A modeling method for assisting classical garden restoration

Abstract

This invention belongs to the field of digital conservation of classical gardens and discloses a modeling method for assisting in the restoration of classical gardens. The method includes a basic data establishment step, a real-time data input step, and a restoration decision support step. The basic data establishment step involves: establishing a restoration knowledge graph; obtaining finite element models of components identified by component IDs; establishing and training a disease identification model, which outputs component ID, disease type, disease level, and disease location based on component image data. The real-time data input step involves: acquiring component image data and inputting it into the disease identification model; establishing a component state model based on the output of the disease identification model and the component finite element model; and outputting deformation data, physical parameter change data, and risk level from the component state model. The restoration decision support step involves: determining whether the component needs restoration based on the output of the component state model and assisting in restoration based on the restoration knowledge graph.

Description

Technical Field

[0001] This invention belongs to the field of digital conservation of classical gardens, and specifically relates to a modeling method for assisting in the restoration of classical gardens. Background Technology

[0002] my country is an ancient civilization with a long history and rich cultural resources, of which classical gardens are a very important part. These garden masterpieces, embodying the wisdom and aesthetics of the ancients, have withstood the vicissitudes of time and face the dual challenges of environmental erosion and human-caused damage in daily life. Traditional artificial restoration techniques are centered on manual inspection. This involves multiple experts carrying inspection tools entering the garden to conduct on-site inspections, checking each component one by one to confirm the erosion and / or damage status of each component and forming a corresponding component condition report. Then, based on their personal experience and the component condition reports, multiple experts jointly determine the restoration materials and restoration techniques. Next, professional construction workers carry out restoration work on the classical garden based on the determined restoration materials and techniques. Finally, multiple experts conduct sensory inspection to accept and confirm the restoration results.

[0003] In recent years, with the development of digital technology, accurate preservation and permanent backup of garden heritage information have been achieved. Intelligent monitoring and scientific decision-making can enhance the professionalism and efficiency of protection and restoration, realizing "equal emphasis on protection and utilization, and symbiosis of inheritance and innovation", thus injecting new vitality into the protection of garden cultural heritage.

[0004] Currently, based on the above, a digital-based approach to garden conservation has emerged: First, several staff members use handheld scanners to scan the garden's components to create a data model. Then, several experts mark the locations of erosion and / or damage on the data model, specifying the exact state of erosion and / or damage. Next, based on the marked data model and their personal experience, the experts determine the restoration materials and techniques. Then, professional construction workers restore the classical garden using the restoration materials and techniques. Finally, several experts confirm the restoration results through sensory inspection.

[0005] Clearly, both of the above methods rely on expert groups to confirm the erosion and / or damage status for each restoration. This makes the restoration method highly dependent on the experience of experts. Moreover, the decision-making process needs to take into account many complex factors such as environment, culture, human factors and the integrity of the components, which makes the restoration decision-making cumbersome and inefficient. Furthermore, the latter has many independent data sources and lacks effective collaboration, which limits the efficiency improvement of related garden restoration work. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a modeling method for assisting in the restoration of classical gardens. By using model assistance, the reliance on experts in the restoration of classical gardens can be effectively reduced, and effective collaboration of multiple data sources can be achieved.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A modeling method for assisting in the restoration of classical gardens includes the following steps: Basic data establishment steps: Obtain restoration record data of classical gardens and establish a restoration knowledge graph based on the restoration record data; reconstruct the garden components in three dimensions to obtain finite element models of the components, with the component ID as the identifier; establish and train a disease identification model, which outputs the component ID, disease type, disease level, and disease location based on the component image data; Real-time data input steps: Acquire component image data of garden components, input component image data into the disease identification model, establish component state model based on the output results of the disease identification model and the component finite element model, and output deformation data, physical parameter change data and risk level of the component state model; Repair decision support steps: The output of the component state model determines whether the component needs to be repaired. If repair is required, the repair is assisted by a repair knowledge graph.

[0008] Preferably, the basic data establishment step further includes: The system acquires image data of classical gardens and extracts multiple garden component datasets based on the image data. The garden component datasets include component shape data and relative position data between components. The component finite element model is associated with the garden component dataset through the component ID. In the real-time data input step, the component state model also outputs contour variation data and position variation data between components. Furthermore, the image data of classical gardens consists of overall photographs of classical gardens taken by drones, while the restoration record data of classical gardens is obtained by scanning historical restoration records and ancient books that record the restoration of gardens.

[0009] Furthermore, the node types of the repair knowledge graph include index nodes and solution nodes. Index nodes include solution ID, component type, defect type, repair time, repair materials, repair method, and effect score. Solution nodes include solution ID, material list, repair steps, cost budget, shape change, position change, and expected repair effect. The effect score is obtained based on the cost budget, shape change, position change, and expected repair effect.

[0010] Furthermore, the effectiveness score = effectiveness score * 0.4 + cost score * 0.4 + cultural score * 0.3. The effect score is based on the expected restoration effect; the cost score is based on the difference between the cost budget and the preset standard budget; and the cultural score is based on the image data, shape changes, and location changes of the classical garden. Preferably, in the repair decision support step, when any one of the following conditions is met—the risk level is a predetermined level, the deformation is greater than the preset deformation, or the change in physical parameters is greater than the preset change in parameters—the component needs to be repaired. Based on the output of the disease identification model, the repair knowledge graph is associated, and at least one garden repair scheme is selected.

[0011] Furthermore, the present invention also includes an effect verification step: Obtain component environmental data within a predetermined period before the restoration of the classical garden; construct expected component status data based on the selected garden restoration plan and component environmental data; obtain component restoration status data after restoration; and obtain restoration effect score by comparing component restoration status data and expected component status data.

[0012] Furthermore, if the repair effect score is greater than or equal to the preset qualified score, it is determined as "repair successful"; if the repair effect score is less than the preset qualified score, an improved plan is manually formulated and the effect verification step is repeated.

[0013] Preferably, when the repair effect score of the corresponding improved solution is greater than or equal to the preset qualified score, it is determined as "repair successful", and a new index node and solution node are added to the repair knowledge graph based on the improved solution.

[0014] A storage medium for assisting in the restoration of classical gardens, wherein a processing program is stored thereon, characterized in that the processing program, when executed by a processor, implements the aforementioned modeling method for assisting in the restoration of classical gardens. Compared with the prior art, the beneficial effects of the present invention are: 1. Because the modeling method for assisting in the restoration of classical gardens in this invention includes a basic data establishment step, a real-time data input step, and a restoration decision support step, the basic data establishment step involves: establishing a restoration knowledge graph based on restoration record data of classical gardens; obtaining finite element models of garden components through three-dimensional reconstruction, with component IDs as identifiers; establishing and training a disease identification model, which outputs component IDs, disease types, disease levels, and disease locations based on component image data; the real-time data input step involves: acquiring component image data of garden components, inputting the component image data into the disease identification model, establishing a component state model based on the output results of the disease identification model and the component finite element models, and the component state model outputting deformation data, physical parameter change data, etc. The invention includes risk levels and repair decision support steps: Based on the output of the component state model, it determines whether the component needs repair. If repair is required, repair is assisted by a repair knowledge graph, which contains previous garden restoration cases that can be referenced. The component finite element model contains information such as the material, physics, pose, distribution rules (specific shape and position relationships between components), and surface imprints of multiple garden components in classical gardens. The disease identification model uses photos of garden components with diseases as a training set. Through expert annotation, the disease identification model can simulate experts' judgment on the disease of components. Therefore, this invention can effectively reduce the dependence of classical garden restoration work on experts through model assistance and achieve effective collaboration of multiple data sources.

[0015] 2. Because the image data of the classical garden in this invention is an overall photograph of the classical garden taken by a drone, and the restoration record data of the classical garden is obtained by scanning historical restoration record documents and ancient books recording garden restoration, the image data of the classical garden is the overall layout of multiple classical gardens, used to obtain the integrity information, coordination information, distribution pattern information and concrete or abstract symbolism between multiple components and in the environment, so as to reflect the cultural and historical connotation value of the classical garden. The restoration knowledge graph of historical restoration record documents and ancient books recording garden restoration provides case references in ancient books and record data. Therefore, the auxiliary model of this invention can further consider the historical and cultural value of the restored garden in the process of assisting restoration, so that the classical garden still has extremely high cultural value in the long term.

[0016] 3. Since the effectiveness score of this invention = effectiveness converted score * 0.4 + cost converted score * 0.4 + cultural converted score * 0.3, the effectiveness converted score is obtained based on the expected restoration effect; the cost converted score is obtained based on the difference between the cost budget and the preset standard budget; and the cultural converted score is obtained based on the image data, shape change and location change of the classical garden, this invention provides a quantifiable evaluation of the restoration case based on effectiveness, cost and culture (value preservation).

[0017] 4. Because the present invention also includes an effect verification step: acquiring component environmental data within a predetermined period before the restoration of the classical garden, constructing expected component state data based on the selected garden restoration plan and component environmental data, acquiring component restoration state data after restoration, and comparing the component restoration state data and expected component state data to obtain a restoration effect score. Specifically, the component restoration state data is a digital twin model of the component based on the fusion of the component's image data and restoration plan data, which is represented by AR. Therefore, the present invention can provide an intuitive and visual effect preview of the restoration effect of the classical garden under the influence of environmental factors while selecting a plan.

[0018] 5. When the repair effect score is greater than or equal to the preset qualified score, the auxiliary model of this invention determines that the repair is "successful"; if the repair effect score is less than the preset qualified score, an improved plan is manually formulated and the effect verification step is re-executed. Specifically, when the score is less than the preset qualified score, the expert group needs to manually analyze the reasons for the insufficient score based on the component repair status data, and then propose an improved plan for re-verification of the effect. Therefore, when the repair knowledge graph cannot provide effective repair reference, this invention allows for the proposal of improved repair plans through the access of experts. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the steps of a modeling method for assisting in the restoration of classical gardens, as described in an embodiment of the present invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the modeling method of this invention for assisting in the restoration of classical gardens. It should be noted that the description of these embodiments is for the purpose of helping to understand this invention, but does not constitute a limitation of this invention.

[0021] like Figure 1 As shown, the modeling method for assisting in the restoration of classical gardens in this embodiment includes a basic data establishment step, a real-time data input step, a restoration decision support step, and an effect verification step.

[0022] Basic data establishment steps: The process involves acquiring image data of classical gardens, extracting multiple garden component datasets based on the image data, and including component shape data and relative position data between components. Specifically, a particular classical garden is decomposed into a combination of multiple components (such as carved columns, artificial rocks, painted decorations, plaques, couplets, etc.), thereby restoring the classical garden based on the restoration of one or more garden components.

[0023] The image data of the classical garden is an overall photo of the classical garden taken by a drone. In this embodiment, a DJI drone is used, with the flight altitude set to 50-80m (adjusted according to the size of the garden) and the overlap set to 80% (to ensure the stitching accuracy between images) to capture an overall layout image of the classical garden, with a resolution of 5472×3648 pixels.

[0024] The restoration records of classical gardens were acquired, and a restoration knowledge graph was built based on these records. The restoration records of classical gardens were obtained by scanning historical restoration documents and ancient books that recorded the restoration of gardens.

[0025] In this embodiment, an Epson scanner is used to scan historical restoration documents of the garden (such as restoration reports, drawings, and material lists) and ancient records (such as relevant construction records in "Yuanye" and "Yingzao Fashi") at a resolution of 600 dpi. At the same time, key information (such as restoration time (e.g., 1998), materials, and processes) is extracted from the documents using OCR technology.

[0026] The node types for repairing the knowledge graph include index nodes and solution nodes. Index nodes include solution ID, component type, defect type, repair time, repair materials, repair method, and effect score. Solution nodes include solution ID, material list, repair steps, cost budget, shape change, position change, and expected repair effect.

[0027] Specifically, the types of components include columns, beams, lintels, purlins, rafters, brackets, stone steps, column bases, brick walls, roof tiles, hanging ornaments, railings, painted decorations, wood carvings, plaques and couplets, roof boards, eaves, and tiles…; the types of defects include decay, insect infestation, cracking, weathering, efflorescence, efflorescence, fissures, hollowing of the base layer, peeling and flaking of painted decorations, powdering of oil finishes, leakage, and biological damage…; the repair materials include pine, fir, cypress, old elm, blue bricks, mortar bricks, barrel tiles, flat tiles, and lime. Mortar, glutinous rice mortar, tung oil, traditional raw lacquer, mineral pigments, carbon fiber cloth (for reinforcement), environmentally friendly agents (for insect prevention)... Repair methods include cleaning, sealing, hand carving, mortise and tenon repair, mortar removal and spot treatment, brick carving repair, one hemp and five mortar, traditional painting, gilding... In this embodiment, the selection of repair materials follows the principle of "original form, original structure, original raw materials, and original process", and the selection of repair methods takes into account both interventional repair and protective repair.

[0028] In this embodiment, the index node is in the form of (001-Column-Wood Rot-1998-Pine Wood-Hand Repair-Good Results), and the matching scheme node is in the form of: Materials list: Grade 2 wood rot column, made of pine wood (moisture content ≤15%) + traditional tung oil (3 coats), specifying material specifications (e.g., pine wood diameter 300mm, length 3m), usage (e.g., pine wood 0.2m³, tung oil 5L), and supplier.

[0029] Process steps: such as "Step 1: Manually clean the rotten parts (down to the healthy wood)" Step 2: Apply tung oil (first coat dries for 24 hours, second coat for 48 hours, third coat for 72 hours) Step 3: Hand sanding (400# sandpaper) Step 4: Apply topcoat (traditional raw lacquer).

[0030] Cost budget: Calculate material costs (e.g., pine wood 1500 yuan / m³, tung oil 80 yuan / L), labor costs (e.g., manual repair 200 yuan / hour), and equipment costs (e.g., small sander 50 yuan / day). The total budget deviation should be ≤5%.

[0031] Expected results: such as "the moisture content of the repaired wood will be stabilized at 12%-15%, the disease level will not be upgraded within 5 years, and the structural load-bearing capacity will be restored to 90% of the original design".

[0032] Shape variation: such as "deformation of curved edges".

[0033] Location change: such as "displacement exceeding 1 meter along the north-south direction".

[0034] Index nodes and scheme nodes are linked by scheme ID. The effect score in the index node is obtained from the cost budget, shape change, position change and expected repair effect in the scheme node. The effect score = effect conversion score * 0.4 + cost conversion score * 0.4 + cultural conversion score * 0.3. The effect conversion score is based on the expected repair effect; the cost conversion score is based on the difference between the cost budget and the preset standard budget; and the cultural conversion score is based on the image data, shape change and position change of the classical garden.

[0035] In this embodiment, the historical restoration data of the gardens were collected from 100 classical gardens across the country (1980-2023). Through manual annotation and NLP automatic extraction, a knowledge graph containing 5,000 nodes and 12,000 edges was constructed and stored in the Neo4j graph database.

[0036] The 3D reconstruction of garden components yields finite element models of the components. These finite element models are identified by component IDs and are associated with the garden component dataset through these component IDs. Specifically, the component IDs are in the form of ("East Wing-Column-001").

[0037] Specifically, a Faro Focus S150 laser scanner was used, with a scanning accuracy set to ±0.1mm and a point cloud density set to 100-150 points / mm² (the upper limit was taken for core components). The scanning range covered the target area (such as the beams and columns of a pavilion) to form a finite element model of the component as a reference model for subsequent repairs. The component ID was used as a matching identifier.

[0038] Establish and train a disease identification model. The disease identification model outputs the component ID, disease type, disease level, disease location, and disease depth (the depth of surface indentation caused by the disease) based on the component image data.

[0039] Specifically, the damage levels are categorized as "Intact," "Minor," "Moderate," "Severe," and "Dangerous." The location of the damage is based on the component ID. For example, the damage location corresponding to the "dougong" component ID includes the mortise and tenon joint, the area near the mortise and tenon joint, and the remaining parts. The damage location corresponding to the "zhu" component ID includes the column head, the area near the column head, and the column body. The damage location corresponding to the "wood carving" component ID includes the relief pattern area and other areas. The damage level is based on a comparison of the color and texture of the damaged and normal parts, as well as learning the location of the damage. For example, Minor: The remaining parts have insect bite area <10%–25%, with a few chips and insect droppings. Basically intact; Severe: The area near the mortise and tenon joint has insect bite area of ​​25%–50%–70%, with obvious damage and localized yellowing; Dangerous: The mortise and tenon joint has insect bite area of ​​25%–50%–70%, with obvious damage and localized yellowing.

[0040] In this embodiment, the disease identification model is based on "improved ResNet50 + attention mechanism" and is optimized for the complex textures of classical garden components. An SE (Squeeze-and-Excitation) attention module is added after the 3rd and 4th convolutional blocks of ResNet50 to enhance the model's ability to extract features from diseased areas (such as wood decay texture and stone cracks) and suppress background interference (such as carvings and paintings).

[0041] Specifically, ancient architecture experts used the "Extended Dataset of Diseases in Classical Chinese Gardens" (containing 20,000 images covering various garden components with different types of diseases in classical gardens) to train a disease identification model. In this embodiment, the training environment was a GPU Tesla V100 (32GB of video memory), the optimizer was Adam (learning rate 0.0001), the loss function was cross-entropy loss, the training epochs were 50, and the final model achieved an accuracy of ≥95% and a confidence coefficient of ≥0.91 on the test set.

[0042] Specifically, in the process of modeling to assist in the restoration of classical gardens, the collected data and models need to be preprocessed, including: raw data verification and data standardization. In this embodiment, raw data verification adopts a combination of "automatic verification + manual sampling"—automatic verification uses algorithms (such as image sharpness algorithms and integrity algorithms) to filter out abnormal data (such as images with a sharpness lower than 0.8 or areas with a missing rate of more than 5%); the manual sampling rate is 10%, with one expert in ancient architecture confirming the results of automatic verification to ensure that abnormal data are accurately marked and to avoid invalid data in subsequent processing; for image data standardization, Gaussian filtering (σ=0.8) is used to denoise the image data. Histogram equalization enhances contrast The process of "image normalization (uniform size to 224×224 pixels for easy model input)" is used to process the detailed images of components; aerial images are automatically stitched together using the SIFT algorithm to generate an overall orthophoto map of the garden (resolution 0.1m / pixel); and the standardized data for the model data is processed using the RANSAC algorithm to remove redundant points (1000 iterations, threshold 0.05mm). Poisson surface reconstruction generates 3D mesh model The process of "model lightweighting (preserving core structural details and controlling the number of faces to within 1 million to facilitate subsequent analysis)" is used. For standardized data, the structured text extracted by OCR is cleaned to remove redundant information (such as irrelevant annotations) from the literature data. Core information (the information needed to build index nodes and scheme nodes) is extracted through entity recognition algorithm (based on BERT model).

[0043] Real-time data input steps: The component image data of the garden components is obtained and input into the defect identification model. In this embodiment, a Sony high-definition camera (with a 90mm macro lens) is used to photograph the components (such as carved columns, artificial rocks, and painted decorations). The shooting distance is controlled at 0.5-1.5m to ensure clear texture. Each component is photographed from 3-5 angles (front, side, and top), and the resolution is set to 9504×6336 pixels.

[0044] Based on the output of the defect identification model and the component finite element model, a component state model is established. The component state model outputs deformation data, physical parameter change data, contour change data, inter-component positional change data, and risk level. Specifically, the component state model maps the defect status of the component onto the component finite element model, which serves as the reference. The physical parameter change data is obtained based on the component ID, defect type, defect location, defect level, defect depth, and deformation data. The risk level is established based on the deformation data, physical parameter change data, contour change data, and inter-component positional change data. In this embodiment, the defect level is calculated using a confidence formula. If it is less than the predetermined confidence level, it is marked as "pending manual confirmation" in the data, meaning that the output still needs to be manually confirmed by experts. Each output of the component state model requires a corresponding report (Excel + PDF), and the defect location and defect level should be visually displayed in the report.

[0045] In this embodiment, the deformation data is in the form of "positive deflection deformation 3mm". For different component IDs, the physical parameters are different. The physical parameter change data is in the form of: load-bearing capacity default / slight / medium / significantly reduced, or shear capacity default / slight / medium / significantly reduced. The contour change data is in the form of: northwest edge contour default / slight / medium / significantly missing, or southeast edge contour slightly / medium / significantly increased. The component position change data is in the form of: position with XX (component ID) default / few / medium / significantly changed. The risk level is in the form of: default, low risk, medium risk, high risk.

[0046] Repair decision support steps: The output of the component state model determines whether the component needs to be repaired. If repair is required, the repair is assisted by a repair knowledge graph.

[0047] When any one of the following conditions is met: the risk level is a predetermined level, the deformation is greater than the preset deformation, or the change in physical parameters is greater than the preset change in parameters, the component needs to be repaired. Based on the output of the disease identification model, the repair knowledge graph is associated, and at least one garden repair plan is selected.

[0048] Specifically, using "current component type (e.g., column) + disease type (e.g., wood rot) + disease level (e.g., level 2)" as search criteria, the GNN algorithm calculates the similarity between historical cases and the current case (based on node attribute matching degree, such as material compatibility and process adaptability), and returns the multiple cases with the highest similarity (similarity ≥ 85% for inclusion criteria, preferably meeting the inclusion criteria, and then the number of cases that meet the inclusion criteria). In this embodiment, after returning multiple cases, an expert group (at least 2 experts) needs to evaluate the multiple solutions and select a solution. Alternatively, the multiple solutions can be rejected and the solutions can be returned again, or the expert group can formulate a new solution based on the multiple cases.

[0049] Effect verification steps: To obtain environmental data of components within a predetermined period before the restoration of classical gardens, specifically, through temperature and humidity sensors, access to a meteorological platform database, and termite monitoring probes on the components to periodically acquire data.

[0050] Based on the selected garden restoration plan and component environmental data, expected component status data is constructed. Specifically, the expected component status data is AR video, and the component restoration status data is a digital twin model of the component based on environmental data, component image data, and restoration plan data.

[0051] Specifically, based on component data, a time-series prediction model using "LSTM (Long Short-Term Memory Network) + Attention Mechanism" is adopted. Environmental data (temperature, humidity, rainfall), current disease level, and structural parameters (obtained from component finite element model) are used as input features to capture the long-term dependency relationship of disease development.

[0052] Obtain the restoration status data of the components of the restored classical garden, and compare the restoration status data with the expected component status data to obtain a restoration effect score.

[0053] Specifically, the repair effectiveness score includes a preliminary assessment and an expert assessment. The preliminary assessment is based on the effectiveness score of the selected scheme (based on the repair knowledge graph), while the expert assessment is conducted by an expert group who watch AR videos and compare them with the expected component status data to score the repair effectiveness.

[0054] If the repair effect score is greater than or equal to the preset qualified score, it is judged as "repair successful"; if the repair effect score is less than the preset qualified score, an improved plan is manually formulated and the effect verification steps are repeated.

[0055] Specifically, when the repair effect score is lower than the preset qualified score, the expert group needs to discuss and analyze the reasons for the failure to repair and propose a modification plan.

[0056] When the repair effect score of the corresponding improved solution is greater than or equal to the preset qualified score, it is judged as "repair successful", and a new index node and solution node are added to the repair knowledge graph based on the improved solution.

[0057] In this embodiment, the implementation process of the modeling method used to assist in the restoration of classical gardens is as follows: Data processing cycle: Traditional digital modeling requires 7-10 days. In this embodiment, through an automated preprocessing process, the data processing time is shortened to 1-2 days, improving efficiency by 71%-86%.

[0058] Survey and analysis cycle: Traditional manual surveys take 15-20 days, while this embodiment requires only 2-3 days without manual surveys, shortening the cycle by 80%-90%; taking a certain classical garden (core area of ​​about 8 hectares) as an example, traditional surveys take 18 days, while this embodiment only takes 2.5 days, saving 15.5 days.

[0059] Reduced labor costs: Existing technologies require 3-5 experts to participate throughout the process, while this embodiment only requires 1-2 experts in the scheme review stage, reducing labor costs by 60%-80%; for a certain garden restoration project (budget of 5 million yuan), the traditional labor cost is about 1 million yuan, while this embodiment only requires 200,000 yuan, saving 800,000 yuan.

[0060] Disease identification accuracy: The accuracy of basic image recognition using traditional technology is ≤70% (for rare diseases), while the improved model in this embodiment has an accuracy of ≥95%, representing an improvement of 35%-36%. For rare diseases such as "bamboo mold", the accuracy of traditional technology is 65%, while this embodiment reaches 94%, significantly reducing the number of missed or misdiagnosed cases.

[0061] Reasonableness of the solution: The rework rate of traditional technology is 15%. This embodiment reduces the rework rate to below 3% through multi-objective optimization and VR pre-visualization, and reduces rework costs by 80%. If this embodiment is promoted nationwide, it can reduce the rework cost of classical garden restoration by about 200 million yuan per year (calculated based on an average of 1,000 projects per year, with a rework cost of 200,000 yuan per project).

[0062] Data consistency: The expert judgment Kappa (confidence level, the same below) coefficient of traditional technology is 0.72, while the Kappa coefficient of this embodiment is ≥0.91, which improves the consistency by 26%-29%, thereby avoiding the situation of "different experts drawing different conclusions".

[0063] Disease prediction capability: Traditional technology can only detect current diseases, while this embodiment can predict the development of diseases in the next 1-3 years with a prediction error of ≤10%. Taking wood-rotted components as an example, this embodiment can predict the disease level from level 2 to level 3 one year in advance, providing a basis for priority repair and avoiding the escalation of structural risks (such as safety hazards caused by the decline in the load-bearing capacity of columns).

[0064] Historical data reuse capability: Traditional technologies cannot link historical restoration data. This embodiment uses a knowledge graph to quickly retrieve similar cases. When formulating a solution, historical results can be referenced (such as "the column restored with pine wood in 1998 has not decayed for 5 years"). The accuracy of material selection is improved by 40%. In a certain garden restoration, a mixture of traditional tung oil and preservative-treated wood was selected through historical association. The moisture content remained stable at 13% one year after restoration, which is better than the single-material solution of traditional technology.

[0065] Adaptability to traditional craftsmanship: In this embodiment, "cultural heritage" is also taken as one of the core objectives (weight 0.3).

[0066] A storage medium for assisting in the restoration of classical gardens, wherein a processing program is stored thereon, characterized in that the processing program, when executed by a processor, implements the aforementioned modeling method for assisting in the restoration of classical gardens.

[0067] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Various modifications or variations that can be made by those skilled in the art without creative effort within the scope of the appended claims are still within the scope of protection of this patent.

Claims

1. A modeling method for assisting in the restoration of classical gardens, characterized in that, Includes the following steps: Basic data establishment steps: Obtain restoration record data of classical gardens and build a restoration knowledge graph based on this restoration record data; The garden components are reconstructed in three dimensions to obtain finite element models of the components, which are identified by component IDs. A disease identification model is established and trained. This model outputs the component ID, disease type, disease level, and disease location based on component image data. Real-time data input steps: Obtain the component image data of the garden component, input the component image data into the disease identification model, establish the component state model based on the output of the disease identification model and the component finite element model, and output the deformation data, physical parameter change data and risk level of the component state model; Repair decision support steps: Based on the output of the component state model, it is determined whether the component needs to be repaired. If repair is required, the repair is assisted by the repair knowledge graph.

2. The modeling method for assisting in the restoration of classical gardens according to claim 1, characterized in that: in, The basic data establishment steps also include: Image data of classical gardens is acquired, and multiple garden component datasets are extracted based on this image data. These garden component datasets include component shape data and relative position data between components. The finite element models of the components are associated with the garden component datasets through the component IDs. In the real-time data input step, the component state model also outputs contour variation data and inter-component position variation data.

3. The modeling method for assisting in the restoration of classical gardens according to claim 2, characterized in that: in, The image data of the classical gardens refers to overall photographs of the classical gardens taken by drones. The restoration records of the classical gardens were obtained by scanning historical restoration documents and ancient books that documented the restoration of the gardens.

4. The modeling method for assisting in the restoration of classical gardens according to claim 2 or 3, characterized in that: in, The node types of the repaired knowledge graph include index nodes and solution nodes. The index nodes include scheme ID, component type, defect type, repair time, repair materials, repair method, and effect score. The scheme nodes include scheme ID, material list, repair steps, cost budget, shape change, position change, and expected repair effect. The effectiveness score is obtained based on the cost budget, the amount of shape change, the amount of position change, and the expected repair effect.

5. The modeling method for assisting in the restoration of classical gardens according to claim 4, characterized in that: in, The effectiveness score is calculated as follows: Effectiveness score * 0.4 + Cost score * 0.4 + Cultural score * 0.

3. The effect conversion score is obtained based on the expected restoration effect; the cost conversion score is obtained based on the difference between the cost budget and the preset standard budget; and the cultural conversion score is obtained based on the image data of the classical garden, the shape change, and the position change.

6. The modeling method for assisting in the restoration of classical gardens according to claim 1, characterized in that: in, In the repair decision support step, when any one of the following is true: the risk level is a predetermined level, the deformation amount is greater than the preset deformation amount, or the change data of the physical parameters is greater than the preset parameter change amount, the component needs to be repaired. Based on the output result of the disease identification model, the repair knowledge graph is associated to select at least one garden repair scheme.

7. The modeling method for assisting in the restoration of classical gardens according to claim 6, characterized in that, Also includes: Effect verification steps: Obtain environmental data of components within a predetermined period prior to the restoration of classical gardens. Based on the selected garden restoration plan and the environmental data of the components, the expected state data of the components is constructed. Obtain the restoration status data of the components of the restored classical garden, and compare the restoration status data of the components with the expected status data of the components to obtain a restoration effect score.

8. The modeling method for assisting in the restoration of classical gardens according to claim 7, characterized in that: in, When the repair effect score is greater than or equal to the preset qualified score, it is determined as "repair successful"; if the repair effect score is less than the preset qualified score, an improved solution is manually formulated and the effect verification step is re-executed.

9. The modeling method for assisting in the restoration of classical gardens according to claim 8, characterized in that: in, When the repair effect score of the corresponding improved solution is greater than or equal to the preset qualified score, it is determined as "repair successful", and a new index node and solution node are added to the repair knowledge graph based on the improved solution.

10. A storage medium for assisting in the restoration of classical gardens, wherein a processing program is stored thereon, characterized in that, When executed by the processor, the process implements the modeling method for assisting in the restoration of classical gardens as described in any one of claims 1-7.

Images