Construction method of rural housing structure safety multi-dimensional characterization index and evaluation system

Abstract

The application discloses a rural housing structure safety multi-dimensional characterization index and an evaluation system construction method, and belongs to the technical field of the cross of building engineering safety appraisal, disaster prevention and mitigation and artificial intelligence. The method comprises the following steps: constructing a rural house structure safety quantitative characterization index system comprising macroscopic elements, microscopic elements and microscopic elements, collecting image data and geographical position information of a target rural house, performing collaborative analysis on the image data through an improved Mask R-CNN network and a panoramic segmentation network based on a recursive layer aggregation structure, extracting quantitative parameter values strictly corresponding to the index system, inputting the extracted quantitative parameter values into a rule engine with built-in judgment rules, automatically executing multi-level safety evaluation by the rule engine, and outputting a final safety level and an evaluation report. The application realizes efficient, objective and standardized rural house safety evaluation through an AI model and a rule engine, and significantly improves the evaluation efficiency and reliability.

Description

Technical Field

[0001] This invention relates to the fields of building engineering safety assessment, disaster prevention and mitigation and artificial intelligence, specifically to a method for constructing a multi-dimensional characterization index and evaluation system for the structural safety of rural housing. Background Technology

[0002] my country has a vast number of rural houses with diverse structural types, and many were built without proper design and construction practices, leading to varying degrees of safety hazards after long-term use. Currently, safety assessments of rural houses are mainly based on the "Technical Guidelines for Safety Assessment of Rural Housing." While this method simplifies the process, it relies heavily on manual on-site inspections and uses qualitative descriptions such as "minor cracks" and "severe tilting" as criteria, resulting in highly subjective and inconsistent assessments that depend heavily on the personal experience of the assessors. Furthermore, manual, house-by-house inspections are inefficient and costly, making it difficult to conduct routine safety surveys of large numbers of rural houses, and even more difficult to rapidly screen and classify the safety of massive numbers of damaged houses during emergency rescue operations following earthquakes or other sudden disasters.

[0003] In recent years, deep learning-based computer vision technology has shown potential in the field of civil engineering health monitoring, such as for crack detection. However, existing research is mostly limited to single, general damage identification, failing to deeply integrate with a complete, professional, and quantitative knowledge system for rural housing safety assessment. The fundamental problem lies in the inability to automatically extract precise indicators that conform to professional specifications from images (such as crack width limits and wall thickness requirements for specific types of components), and even more so in the inability to embed the extracted data into a professional assessment process that includes multi-level indicators, complex weight calculations, and progressive decision-making logic. Therefore, developing a technical solution that can automate and intelligently execute professional assessment standards has become an urgent need to solve the challenges of rural housing safety supervision.

[0004] To address the aforementioned issues, there is an urgent need for a method to construct a multi-dimensional characterization index and evaluation system for the structural safety of rural housing, thereby resolving the problems associated with traditional methods. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing a multi-dimensional characterization index and evaluation system for the structural safety of rural housing. By automatically acquiring quantitative parameters through an AI model and using a rule engine for intelligent decision-making, the invention achieves high efficiency, objectivity, and standardization in the safety assessment of rural housing, significantly improving the efficiency and reliability of the assessment.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety, including:

[0008] Step 1: Construct a quantitative characterization index system for the structural safety of rural houses, comprising macro-level elements, micro-level elements, and micro-level elements, along with corresponding judgment rules;

[0009] Step 2: Collect image data and geographic location information of the target farmhouse. Through the improved Mask R-CNN network and the panoramic segmentation network based on the recursive layer aggregation structure, perform collaborative analysis on the image data and extract quantitative parameter values ​​that strictly correspond to the indicator system.

[0010] Step 3: Input the extracted quantitative parameter values ​​into the rule engine with built-in judgment rules. The rule engine will automatically perform multi-level security assessment and output the final security level and assessment report.

[0011] Furthermore, the macro-level indicator is site hazard, which is classified into three levels: hazardous, potential hazardous, and safe based on geological disaster risk; the micro-level indicator includes structural disaster prevention measures indicators and overall structural safety indicators; the micro-level indicator includes quantitative descriptions and grading thresholds for cracks, deformation, cross-sectional loss, and node connection status of different material components.

[0012] Furthermore, the structural disaster prevention measures indicators in the detailed elements specify height and number of stories limits, minimum wall thickness limits, and seismic structural requirements for masonry, base frame, reinforced concrete, wood, stone, rammed earth walls, and steel structures, respectively.

[0013] Furthermore, the overall structural safety indicators among the micro-elements include the overall tilt rate and the uneven settlement degree; the overall tilt rate is obtained by calculating the ratio of the horizontal offset value of the observation point to its height; the uneven settlement degree is obtained by calculating the ratio of the vertical deformation difference of the observation point to its distance.

[0014] Furthermore, in step 2, the input of the improved Mask R-CNN network is a single image of the target farmhouse, and the output is a bounding box, a pixel-level instance segmentation mask, an element type label, and a material type label for each detected structural component, forming a preliminary list of element instances.

[0015] Furthermore, the improved Mask R-CNN network is specifically as follows:

[0016] Based on the original Faster R-CNN network, ROI Align operation is used instead of ROI Pooling;

[0017] Based on the original Faster R-CNN network, an improved ResNet50 and a feature pyramid network are used as the backbone for feature extraction.

[0018] A fully convolutional network is introduced into the segmentation branch of the original Faster R-CNN network.

[0019] Furthermore, the input to the panoramic segmentation network is a single image of the target farmhouse, and the output is a panoramic segmentation map with a semantic label and an instance ID for each pixel. The semantic label categories include: multiple structural component categories, multiple damage categories, and background categories; wherein the damage categories include at least cracks, peeling, weathering, and corrosion.

[0020] Furthermore, the panoramic segmentation network includes a feature extraction backbone network, a semantic segmentation head, an instance segmentation head, and a panoramic fusion head.

[0021] Furthermore, step 2 also includes:

[0022] By associating and merging the initial list of component instances with the panoramic segmentation map, a precise attribution relationship is established between the damaged area and its associated component.

[0023] Furthermore, step 2 also includes:

[0024] Based on the acquired multi-view sequence images, a high-precision three-dimensional dense point cloud model of the target farmhouse is generated by using the structure of motion recovery (SFM) and multi-view stereo (MVS) algorithms. By mapping the two-dimensional images to the three-dimensional point cloud, the fused and identified components and damaged areas are mapped to three-dimensional space, and the physical dimensions of the components, the physical scale of the damage, and the overall tilt and uneven settlement parameters of the house are directly measured.

[0025] In summary, the present invention has at least one of the following beneficial technical effects:

[0026] 1. This invention automates and objectifies the evaluation process. By working in tandem with an improved Mask R-CNN network and a recursive layer aggregation panoramic segmentation network, it automatically completes the entire process from component identification and material judgment to damage quantification and extraction. This replaces manual visual inspection and measurement, completely eliminates the influence of subjective human factors, and ensures the objectivity and consistency of the evaluation results.

[0027] 2. This invention significantly improves assessment efficiency and coverage. It reduces the traditional process of manual on-site inspection, recording, and calculation, which requires hours, to automated processing within minutes. It is particularly suitable for large-scale rural housing safety surveys and rapid post-disaster emergency screenings, enabling preliminary safety assessments of tens of thousands of houses in a very short time, providing unprecedented efficiency support for management decisions.

[0028] 3. This invention ensures the professionalism and accuracy of the assessment conclusions. The core of the entire system's decision-making is a rule engine built on professional standards, which strictly follows a progressive assessment logic from macro to micro and then to macro levels, and applies a scientific weighted statistical model. This makes the automated assessment results comparable to, or even more rigorous than, those of professional appraisers.

[0029] 4. A traceable and reliable assessment record has been established: The structured report automatically generated by the system not only includes the safety level conclusion but also fully records all the evidence chain leading to that conclusion, including image annotations of key damages, extracted quantitative parameters, triggered judgment rules, and intermediate calculation results. This achieves complete transparency and traceability in the assessment process, greatly enhancing the credibility and legal reference value of the assessment results. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0031] Figure 2 This is a schematic diagram of the improved Mask R-CNN network structure. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0033] like Figure 1 As shown, this invention provides a method for constructing a multi-dimensional characterization index and evaluation system for the structural safety of rural housing, including:

[0034] Step 1: Construct a quantitative characterization index system for the structural safety of rural houses, comprising macro-level elements, micro-level elements, and micro-level elements, along with corresponding judgment rules;

[0035] Step 2: Collect image data and geographic location information of the target farmhouse. Through the improved Mask R-CNN network and the panoramic segmentation network based on the recursive layer aggregation structure, perform collaborative analysis on the image data and extract quantitative parameter values ​​that strictly correspond to the indicator system.

[0036] Step 3: Input the extracted quantitative parameter values ​​into the rule engine with built-in judgment rules. The rule engine will automatically perform multi-level security assessment and output the final security level and assessment report.

[0037] In step 1, a quantitative characterization index system for the structural safety of rural houses, comprising macro-level, micro-level, and macro-level elements, and corresponding judgment rules, is constructed, specifically as follows:

[0038] The following sections will introduce the macro-level factors, the meso-level factors, and the micro-level factors respectively:

[0039] 1. Macro-level factors: Site hazards

[0040] Macro-level factors are used to assess the geological safety of building sites and are the primary prerequisite for the assessment.

[0041] Its core indicator is the site hazard level;

[0042] Its quantitative grading and judgment rules are as follows:

[0043] (1) Hazard: The site meets any of the following conditions:

[0044] 1) Landslides, collapses, subsidence, and ground fissures may occur;

[0045] 2) Located in areas prone to flooding, flash floods, or debris flows;

[0046] 3) Located in an area with highly developed karst and cave systems;

[0047] 4) The goaf has shown a clear trend of deformation and subsidence;

[0048] (2) Potential hazards: The site has risks that are not yet stable or identified, such as unstable goaf areas, flood-prone areas, and geological hazard points with unknown risks.

[0049] (3) Safety: The above-mentioned dangerous or potential dangerous conditions are not met.

[0050] 2. Detailed factors: Overall structural safety

[0051] This element assesses the building's overall disaster resilience and current condition, and includes two sub-indicators:

[0052] (1) Structural disaster prevention measures indicators

[0053] Objective: To assess whether the structure meets basic seismic and disaster prevention design requirements;

[0054] Judgment Logic: Based on the "Standard for Seismic Appraisal of Buildings" (GB50023-2009), a differentiated compliance checklist is set for seven types of rural houses: masonry, frame, wood, stone, rammed earth, reinforced concrete, and steel structure. The core of the checklist includes: height and number of stories limits, minimum thickness limits for load-bearing walls (columns), and requirements for specific seismic structural measures (such as ring beams, structural columns, scissor bracing, bottom chord tie rods, and shear walls). If all of them are met, the house is considered "safe"; if any item is not met, the house is considered "dangerous".

[0055] (2) Overall structural safety indicators

[0056] Objective: To quantitatively assess the overall stability of a building through geometric deformation;

[0057] Core parameters and formulas:

[0058] Overall tilt rate ( i h ): .in, d This represents the horizontal offset value of the upper observation point. H The corresponding height limits are set according to the "Standards for Identification of Dangerous Buildings";

[0059] Uneven settlement ( i t ): .in, h The vertical deformation difference at the observation point. L For horizontal distance, the danger limit is i t >0.002.

[0060] Overall assessment: "Safe" is defined as both indicators being within limits, while "dangerous" is defined as either indicator exceeding the limit.

[0061] 3. Microscopic factors: Component damage safety

[0062] This element delves into the component level and is the most refined part of the system;

[0063] Comprehensive coverage: Damage indicators are developed for six types of materials: masonry, reinforced concrete, wood, stone, raw earth, and steel, for each component (wall, column, beam, slab, and joint).

[0064] Contextualized thresholds: Two sets of quantitative grading thresholds are set for "routine safety management" and "post-disaster emergency assessment", with the latter being more lenient to facilitate rapid screening;

[0065] Fully quantifiable description: All damage is defined by physical quantities, such as crack width (mm), percentage of spalled area (%), deformation ratio (e.g., deflection / span), etc., completely eliminating ambiguous descriptions.

[0066] The following section introduces the comprehensive judgment model from component to structure, which includes:

[0067] 1. Component status synthesis rule: If any damage index of a single component is judged as "dangerous", then the component is "dangerous"; if all indices are "safe", then the component is "safe"; the rest are "potentially dangerous".

[0068] 2. Comprehensive model for determining the microscopic safety level of a structure:

[0069] Calculate the proportion of dangerous components on each floor ( R i ): Weights are introduced (walls and columns 3.0, beams 1.5, slabs 1.0) to reflect the differences in importance of different components. The formula is as follows:

[0070] (1)

[0071] In the formula, , , These represent the number of dangerous structural members (walls, columns, beams, and slabs) on the i-th floor. , , These represent the number of wall columns, beams, and slab components in the i-th layer, respectively.

[0072] Calculate the overall proportion (R) of dangerous components in the entire building: calculate the proportion of dangerous components in the entire building with equal weights;

[0073] Determining the microscopic level:

[0074] Risk: If R ≥ 25%; or if 5% ≤ R < 25% and R is satisfied i ≥25% of the floors must account for no less than 1 / 3 of the total number of floors;

[0075] Potential hazards: If 5% ≤ R < 25% but the above floor distribution is not met; or if R < 5% but R exists. i ≥25% of the floors.

[0076] Safety: If R < 5% and R is true for all floors i <25%.

[0077] The overall decision-making logic of the entire method is defined as a strict, irreversible, progressive process:

[0078] Macro-level decision: If the site is deemed "dangerous", terminate immediately and output "dangerous" overall.

[0079] Detailed Decision Making: If the site is not hazardous, proceed with a detailed assessment. If the detailed assessment indicates "hazard," immediately terminate the assessment and output "hazardous" for the overall site.

[0080] Micro-level decision-making: If the level is "safe" at a detailed level, then a comprehensive assessment of micro-level factors is conducted, and the micro-level safety level is used as the final overall safety level of the building.

[0081] Output: The final level is divided into three levels: "Safe", "Potentially Dangerous", and "Dangerous", with corresponding handling recommendations.

[0082] In step 2, image data and geographic location information of the target farmhouse are collected. An improved Mask R-CNN network and a panoramic segmentation network based on a recursive layer aggregation structure are used to collaboratively analyze the image data and extract quantitative parameter values ​​that strictly correspond to the indicator system. Specifically:

[0083] I. Collecting data from multiple sources, specifically including:

[0084] Aerial imagery: High-resolution sequence of multi-view images of buildings collected by drones;

[0085] Ground imagery: Images of components and detailed damage captured by mobile terminals;

[0086] Location data: Records the latitude and longitude coordinates of the house;

[0087] Reference calibration: Place a calibration object of known size on the key measurement parts.

[0088] II. Component Instance Recognition Based on an Improved Mask R-CNN Network

[0089] 1. Input: A single image of the target farmhouse;

[0090] 2. Output: Generate a list of component instances, each instance containing: bounding box, pixel-level mask, component type label, and material type label;

[0091] 3. A detailed introduction to the improved Mask R-CNN network:

[0092] The improved Mask R-CNN network architecture is as follows: Figure 2 As shown, the improved Mask R-CNN network adds a branch for semantic segmentation. The entire model mainly consists of a feature extraction network, RPN, ROI Align module, and a three-branch prediction module. The input image first extracts multi-scale feature maps through the improved ResNet50 and FPN to enhance the detection capability of targets at different scales. The feature maps are passed to the RPN to generate candidate regions. The classification branch determines whether the candidate region contains the target, and the bounding box regression branch adjusts the position and size of the candidate region. After passing through the ROI Align module, the candidate region is extracted as a fixed-size feature for subsequent processing. The three-branch prediction module predicts the target category through fully connected layers to further optimize the accurate location of the target. A pixel-level target segmentation mask is generated through a fully convolutional network. The improved Mask R-CNN replaces the original ROIPooling with ROIAlign to solve the candidate region alignment error problem. At the same time, the segmentation branch further improves the understanding of the target shape and boundary, making it perform better in target detection and segmentation tasks.

[0093] III. Refined Scene Analysis Based on Recursive Layer Aggregation Panoramic Segmentation Network

[0094] 1. Input: A single image of the target farmhouse;

[0095] 2. Output: Generates a pixel-level panoramic segmentation map. Each pixel has a (semantic label, instance ID) pair. The semantic label covers detailed component categories and damage categories (such as cracks, peeling, and corrosion). This network is good at recognizing complex morphological damage such as slender and mesh-like structures.

[0096] 3. Specific structure of the panoramic segmentation network

[0097] The overall structure of the panoramic segmentation network consists of four parts: a feature extraction backbone network, a semantic segmentation head, an instance segmentation head, and a panoramic fusion head. The feature extraction backbone network uses a ResNet structure with recursive layer aggregation and a bidirectional FPN structure. The FPN structure is followed by a channel diversification block. The ResNet structure with recursive layer aggregation employs a recurrent network structure, which avoids model parameter redundancy while fusing information from different layers for feature extraction in the current layer. The bidirectional FPN structure allows feature information to flow bidirectionally, avoiding the limitation of the standard FPN structure's unidirectional information flow from shallow to deep layers, which hinders the aggregation of multi-scale features. The channel diversification block (CDB) takes the output of the bidirectional FPN structure as input and enhances global contextual information by establishing global-level channel attention relationships, while also considering the importance of each channel's information, making the network focus more on salient features.

[0098] The semantic segmentation head extracts fine-grained features and contextual features, and then performs mismatch correction to fuse features of different sizes. Fine-grained features and contextual features are extracted by a Large Scale Feature Extractor (LSFE) and a Dense Prediction Cell (DPC), respectively, and then fused using a mismatch correction (MC) module. The backbone network extracts {P4, P8, P...} 16 P 32}, where {P4, P8} are used to extract fine features, {P 16 P 32 The input is used to extract context information. Before the context information is extracted from the input to the DPC, it is processed through a branch consisting of a skip connection and a global attention module.

[0099] The instance segmentation head adopts the structure of the EfficientPS network, replacing the convolution, BN operation and Re-LU activation function in MaskR-CNN with depthwise separable convolution, iABN synchronization layer and LeakyReLU activation function.

[0100] The panoramic fusion head also follows the EfficientPS network. It first reduces the number of instance objects by filtering and sorting through confidence thresholds and checking for instance overlap. Then, it adaptively adjusts the fusion by combining the confidence of the instance head and the semantic head.

[0101] Recursive layer aggregation of ResNet structure and channel diversification modules are existing technologies and will not be introduced here.

[0102] IV. Multi-model output fusion and two-dimensional parameter extraction

[0103] Fusion logic: Using the instance list output by Mask R-CNN as the main index, the pixel-level semantic information provided by the panoramic segmentation map is used for verification and refinement. For example, the precise outline of a wall is determined and the set of all pixels marked as "cracks" in its area is extracted, thereby establishing the precise attribution relationship between damage and components.

[0104] Initial extraction of two-dimensional parameters: Based on the fused results, parameters such as damage size and location in the image coordinate system can be initially calculated.

[0105] V. Precise Extraction of Geometric Parameters Based on 3D Reconstruction

[0106] 1. 3D model generation: For UAV image sequence, the Structure of Motion Recovery (SFM) and Multi-View Stereo (MVS) algorithms are used to reconstruct a high-precision 3D dense point cloud model;

[0107] 2. Two-dimensional-three-dimensional association: Establish a spatial mapping between image pixels and three-dimensional point clouds through camera parameters.

[0108] 3. Precise parameter calculation:

[0109] (1) Component geometric dimensions: The wall thickness, beam height, etc. are directly measured in the three-dimensional point cloud for the verification of disaster prevention measures.

[0110] (2) Damage physical scale: Map the damage identified in the image to three-dimensional space and calculate its actual physical size (true width / length of crack, true area of ​​peeling).

[0111] (3) Overall deformation parameters: The overall tilt parameters (d, H) and uneven settlement parameters (h, L) are accurately fitted and calculated from the overall three-dimensional model.

[0112] VI. Spatial Information Integration

[0113] Spatial overlay analysis of building location coordinates with geological hazard GIS risk layers provides objective auxiliary basis for macro-level site hazard assessment.

[0114] VII. Final Output: Structured Parameter Dataset

[0115] By integrating the outputs of all the above processes, a structured parameter dataset that is strictly isomorphic to the indicator system of the first step is generated, which includes all quantitative parameters at the macro, micro, and granular levels, and serves as the input for step 3.

[0116] In step 3, the extracted quantified parameter values ​​are input into a rule engine with built-in judgment rules. The rule engine automatically performs multi-level security assessments and outputs the final security level and assessment report, specifically as follows:

[0117] This step seamlessly integrates the knowledge from step 1 with the data from step 2 through a software-based rules engine, achieving a fully automated, standardized, and traceable security assessment. Specifically, it includes:

[0118] I. Construction and Knowledge Injection of the Rule Engine

[0119] Develop or configure a rules engine system;

[0120] The entire quantitative indicator system, judgment rules, calculation formulas (such as R-value calculation) and comprehensive evaluation process constructed in step 1 are fully encoded and loaded into the engine to form its "professional knowledge base".

[0121] II. Automated Evaluation and Execution Process

[0122] The engine automatically triggers after receiving the structured parameter dataset generated in step 2:

[0123] 1. Automatic determination of macro-level factors: The engine calls upon site hazard rules, combines geographical location and GIS analysis results, and automatically determines the hazard level. If it is "hazardous", the termination rule is immediately triggered, skipping all subsequent calculations and directly outputting the overall "hazardous" conclusion.

[0124] 2. Automatic determination of details: If the macroscopic situation is not dangerous, the engine will start the detail determination submodule.

[0125] (1) Disaster prevention measures verification: Based on the structural type, the measured values ​​of the number of floors, wall thickness and other values ​​are automatically compared with the standard limits to determine the compliance of the construction measures.

[0126] (2) Overall security calculation: reading or calculating i h and i t It is automatically compared with the limit value.

[0127] (3) If the detailed inspection determines that it is "dangerous", then the termination rule is triggered again and the overall "dangerous" is output.

[0128] 3. Automatic determination of micro-factors (core calculation): If the micro-factor is "safe", the engine enters the micro-factor determination loop.

[0129] (1) Component traversal and rule matching: Traverse each component and automatically match the corresponding quantitative index threshold table according to its material type.

[0130] (2) Item-by-item comparison and state synthesis: The quantification value of each damage of the component is automatically compared with the threshold, and the component state is synthesized by applying the "one-vote veto" rule.

[0131] (3) Global statistics and level determination: Automatically count dangerous components, calculate R value, and determine micro safety level based on R value and its distribution rules.

[0132] 4. Final Decision: The engine summarizes the results from all three levels, strictly follows the progressive process defined in the first step, and makes a final decision.

[0133] III. Automatic Generation of Structured Reports

[0134] Throughout the inference process, the engine automatically logs all key decisions: triggered rules, parameter values ​​used, comparison thresholds, intermediate calculation results (such as R values), etc.

[0135] The report generation module automatically associates these logs with the original evidence from step 2 (such as images annotated with damage and 3D model views), populates them into a standardized template, and instantly generates a structured safety assessment report. The report is detailed, including conclusions, supporting evidence, and recommendations, and achieves complete traceability of the assessment process.

[0136] To illustrate the technical solution and effectiveness of the present invention in detail, a specific embodiment is described below.

[0137] Example: Daily safety assessment of a two-story masonry structure farmhouse in a certain area.

[0138] 1. Data Collection:

[0139] The operator controlled a drone to fly around the farmhouse, automatically collecting 120 high-definition images from multiple perspectives, including the roof and four facades. Simultaneously, a handheld smart tablet was used to focus on photographing the interior and exterior walls, beams, columns, and corners, acquiring 50 detailed images. The equipment automatically recorded the house's latitude and longitude coordinates. When photographing a wall with cracks, a standard ruler was placed next to the crack as a calibration object.

[0140] 2. Structured parameter extraction:

[0141] All image data was uploaded to the cloud processing platform. First, the improved Mask R-CNN model was run. In a single main facade image, the model successfully identified and segmented 8 instances of "load-bearing walls," all of which were classified as "masonry"; identified 4 instances of "beams," all of which were classified as "reinforced concrete"; and accurately identified non-structural components such as windows and doors.

[0142] Subsequently, a panoramic segmentation network based on recursive layer aggregation was invoked to process the same image. The network output a pixel-level panoramic image, which not only refined the wall outlines but also accurately segmented multiple thin, continuous "crack" pixel regions within one wall instance area. These cracks were not detected individually in Mask R-CNN.

[0143] Model fusion was performed, confirming that the crack belongs to the masonry load-bearing wall numbered "Wall_01". Using a scale in the image, the physical width of the maximum crack on this wall was calculated to be 1.8 mm.

[0144] A high-precision 3D point cloud model of the farmhouse was generated by reconstructing a sequence of UAV images. From this model, it was directly measured that the house as a whole had no significant tilt (calculated). i h (Much smaller than the limit), the minimum thickness of each load-bearing wall is 260 mm (greater than the standard limit of 240 mm). Simultaneously, through correlation mapping, the actual location and direction of the cracks in the two-dimensional image in three-dimensional space were confirmed.

[0145] By comparing the coordinates of the house with the local geological hazard database, the analysis results show that the site does not belong to a high-risk area such as landslides or debris flows, and the macroscopic site hazard is initially judged to be "safe".

[0146] Finally, a structured parameter dataset of the house is generated, containing all the aforementioned quantitative information.

[0147] 3. Automated intelligent assessment:

[0148] After receiving the dataset, the rule engine begins automatic reasoning.

[0149] Macro-level assessment: Based on the GIS analysis results, the site's hazard level is automatically determined to be "safe".

[0150] Detailed assessment: The engine reads the structure type as "masonry" and the number of stories as "2". Data verification: The number of stories (2 stories) is within limits; the minimum wall thickness (260mm) is greater than the limit (240mm); image recognition shows no obvious missing ring beam structural columns. Therefore, the "Disaster Prevention Measures Index" is automatically judged as "Safe". The tilt and settlement data read from the 3D model are both within limits, and the "Overall Safety Index" is also judged as "Safe". Therefore, the comprehensive assessment of detailed elements is "Safe".

[0151] Microscopic Determination: The engine traverses all components. For the "Wall_01" wall, based on its material (masonry) and scene (daily management), the engine invokes the "vertical cracks in masonry walls" indicator rule from the knowledge base. This rule defines a crack width greater than 2 mm as "dangerous." The extracted crack width value is 1.8 mm, and the engine automatically determines this indicator as "potentially dangerous." All other damage indicators for this wall are "safe." According to the "one-vote veto" principle, the wall component's status is ultimately judged as "potentially dangerous." Upon verification, all other component statuses are "safe."

[0152] The engine automatically calculated the following: There are a total of 10 wall and column components in the building, of which 0 are hazardous and 1 is potentially hazardous (i.e., Wall_01). All beams and slabs are safe. Substituting the values ​​into the weighting formula, the overall proportion of hazardous components in the building is R≈3.6% (less than 5%), and there are no hazardous floors.

[0153] Based on the micro-level determination rules (R<5% and no unsafe floors), the engine determines that the micro-level safety of the house is "safe".

[0154] Final decision: Based on the comprehensive results of macro (safety), micro (safety), and detailed (safety) assessments, the engine outputs that the overall safety level of the farmhouse is "safe".

[0155] 4. Report generation:

[0156] The system automatically generated a "Structural Safety Assessment Report for Rural Houses," which clearly concluded that the house was "safe." The report appendix included the original drawing of the "Wall_01" wall, with identified cracks marked in red and noted next to them as "Maximum crack width: 1.8mm, judgment: potential danger." The report also listed key criteria for the assessment, such as "Minimum wall thickness: 260mm ≥ 240mm, compliant" and "Overall tilt rate: not exceeded." The final recommendation was: "The overall structure of the house is safe and can continue to be used normally. Regular observation of the cracks in the north exterior wall is recommended."

[0157] Example Implementation: This example fully demonstrates the entire process of the method of the present invention, from data acquisition to report generation. This method successfully transforms a vague qualitative inspection ("There is a crack in the wall") into a precise quantitative analysis ("The crack is 1.8mm wide, judged as a potential hazard according to regulations"), and ultimately draws an overall conclusion through a scientific comprehensive model. The entire process does not require a professional structural engineer to be on-site; it is completed automatically by equipment and the system, taking less than one-tenth the time of traditional methods. Furthermore, the conclusions are clear, the evidence is conclusive, and the logic is rigorous, fully demonstrating the practicality, reliability, and efficiency of the present invention.

[0158] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0160] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0161] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0162] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A method for constructing a multi-dimensional characterization index and evaluation system for the structural safety of rural housing, characterized in that, include: Step 1: Construct a quantitative characterization index system for the structural safety of rural houses, comprising three levels: macro-level elements, meso-level elements, and micro-level elements, along with corresponding judgment rules. The macro-level element index is the site hazard, which is divided into three levels: dangerous, potential dangerous, and safe, based on geological disaster risk. The meso-level element index includes structural disaster prevention measures index and overall structural safety index. The micro-level element index includes quantitative descriptions and grading thresholds for cracks, deformation, cross-sectional loss, and node connection status of different material components. Step 2: Collect image data and geographic location information of the target farmhouse. Through the improved Mask R-CNN network and the panoramic segmentation network based on the recursive layer aggregation structure, perform collaborative analysis on the image data and extract the quantitative parameter values ​​corresponding to the indicators in the indicator system. The collaborative analysis includes: associating and fusing the component instance list output by the improved Mask R-CNN network with the panoramic segmentation map output by the panoramic segmentation network; using the component instance list as the main index; and using the pixel-level semantic information provided by the panoramic segmentation map to verify and refine the component instances, thereby establishing a precise attribution relationship between the damaged area and its component. Step 3: Input the extracted quantitative parameter values ​​into the rule engine with built-in judgment rules. The rule engine will automatically perform multi-level security assessment and output the final security level and assessment report.

2. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 1, characterized in that, The structural disaster prevention measures indicators in the detailed elements specify height and number of stories limits, minimum wall thickness limits, and seismic structural requirements for masonry, base frame, reinforced concrete, wood, stone, rammed earth walls, and steel structures, respectively.

3. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 2, characterized in that, The overall structural safety indicators among the micro-elements include the overall tilt rate and the uneven settlement degree; the overall tilt rate is obtained by calculating the ratio of the horizontal offset value of the observation point to its height; the uneven settlement degree is obtained by calculating the ratio of the vertical deformation difference of the observation point to its distance.

4. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 2, characterized in that, In step 2, the input of the improved Mask R-CNN network is a single image of the target farmhouse, and the output is a bounding box, a pixel-level instance segmentation mask, an element type label, and a material type label for each detected structural component, forming a preliminary list of element instances.

5. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 4, characterized in that, The improved Mask R-CNN network is specifically as follows: Based on the original Faster R-CNN network, ROI Align operation is used instead of ROI Pooling; Based on the original Faster R-CNN network, an improved ResNet50 and a feature pyramid network are used as the backbone for feature extraction. A fully convolutional network is introduced into the segmentation branch of the original Faster R-CNN network.

6. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 5, characterized in that, The panoramic segmentation network takes a single image of a target farmhouse as input and outputs a panoramic segmentation map where each pixel has both a semantic label and an instance ID. The semantic label categories include: multiple structural component categories, multiple damage categories, and background categories; the damage categories include at least cracks, peeling, weathering, and corrosion.

7. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 6, characterized in that, The panoramic segmentation network includes a feature extraction backbone network, a semantic segmentation head, an instance segmentation head, and a panoramic fusion head.

8. The method for constructing a multi-dimensional characterization index and evaluation system for rural housing structural safety according to claim 7, characterized in that, Step 2 also includes: Based on the acquired multi-view sequence images, a high-precision three-dimensional dense point cloud model of the target farmhouse is generated by using the structure of motion recovery (SFM) and multi-view stereo (MVS) algorithms. By mapping the two-dimensional images to the three-dimensional point cloud, the fused and identified components and damaged areas are mapped to three-dimensional space, and the physical dimensions of the components, the physical scale of the damage, and the overall tilt and uneven settlement parameters of the house are directly measured.

Images