molding is proportional to the span. The total number of bands for a multispectral imaging device. For band number, For the first The region in Average spectral reflectance in the band, For the current historical buildings in the first The standard spectral reflectance in the band; the specific method for obtaining the brightness, red-green hue, and yellow-blue hue is to convert multispectral image data into the CIELAB color space of the current historical building, and calculate the average brightness, average red-green hue, and average yellow-blue hue of all pixels in the color space as brightness, red-green hue, and yellow-blue hue, all in dimensionless units; the specific method for obtaining the standard reference brightness, standard reference red-green hue, and standard reference yellow-blue hue is to measure the undamaged facade using a spectrophotometer, all in dimensionless units; the total number of decorative moldings is the total number of moldings on the current building facade, specifically obtained by... The clustering and recognition algorithm for the feature point cloud of the molding is automatically statistically analyzed, with the unit being dimensionless; the building bay width is the boundary point cloud of the current area, specifically obtained by extracting the left and right boundary point clouds of the current area from the 3D point cloud data and calculating the horizontal coordinate difference, with the unit being millimeters; the actual width is in millimeters; the reference width-to-bay ratio is obtained by measuring the actual width and actual width-to-bay ratio of the undamaged facade, and the ratio of the actual width to the actual width-to-bay ratio is the reference width-to-bay ratio, with the unit being dimensionless; the style feature parameter matrix has a dimension of 3*4, with the number of rows representing color feature parameters, line-corner ratio parameters, and material feature parameters, and the number of columns representing the preset number of areas.

[0020] Based on the landscape feature parameter matrix, a unified mapping is performed to a landscape feature tensor through feature-by-feature independent nonlinear embedding. This landscape feature tensor is a parameter vector uniformly representing different landscape features in different regions. Features are aggregated for each region based on the landscape feature tensor, and global landscape vectors for different regions are obtained using max-pooling. These global landscape vectors are the dominant semantic feature vectors of facade landscape damage in the current region. The global landscape vectors of different regions are then binarized and encoded using locality-sensitive hashing to obtain landscape hash codes for different regions. These hash codes are binary sequences of the global landscape vectors for all regions. The calculation formulas for the landscape feature tensor, global vectors, and landscape hash codes are as follows: ; ; ; in, In the first The first in the region The landscape feature tensor of landscape features For the first Global landscape vector of each region For the first The landscape hash code of each region For activation function, For the weight vector, For bias vectors, For the first Region 1 Measured values ​​of several landscape feature parameters For symbolic functions, For the first The normal vector of the landscape segmentation line corresponding to each hash code. For the first The intercept of the landscape segmentation line corresponding to each hash code. This is the sequence number of the hash code. This is the minimum valid hash code length.

[0021] It should be explained that the landscape feature tensor has a dimension of 2*1, the landscape feature parameter matrix has a dimension of 3*4, the weight vector has a dimension of 2*1, including three sets of weight vectors corresponding to the three feature parameters, and the bias vector has a dimension of 2*1, including three sets of bias vectors corresponding to the three feature parameters. The specific method for obtaining the weight vector and bias vector is as follows: construct a landscape feature triplet dataset based on the landscape feature types in historical cases, use the triplet loss as the optimization objective, minimize the distance between the anchor point and the positive sample and maximize the distance between the anchor point and the negative sample, where the anchor point is the current case, the positive sample is a similar case, and the negative sample is a dissimilar case. The final weight vector and bias vector are obtained through the optimization objective. The global landscape vector has a dimension of 2*1, and each region contains a global landscape vector. The components are: landscape feature tensor, landscape feature parameter matrix, weight vector, bias vector, and global landscape. The units of the measured values ​​of vectors and landscape feature parameters are dimensionless; the minimum effective hash code length is the minimum number of hash codes required, specifically obtained by selecting the optimal value based on the retrieval coverage rate, where the retrieval coverage rate is the ratio of the number of retrievals to the total number of retrievals. If the current retrieval coverage rate is less than 0.5, the maximum number of hash codes is selected as the minimum effective hash code length; the dimension of the normal vector of the landscape segmentation line is 2*1, and different regions correspond to different normal vectors. Specifically, the normal vectors all follow a standard normal distribution, and the normal vectors corresponding to different regions can be generated through the standard normal distribution and the Mason rotation algorithm; the intercept of the landscape segmentation line is a parameter, and different regions correspond to different intercepts. Specifically, the intercepts all follow a uniform distribution from negative one to positive one, and the intercepts corresponding to different regions can be generated through the uniform distribution from negative one to positive one and the Mason rotation algorithm.

[0022] Crack data for different areas of the facade is obtained using a crack width measuring instrument, and hollow data for different areas of the facade is obtained using infrared imaging. The crack data includes the maximum width and maximum depth of cracks in different areas, and the hollow data includes the maximum bulge height and hollow area in different areas. The degree of crack damage and the degree of hollow damage in different areas are calculated based on the crack data and hollow data, respectively. Structural damage feature vectors for different areas are constructed based on the degree of crack damage and hollow damage. The structural damage feature vectors are feature vectors generated by cracks and hollows in different areas of the facade. The structural damage feature vectors of different areas are binary encoded using local sensitive hashing to obtain structural hash codes for different areas. The structural hash codes are binary sequences of the structural damage feature vectors of all areas. The calculation formulas for the degree of crack damage, the degree of hollow damage, the structural damage feature vectors, and the structural hash codes are as follows: ; ; ; ; in, For the first The extent of crack damage in each area For the first The degree of hollow damage in each area For the first Structural damage feature vectors of each region For the first The structural hash code of each region, For the first Maximum crack width in each region This represents the critical threshold for crack width. For the first Maximum crack depth in each region This represents the critical threshold for crack depth. For the first The area of ​​hollow spots in each region This represents the critical threshold for the area of ​​voids. For the first The maximum height of the uplift in each area The critical threshold for the height of the bulge. As the first weight, As the second weight, As the third weight, As the fourth weight, For the first The normal vector of the structural dividing line corresponding to each hash code. For the first The intercept of the structural dividing line corresponding to each hash code.

[0023] It should be explained that the specific method for obtaining the critical threshold for crack width is determined according to the classification provisions of the current national standards regarding the impact of crack width on structural safety in concrete structures, with the unit being millimeters; the specific method for obtaining the critical threshold for crack depth is based on the General Specifications for Appraisal and Reinforcement of Existing Buildings, with the unit being millimeters; the specific method for obtaining the critical threshold for hollow area is to pre-train the materials of the current building facade to obtain the maximum hollow area of ​​the current building that has experienced detachment as the critical threshold for hollow area; the first weight and the second weight are obtained by the ratio of the number of cases with different crack damage types in the fixed total number of cases to the total number of cases, specifically obtained by: based on 100 historical cases The ratio of the number of cases where facade damage was caused by crack width to the total number of cases is used as the first weight; the ratio of the number of cases where facade damage was caused by crack depth to the total number of cases is used as the second weight; the third and fourth weights are obtained by the ratio of the number of cases with different damage types in a fixed total number of cases to the total number of cases. Specifically, the ratio of the number of cases where facade damage was caused by hollow area to the total number of cases in 100 historical cases is used as the third weight; the ratio of the number of cases where facade damage was caused by bulge height to the total number of cases is used as the fourth weight; the normal vector and intercept of the structural dividing line are obtained in the same way as those of the landscape dividing line.

[0024] Based on the landscape hash codes, structural hash codes, and cases in the restoration case database for different regions, the landscape hash distance and structural hash distance are calculated using the Hamming distance method. A preset distance threshold is used. The restoration case database is then searched based on the landscape hash distance, structural hash distance, and the distance threshold. The search criteria are: if both the landscape hash distance and structural hash distance are less than or equal to the distance threshold, the current restoration case is selected, and a first set of similar cases is obtained. A preset similarity quantity threshold is also used to filter the first set of similar cases, resulting in a second set of similar cases. If the number of cases in the first set is less than the similarity quantity threshold, the distance threshold is adjusted. The formulas for calculating the landscape hash distance and structural hash distance are as follows: ; ; in, For renovation cases With the The landscape hash distance of each region For renovation cases The appearance hash code, For the first The landscape hash code of each region For renovation cases With the The structural hash distance of each region For renovation cases The structure hash code, For the first The structural hash code of each region, This is the sequence number of the hash code. This is the minimum valid hash code length.

[0025] It should be explained that the landscape hash distance is the Hamming distance between the landscape hash codes of the restoration case and the target area, and the structure hash distance is the Hamming distance between the structure hash codes of the restoration case and the target area. The preset distance threshold is a distance threshold used to filter the landscape hash distance and structure hash distance. Specifically, it is obtained by performing a grid search on the hash codes according to the minimum effective hash code length, with the distance range being 0 to 5. The maximum distance obtained through the case accuracy is used as the distance threshold. The specific method for obtaining the preset similarity number threshold is the same as that for the feature parameter type. If the number of cases in the first similar case set is less than the similarity number threshold, the distance threshold is adjusted by incrementing the distance threshold by one.

[0026] A maintenance knowledge graph is constructed based on a repair case database. The maintenance knowledge graph consists of case nodes, feature nodes, damage nodes, and repair nodes from different repair cases as a set of nodes. The relationships between nodes are obtained from the repair case database as an edge set between the node sets. Cases from a second set of similar cases are injected into the maintenance knowledge graph as activation nodes. A preset edge number threshold is set, and a local maintenance subgraph is constructed by obtaining a set of nodes starting from the activation node and having an edge number less than or equal to the edge number threshold. The local maintenance subgraph includes case nodes, feature nodes, damage nodes, and repair nodes that have similar damage to the current second set of similar cases.

[0027] It needs to be explained that the knowledge graph is constructed by connecting the set of nodes and the set of associated edges; the case nodes are obtained by traversing each case in the repair case database and creating a unique case node for each case; the preset edge number threshold is obtained by performing a grid search on the compensation. When the expansion step size is 1, only the first-level neighbor nodes directly connected to the active node can be obtained, and the constructed subgraph only contains the direct association information of the active node itself; when the expansion step size is 2, the active node is reached from the associated edge to the first-level neighbor node, and then the first-level neighbor node is reached from the associated edge to the second-level neighbor node. The second-level neighbor nodes include other case nodes associated with cases in the second similar case set, so a step size of 2 is chosen.

[0028] Based on the original data of the facades of historical buildings, a facade benchmark parameter matrix is ​​constructed. The facade damage index for different areas is calculated based on this matrix. A comprehensive damage assessment vector for different areas is constructed based on the facade damage index, crack damage degree, and hollowing damage degree. Initial screening of repair nodes is performed on local maintenance sub-maps to obtain initial local maintenance sub-maps. The matching degree of different repair techniques for different areas is calculated based on the comprehensive damage assessment vector and the global facade vector. The initial local maintenance sub-maps are then further screened based on the matching degree and sorted in descending order to obtain a maintenance and repair set. This maintenance and repair set represents the set of repair techniques most suitable for the current facade area. A minimum quantity threshold is preset, and the maintenance and repair set is judged based on this minimum quantity threshold to obtain a candidate maintenance set. The formulas for calculating the facade damage index, comprehensive damage assessment vector, and matching degree are as follows: ; ; ; in, For the first Area use of number The degree of matching between the repair techniques For the first The landscape damage index of the area For the first Comprehensive damage assessment vector for the region For example The Comprehensive damage assessment vector for the region This refers to the case node number. To maintain the initial screening sub-graph, the first step was to use the second step. A collection of nodes in the repair process. For bandwidth parameters, The square of the Euclidean distance. For the first Global landscape vector of the region For the first A global landscape vector of historical cases. For the first The extent of crack damage in each area For the first The degree of hollow damage in each area The first in the landscape reference parameter matrix Region 1 The baseline values ​​for each landscape feature parameter, These are the landscape feature parameters, including color feature parameters, line proportion parameters, and material feature parameters, with a value of 3; For the first Region 1 Measured values ​​of several landscape feature parameters For the first Feature weights of each landscape feature parameter This is the first protection zero constant.

[0029] It needs to be explained that the specific method for obtaining the minimum number threshold is as follows: traverse each case node in the knowledge graph, starting from the current node, move one step along the associated edge to reach the damaged node and feature node directly associated with the current node, then move one step along the associated edge from the damaged node and feature node to reach all associated repair nodes, count the cumulative number of different repair nodes visited in this process, and take the minimum value of the statistical results for all case nodes as the minimum number threshold; the specific determination process for the candidate maintenance set is as follows: compare the index in the maintenance and repair set with the minimum number threshold, and obtain the elements whose index is less than or equal to the minimum number threshold to construct the candidate maintenance set; the matching degree and the appearance damage index are dimensionless parameters; the appearance benchmark parameter matrix is ​​the original appearance feature parameters of the current historical appearance building facade, the appearance damage index is the comprehensive deviation between the current appearance features and the benchmark appearance features of different regions; the matching degree is the matching degree of the current region with the repair process. The degree of matching is as follows: the comprehensive damage assessment vector is a vector for assessing the degree of damage to the facade in different areas; the candidate maintenance set is a set of maintenance and repair methods adapted to the current facade area; the specific method for obtaining the benchmark values ​​of the color feature parameters and the line-angle ratio parameters in the style feature parameters is to obtain them through the calculation formulas of different style feature parameters, and change the parameters in the formulas to the standard parameters of the original building facade; the measured values ​​of the material feature parameters are obtained directly through spectral testing of the original materials of the same type of historical buildings; the specific method for obtaining the bandwidth parameter is to calculate the Euclidean distance between the comprehensive damage assessment vectors of any two samples from the historical case database, and then calculate the average value of all distances as the bandwidth parameter; the specific method for obtaining the feature weights is that the feature weights of the three types of style features are the same and sum to one, and the feature weights of the three types of style features are obtained by averaging; the specific method for obtaining the first prevention and control zero constant is to fix it to a value of 10 to the power of -8.

[0030] Material nodes are removed from the local maintenance subgraph to obtain the first subgraph, which consists of case nodes, repair nodes, feature nodes, and damage nodes in the local maintenance subgraph. All repair nodes are traversed, and repair nodes that meet the screening criteria are selected. The screening criteria are: the current repair node must be connected to both feature nodes and damage nodes. The case nodes corresponding to all repair nodes that meet the screening criteria are used as historical case tags for the current repair node. The historical case tags of the current repair node are historical cases that represent the current repair process. All repair nodes with historical case tags are collected as the initial screening subgraph for local maintenance.

[0031] Based on historical cases, predicted residual values ​​are obtained for different regions using different candidate maintenance schemes from the candidate maintenance set. These predicted residual values ​​represent the residual values ​​of the landscape feature parameters after restoration in different regions. Landscape fidelity is calculated based on the predicted residual values. The maintenance costs of different candidate maintenance schemes are obtained. Multi-objective optimization is then performed on the current candidate maintenance set based on landscape fidelity and maintenance costs to obtain the optimal maintenance scheme. The formulas for calculating landscape fidelity and multi-objective optimization are as follows: ; ; in, To adopt The fidelity of the design's appearance. For the first The candidate maintenance set of the region One element, For the first Region 1 Measured values ​​of several landscape feature parameters For the renovation of the first Region 1 Predicted residual values ​​of each landscape feature parameter The second prevention is zero constant. For multi-objective optimization, the optimal candidate maintenance scheme, As the first optimization weight, As the second optimization weight, To adopt The maintenance cost of the solution The highest cost among the candidate maintenance sets, To obtain the position index of the maximum value.

[0032] It should be explained that the method for obtaining the predicted residual value is as follows: For each element in the candidate maintenance set, find all case nodes in the local maintenance subgraph of the maintenance knowledge graph that have historically adopted the current candidate maintenance scheme, calculate the change in each landscape feature parameter before and after the repair as the improvement amount, take the arithmetic mean of the improvement amount of each landscape feature parameter of all historical cases to obtain the average improvement amount of the candidate maintenance scheme in each landscape feature dimension, and subtract the corresponding average improvement amount from the current measured value to obtain the predicted residual value; the landscape fidelity is the repair effect of the candidate maintenance scheme on the current landscape damage; the specific optimization weights and second optimization weights are... The method for obtaining the body is as follows: multiple sets of different first and second optimization weights are used to perform backtracking optimization calculations on these cases, the calculation results are compared with the actual selection results, and the set of weights with the highest consistency rate with the actual selection results is selected as the calibrated first and second optimization weights, and the sum of the first and second optimization weights is one; the method for obtaining the maintenance cost is as follows: read the unit area cost corresponding to the current solution from the repair node of the maintenance knowledge graph, multiply it by the maintenance area of ​​the current region, and obtain the total cost of the current solution in the current region; the specific method for obtaining the second prevention and removal zero constant is to fix it at a value of 10 to the power of negative 8.

[0033] Example 2: Based on Example 1, a building facade maintenance system based on multi-information fusion further includes: The data acquisition module collects point cloud data and image data of historical buildings, divides the facades of historical buildings into different areas, and constructs a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. The hash index module calculates the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtains the landscape hash code of different regions based on the landscape feature tensor of different regions; it constructs the structural damage feature vector of different regions, and obtains the structural hash code of different regions through the structural damage feature vector of different regions. The knowledge graph module obtains the first set of similar cases from the repair case database based on the appearance hash code and the structure hash code. It then filters the first set of similar cases by setting a preset similarity threshold to obtain the second set of similar cases. Based on the repair case database, it constructs and maintains a knowledge graph. Based on the maintenance knowledge graph and the second set of similar cases, it constructs a local maintenance subgraph for the current region. The candidate scheme module calculates the landscape damage index for different regions, obtains the comprehensive damage assessment vector for different regions based on the landscape damage index, and filters the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. The optimal solution module calculates the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtains the optimal maintenance solution through multi-objective optimization.

[0034] Although the invention has been described with reference to exemplary embodiments, it should be understood that the invention is not limited to the disclosed exemplary embodiments. The scope of the following claims should be given the broadest interpretation so as to cover all variations and equivalent structures and functions.

Claims

1. A method for maintaining building facades based on multi-information fusion, characterized in that, Includes the following steps: S1: Collect point cloud data and image data of historical buildings, divide the facades of historical buildings into different areas, and construct a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. S2: Calculate the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtain the landscape hash code of different regions based on the landscape feature tensor of different regions; construct the structural damage feature vector of different regions, and obtain the structural hash code of different regions through the structural damage feature vector of different regions. S3: Obtain the first set of similar cases from the renovation case database based on the appearance hash code and structure hash code, filter the first set of similar cases by preset similarity threshold, and obtain the second set of similar cases; A maintenance knowledge graph is constructed based on the repair case database, and a local maintenance subgraph for the current area is constructed based on the maintenance knowledge graph and the second set of similar cases. S4: Calculate the landscape damage index for different regions, obtain the comprehensive damage assessment vector for different regions based on the landscape damage index, and filter the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. S5: Calculate the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtain the optimal maintenance plan through multi-objective optimization.

2. The building facade maintenance method based on multi-information fusion according to claim 1, characterized in that, The process involves collecting point cloud data and image data of historical buildings, dividing the facades of these buildings into different regions, and constructing a feature parameter matrix and a restoration case database based on the point cloud data, image data, and different regions. This includes: acquiring point cloud data, texture data, and image data of historical buildings using scanners and multispectral imaging technology; the point cloud data being spatial coordinates and color information in a spatial coordinate system constructed with the center point of the current facade; the image data being reflectance vectors in spectral bands; and the texture data being the texture of the facade in the current spatial coordinates; dividing the facades of historical buildings into regions based on a preset number of regions; acquiring color feature parameters, line-angle ratio parameters, and material feature parameters for each corresponding facade region based on the point cloud data, image data, texture data, and corresponding facade regions; constructing a feature parameter matrix for this region based on the color feature parameters, line-angle ratio parameters, and material feature parameters; and constructing a restoration case database for the current historical buildings based on their restoration records. The restoration case database contains historical restoration records of the facades of the current historical buildings.

3. The building facade maintenance method based on multi-information fusion according to claim 1, characterized in that, The step of calculating the landscape feature tensor of different regions based on the landscape feature parameter matrix and obtaining the landscape hash code of different regions based on the landscape feature tensor of different regions includes: mapping the landscape feature parameter matrix into a unified landscape feature tensor through feature-by-feature independent nonlinear embedding, wherein the landscape feature tensor is a parameter vector that uniformly represents different landscape features of different regions; performing feature aggregation on each region based on the landscape feature tensor; obtaining the global landscape vector of different regions by using max pooling, wherein the global landscape vector is the dominant semantic feature vector of facade landscape damage in the current region; and binarizing the global landscape vector of different regions through local sensitive hashing to obtain the landscape hash code of different regions, wherein the landscape hash code is a binary sequence of the global landscape vector of all regions, wherein the calculation formulas for the landscape feature tensor, the global landscape vector, and the landscape hash code are as follows: ； ； ； in, In the first The first in the region The landscape feature tensor of landscape features For the first Global landscape vector of each region For the first The landscape hash code of each region For activation function, For the weight vector, For bias vectors, For the first Region 1 Measured values ​​of several landscape feature parameters For symbolic functions, For the first The normal vector of the landscape segmentation line corresponding to each hash code. For the first The intercept of the landscape segmentation line corresponding to each hash code. This is the sequence number of the hash code. This is the minimum valid hash code length.

4. The building facade maintenance method based on multi-information fusion according to claim 1, characterized in that, The construction of structural damage feature vectors for different regions, and the acquisition of structural hash codes for different regions through these feature vectors, includes: acquiring crack data for different regions of the facade using a crack width measuring instrument; acquiring void data for different regions of the facade using infrared imaging; the crack data being the maximum crack width and maximum crack depth for different regions; and the void data being the maximum bulge height and void area for different regions; calculating the degree of crack damage and the degree of void damage for different regions based on the crack and void data; constructing structural damage feature vectors for different regions based on the degree of crack damage and the degree of void damage; and performing binary encoding on the structural damage feature vectors for different regions using local sensitive hashing to obtain structural hash codes for different regions. The structural hash codes are binary sequences of the structural damage feature vectors for all regions. The calculation formulas for the degree of crack damage, the degree of void damage, the structural damage feature vectors, and the structural hash codes are as follows: ； ； ； ； in, For the first The extent of crack damage in each area For the first The degree of hollow damage in each area For the first Structural damage feature vectors of each region For the first The structural hash code of each region, For the first Maximum crack width in each region This represents the critical threshold for crack width. For the first Maximum crack depth in each region This represents the critical threshold for crack depth. For the first The area of ​​hollow spots in each region This represents the critical threshold for the area of ​​voids. For the first The maximum height of the uplift in each area The critical threshold for the height of the bulge. As the first weight, As the second weight, As the third weight, As the fourth weight, For the first The normal vector of the structural dividing line corresponding to each hash code. For the first The intercept of the structural dividing line corresponding to each hash code.

5. A method for maintaining building facades based on multi-information fusion according to claim 1, characterized in that, The process of obtaining a first set of similar cases from a restoration case database based on landscape hash codes and structural hash codes, and then filtering the first set of similar cases based on a preset similarity threshold to obtain a second set of similar cases includes: calculating landscape hash distance and structural hash distance respectively using the Hamming distance calculation method based on landscape hash codes, structural hash codes, and cases in the restoration case database from different regions; setting a preset distance threshold; searching the restoration case database based on the landscape hash distance, structural hash distance, and distance threshold; the search condition is: if both the landscape hash distance and the structural hash distance are less than or equal to the distance threshold, then the current restoration case is selected to obtain the first set of similar cases; setting a preset similarity threshold, filtering the first set of similar cases based on the similarity threshold to obtain the second set of similar cases; if the number of cases in the first set of similar cases is less than the similarity threshold, then the distance threshold is adjusted; wherein the calculation formulas for landscape hash distance and structural hash distance are: ； ； in, For renovation cases With the The landscape hash distance of each region For renovation cases The appearance hash code, For the first The landscape hash code of each region For renovation cases With the The structural hash distance of each region For renovation cases The structure hash code, For the first The structural hash code of each region, This is the sequence number of the hash code. This is the minimum valid hash code length.

6. The building facade maintenance method based on multi-information fusion according to claim 1, characterized in that, The step of constructing a maintenance knowledge graph based on a repair case database and constructing a local maintenance subgraph for the current region based on the maintenance knowledge graph and a second set of similar cases includes: constructing a maintenance knowledge graph based on a repair case database, wherein the maintenance knowledge graph is a set of nodes of the maintenance knowledge graph consisting of case nodes, feature nodes, damaged nodes, and repair nodes in different repair cases; obtaining the association relationships between nodes based on the repair case database as an edge set between the node sets; injecting cases from the second set of similar cases as activation nodes into the maintenance knowledge graph; setting a preset edge number threshold; and constructing a local maintenance subgraph by obtaining a set of nodes starting from the activation node and having an associated edge number less than or equal to the edge number threshold. The local maintenance subgraph includes case nodes, feature nodes, damaged nodes, and repair nodes that have similar damage to the current second set of similar cases.

7. A method for maintaining building facades based on multi-information fusion according to claim 1, characterized in that, The process involves calculating the landscape damage index for different areas, obtaining a comprehensive damage assessment vector for different areas based on the landscape damage index, and filtering local maintenance sub-maps to obtain a candidate maintenance set. This includes: constructing a landscape benchmark parameter matrix based on original data of the historical building facades; calculating the landscape damage index for different areas based on the landscape benchmark parameter matrix; constructing a comprehensive damage assessment vector for different areas based on the landscape damage index, crack damage degree, and hollowing damage degree; performing initial screening of repair nodes on the local maintenance sub-maps to obtain initial screening sub-maps; calculating the matching degree of different repair techniques for different areas based on the comprehensive damage assessment vector and global landscape vector; refining the initial screening sub-maps based on the matching degree and sorting them in descending order of matching degree to obtain a maintenance and repair set. This maintenance and repair set represents the set of repair techniques most suitable for the current facade area. A minimum quantity threshold is preset, and the maintenance and repair set is judged based on the minimum quantity threshold to obtain a candidate maintenance set. The formulas for calculating the landscape damage index, comprehensive damage assessment vector, and matching degree are as follows: ； ； ； in, For the first Area use of number The degree of matching between the repair techniques For the first The landscape damage index of the area For the first Comprehensive damage assessment vector for the region For example The Comprehensive damage assessment vector for the region This refers to the case node number. To maintain the initial screening sub-graph, the first step was to use the second step. A collection of nodes in the repair process. For bandwidth parameters, The square of the Euclidean distance. For the first Global landscape vector of the region For the first A global landscape vector of historical cases. For the first The extent of crack damage in each area For the first The degree of hollow damage in each area The first in the landscape reference parameter matrix Region 1 The baseline values ​​for each landscape feature parameter, These are the landscape feature parameters, including color feature parameters, line proportion parameters, and material feature parameters, with a value of 3; For the first Region 1 Measured values ​​of several landscape feature parameters For the first Feature weights of each landscape feature parameter This is the first protection zero constant.

8. A method for maintaining building facades based on multi-information fusion according to claim 7, characterized in that, The initial screening of repair nodes in the local maintenance sub-graph to obtain the initial screening sub-graph includes: removing material nodes from the local maintenance sub-graph to obtain a first sub-graph, which consists of case nodes, repair nodes, feature nodes, and damaged nodes in the local maintenance sub-graph; traversing all repair nodes and filtering out repair nodes that meet the filtering criteria, which is that the current repair node must be connected to both feature nodes and damaged nodes; using the case nodes corresponding to all repair nodes that meet the filtering criteria as historical case tags for the current repair node, where the historical case tags represent historical cases of the current repair process; and collecting all repair nodes with historical case tags as the initial screening sub-graph for local maintenance.

9. A method for maintaining building facades based on multi-information fusion according to claim 1, characterized in that, The step of calculating the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtaining the optimal maintenance scheme through multi-objective optimization, includes: obtaining predicted residual values ​​for different areas based on different candidate maintenance schemes adopted in historical cases, wherein the predicted residual values ​​are the residual values ​​of landscape feature parameters after repair in different areas; calculating the landscape fidelity based on the predicted residual values; obtaining the maintenance cost of different candidate maintenance schemes based on different candidate maintenance schemes; and performing multi-objective optimization on the current candidate maintenance set based on the landscape fidelity and maintenance cost to obtain the optimal maintenance scheme. The calculation formulas for landscape fidelity and multi-objective optimization are as follows: ； ； in, To adopt The fidelity of the design's appearance. For the first The candidate maintenance set of the region One element, For the first Region 1 Measured values ​​of several landscape feature parameters For the renovation of the first Region 1 Predicted residual values ​​of each landscape feature parameter The second prevention is zero constant. For multi-objective optimization, the optimal candidate maintenance scheme is... As the first optimization weight, As the second optimization weight, To adopt The maintenance cost of the solution The highest cost among the candidate maintenance sets, To obtain the position index of the maximum value.

10. A building facade maintenance system based on multi-information fusion, used to implement the building facade maintenance method based on multi-information fusion as described in claims 1-9, further comprising: The data acquisition module collects point cloud data and image data of historical buildings, divides the facades of historical buildings into different areas, and constructs a feature parameter matrix and a renovation case database based on the point cloud data, image data and different areas. The hash index module calculates the landscape feature tensor of different regions based on the landscape feature parameter matrix, and obtains the landscape hash code of different regions based on the landscape feature tensor of different regions; it constructs the structural damage feature vector of different regions, and obtains the structural hash code of different regions through the structural damage feature vector of different regions. The knowledge graph module obtains the first set of similar cases from the renovation case database based on the appearance hash code and the structure hash code. It then filters the first set of similar cases by setting a threshold for the number of similar cases to obtain the second set of similar cases. A maintenance knowledge graph is constructed based on the repair case database, and a local maintenance subgraph for the current area is constructed based on the maintenance knowledge graph and the second set of similar cases. The candidate scheme module calculates the landscape damage index for different regions, obtains the comprehensive damage assessment vector for different regions based on the landscape damage index, and filters the local maintenance sub-maps based on the comprehensive damage assessment vector to obtain the candidate maintenance set. The optimal solution module calculates the landscape fidelity and maintenance cost based on the candidate maintenance set, and obtains the optimal maintenance solution through multi-objective optimization.

Images