A Tag-Based Method and System for Cultural Relics Conservation Resource Data Governance

Abstract

This application provides a tag-based method and system for the governance of cultural relic protection resource data, relating to the field of data processing technology. This application identifies diseased units in multi-phase 3D models and constructs vectors for size and morphological changes. It then extracts displacement trajectories of key surface points to generate deformation data, associates these with composite attributes to construct a spatiotemporal graph, and mines the spatiotemporal relationships of diseased units through convolution operations to output feature data. This feature data is mapped to a knowledge graph vector space encapsulating disease evolution patterns and causal relationships. Evolution patterns with semantic distances meeting requirements are matched, and semantic tags for diseased units are determined by combining pattern attributes and causal relationship analysis. Finally, effective clues are generated based on the tags and spatiotemporal relationships. This method enables precise mining of the changing characteristics of diseased units in rammed earth sites and analysis of their evolutionary patterns, providing data support for cultural relic protection.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a tag-based method and system for the governance of cultural relic protection resource data. Background Technology

[0002] Data governance methods for cultural relic conservation resources are a core supporting technology for the digital preservation of cultural relics, enabling the standardized processing and value extraction of relevant data. This method is widely applied in the protection of vulnerable cultural relics such as rammed earth sites, providing data support for monitoring cultural relic damage, predicting evolution, and formulating protection plans, and has broad application prospects.

[0003] Currently, existing tag-based data governance methods for cultural relic protection resources mostly determine the tags of cultural relic defects through manual annotation or simple feature matching, and often combine single-dimensional cultural relic data for analysis. These methods have been applied in some cultural relic protection scenarios, aiming to achieve orderly management of cultural relic protection resources through tag-based management.

[0004] However, existing technologies struggle to simultaneously consider the dynamic changes and spatiotemporal correlations of cultural relic damage, resulting in insufficient accuracy in damage labeling and an inability to effectively uncover the evolutionary patterns of damage. Consequently, it becomes difficult to generate effective clues to support precise cultural relic protection. Therefore, existing technologies suffer from insufficient precision in the governance of cultural relic protection resource data, making it difficult to effectively support precise cultural relic protection. Summary of the Invention

[0005] The purpose of this application is to provide a tag-based method and system for the governance of cultural relic protection resource data, in order to solve the problems of insufficient accuracy in the governance of cultural relic protection resource data in the prior art and the inability to effectively support the precise protection of cultural relics.

[0006] To address the aforementioned technical problems, firstly, this application provides a tag-based method for managing cultural relic conservation resource data, comprising:

[0007] The diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site are identified, and a change vector is constructed for each diseased unit. The change vector is used to reflect the changes in the size and shape of the diseased unit over time.

[0008] The displacement trajectory corresponding to the surface key points of the target rammed earth site is determined, and deformation data of the surface key points is generated based on the displacement trajectory. The deformation data is used to reveal the movement pattern of the surface key points within a preset period.

[0009] The change vector is associated with the deformation data at the corresponding spatiotemporal location to form a composite attribute of the node, and a spatiotemporal graph is constructed based on the composite attribute. In the spatiotemporal graph, the spatiotemporal relationship between disease units is mined through convolution operation to output feature data.

[0010] The feature data is mapped to the vector space of a pre-constructed disease knowledge graph, which encapsulates the disease evolution pattern and the relationship between causes.

[0011] Within the vector space, multiple disease evolution patterns that satisfy a preset condition in semantic distance from the feature data are searched. Based on the attributes of the multiple disease evolution patterns and the causal relationship, the disease units corresponding to the feature data are analyzed to determine the semantic labels corresponding to the disease units.

[0012] Based on the semantic tags and the spatiotemporal relationships contained in the feature data, valid clues are generated.

[0013] Optionally, a spatiotemporal graph is constructed based on the composite attributes. Convolution operations are then used to mine the spatiotemporal relationships between diseased units within the spatiotemporal graph, and feature data is output, including:

[0014] Using key points on the surface of the target rammed earth site as nodes, a spatiotemporal map is constructed;

[0015] By performing convolution operations on the spatial and temporal connections of the spatiotemporal graph based on the composite attributes of the nodes, the spatial transmission relationship and temporal causal order between the disease units are mined, and feature data is output. The feature data is used to characterize the spatial transmission relationship and temporal causal order between the disease units.

[0016] Optionally, by performing convolution operations on the spatial and temporal connections of the spatiotemporal graph based on the composite attributes of the nodes, the spatial transmission relationships and temporal causal order among the disease units are mined, and feature data is output, including:

[0017] In the spatiotemporal graph, for each spatial connection, the composite attributes of adjacent nodes are summed using a weighted average.

[0018] In the spatiotemporal graph, for each temporal connection, the composite attributes of the same node at different time points are weighted and summed.

[0019] The weighted summation results of spatial connections and time connections are fused to obtain the updated attributes of each node. The transmission relationships and causal order between disease units are identified through the updated attributes, and the transmission relationships and causal order are combined into feature data.

[0020] Optionally, mapping the feature data to the vector space of a pre-constructed disease knowledge graph includes:

[0021] Obtain a pre-constructed disease knowledge graph, extract the numerical representations of all nodes and edges from the disease knowledge graph, and organize the numerical representations of all nodes and edges into a vector space.

[0022] For the feature data, the numerical weight of each dimension is calculated, and the feature data is mapped by matching the numerical weight of the feature data with the corresponding dimension in the vector space.

[0023] Optionally, within the vector space, multiple disease evolution patterns that satisfy a preset condition in semantic distance to the feature data are searched, and the disease units corresponding to the feature data are analyzed based on the attributes of the multiple disease evolution patterns and the causal relationship to determine the semantic labels corresponding to the disease units, including:

[0024] Within the vector space, the numerical difference between the feature data and each disease evolution mode is calculated, and multiple disease evolution modes whose numerical differences satisfy a preset threshold are selected.

[0025] For the disease units corresponding to the feature data, common attributes and related causal relationships of the multiple disease evolution patterns are extracted. Pattern matching is achieved by comparing the attributes of the disease units with the common attributes and causal relationships item by item, and semantic labels are selected and assigned to the disease units based on the matching results.

[0026] Optionally, based on the semantic tags and the spatiotemporal relationships contained in the feature data, valid clues are generated, including:

[0027] Spatiotemporal relationships are extracted from the feature data, and the spatiotemporal relationships are combined with the semantic tags;

[0028] For each diseased unit, the corresponding evolutionary pattern is identified based on the combined results, potential risk points are inferred through the spatiotemporal relationship, and the diseased unit, the evolutionary pattern, and the potential risk points are integrated into effective clues.

[0029] Optionally, the diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site are determined, and a change vector is constructed for each diseased unit, including:

[0030] A multi-phase three-dimensional model of the target rammed earth site is obtained, and the boundary contours of key surface points are extracted from the multi-phase three-dimensional model. The disease units are divided by comparing the differences between the boundary contours of key surface points in the three-dimensional models of adjacent periods.

[0031] Calculate the volume value and shape parameters of each disease unit in the three-dimensional model at each period, arrange the volume value and shape parameters corresponding to each disease unit into a sequence in chronological order, calculate the change amplitude by the difference between adjacent elements in the sequence, and combine the change amplitude into a change vector.

[0032] Optionally, determining the displacement trajectory corresponding to the key points on the surface of the target rammed earth site, and generating deformation data of the key points on the surface based on the displacement trajectory, including:

[0033] Continuous three-dimensional laser scanning was performed on key points on the surface of the target rammed earth site to obtain point cloud sequences at multiple time points;

[0034] For each key surface point in the key surface points, the coordinate positions of the key points at different time points are extracted from the point cloud sequence, and the displacement trajectory is calculated by the continuous change of the coordinate positions.

[0035] The displacement trajectories of all surface points are summarized, the average velocity and direction changes of the displacement trajectories in the local area are calculated, and the average velocity and direction changes of all surface points are combined into the deformation data of the key surface points.

[0036] Optionally, after generating valid clues, the method further includes: prioritizing the protection resources of the target rammed earth site based on the potential risk points in the valid clues, and dynamically optimizing the resource allocation by comparing and adjusting the resource requirements of high-risk points with those of low-risk points.

[0037] Secondly, this application provides a tag-based data governance system for cultural relic protection resources, including:

[0038] The first determining module is used to determine the diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site, and to construct a change vector for each diseased unit. The change vector is used to reflect the changes in the size and shape of the diseased unit over time. The module also determines the displacement trajectory corresponding to the surface key points of the target rammed earth site, and generates deformation data of the surface key points based on the displacement trajectory. The deformation data is used to reveal the movement pattern of the surface key points within a preset period.

[0039] The output module is used to associate the change vector with the deformation data at the corresponding spatiotemporal location to form a composite attribute of the node, and to construct a spatiotemporal graph based on the composite attribute. In the spatiotemporal graph, convolution operation is used to mine the spatiotemporal relationship between disease units to output feature data.

[0040] The mapping module is used to map the feature data to the vector space of a pre-constructed disease knowledge graph, wherein the vector space encapsulates the disease evolution pattern and the relationship between causes.

[0041] The second determining module is used to find multiple disease evolution patterns that satisfy a preset condition in the vector space and have a semantic distance to the feature data, and to analyze the disease units corresponding to the feature data according to the attributes of the multiple disease evolution patterns and the causal relationship, so as to determine the semantic label corresponding to the disease unit.

[0042] The generation module is used to generate valid clues based on the semantic tags and the spatiotemporal relationships contained in the feature data.

[0043] Thirdly, this application provides an electronic device, comprising:

[0044] Memory, used to store computer programs;

[0045] A processor is configured to execute the computer program to implement the steps of the tag-based cultural heritage conservation resource data governance method as described in the first aspect above.

[0046] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the tag-based cultural relic protection resource data governance method described in the first aspect above.

[0047] The tag-based data governance method for cultural relic protection resources provided in this application can intuitively capture the temporal changes in the size and shape of diseased units by identifying diseased units in multi-phase 3D models of target rammed earth sites and constructing change vectors for each diseased unit; by determining the displacement trajectories of key points on the surface of the target rammed earth site and generating deformation data, it can clearly grasp the movement patterns of key points on the surface within a set period; by associating change vectors with deformation data at corresponding spatiotemporal locations to form node composite attributes, and then constructing a spatiotemporal graph based on composite attributes and mining the spatiotemporal relationships between diseased units through convolution operations to output feature data, it can achieve deep fusion and related information mining of diseased unit-related data; by mapping feature data to a pre-constructed disease knowledge graph vector space, it can provide knowledge support for subsequent disease evolution pattern matching; by matching disease evolution patterns with semantic distance requirements in the vector space and analyzing and determining semantic tags of diseased units, it can accurately define the attributes of diseased units; and by generating effective clues based on semantic tags and spatiotemporal relationships in feature data, it can provide direct data basis for cultural relic protection decisions.

[0048] Furthermore, a spatiotemporal graph is constructed using key points on the surface of the target rammed earth site as nodes. Then, convolution operations are performed on the spatial and temporal connections of the spatiotemporal graph based on the composite attributes of the nodes to mine the spatial transmission relationships and temporal causal order among the diseased units, thereby outputting feature data representing these relationships and orders. This step accurately constructs a spatiotemporal correlation framework for diseased units, and through targeted convolution operations, deeply mines the inherent laws of disease propagation and evolution, making the output feature data more targeted and interpretable, laying a high-quality data foundation for subsequent semantic label determination and effective clue generation. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a tag-based method for managing cultural relic conservation resource data, provided in an embodiment of this application;

[0051] Figure 2 A schematic diagram illustrating the composite attribute association of a tag-based method for managing cultural relic protection resource data, provided in an embodiment of this application;

[0052] Figure 3 A schematic diagram illustrating the construction of composite node attributes in a tag-based method for governing cultural relic protection resource data, as provided in an embodiment of this application.

[0053] Figure 4 A schematic diagram of feature data mining for a tag-based method for governing cultural relic protection resource data, provided in an embodiment of this application;

[0054] Figure 5 This is a schematic diagram of a tag-based cultural relic protection resource data governance system provided in an embodiment of this application. Detailed Implementation

[0055] In the field of data governance for cultural relic conservation resources, existing tag-based governance methods have significant shortcomings. They largely rely on manual annotation or simple feature matching to determine disease labels, and only combine single-dimensional data for analysis, making it difficult to consider the dynamic changes in cultural relic diseases and the spatiotemporal correlations between different disease units. This directly leads to poor accuracy in disease label determination, an inability to effectively uncover disease evolution patterns, and consequently, a lack of effective clues to support precise cultural relic conservation, thus hindering the in-depth advancement of digital cultural relic conservation work.

[0056] To address the aforementioned issues, this application proposes a label-based data governance method for cultural relic conservation resources. The core of this method lies in data governance focused on the dynamic changes and spatiotemporal correlations of cultural relic damage. By extracting information on the changes in damage units and deformation data of key surface points from multi-period models of cultural relics, the two are correlated to construct a spatiotemporal graph containing composite attributes. The spatiotemporal relationships between damage units are then mined to obtain precise feature data. This is further combined with a pre-constructed damage knowledge graph to match appropriate damage evolution patterns and determine semantic labels for damage units, ultimately generating effective protection clues. This method overcomes the shortcomings of existing technologies in simultaneously addressing dynamic changes and spatiotemporal correlations. It forms a complete and precise governance chain from data association and feature mining to label matching, effectively improving the accuracy of cultural relic conservation resource data governance, providing reliable data support for precise cultural relic protection, and solving the problem that existing technologies cannot support precise protection.

[0057] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] The core of this application is to provide a tag-based method for managing cultural relic conservation resource data, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0059] S101. Determine the diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site, and construct a change vector for each diseased unit.

[0060] Among them, the multi-phase 3D model refers to a collection of multiple 3D models constructed after collecting data on the same rammed earth site at different time points, containing intuitive data such as the surface morphology and spatial structure of the site at different periods. A diseased unit refers to a damaged area with a specific boundary contour on the surface of the site, such as cracks or erosion zones. The change vector is a set composed of data on the size and morphological changes of diseased units at different periods, used to reflect the changes in the size and morphology of diseased units over time.

[0061] Optionally, step S101 may specifically include the following steps:

[0062] S1011. Obtain multi-phase three-dimensional models of the target rammed earth site, extract the boundary contours of key surface points from the multi-phase three-dimensional models, and divide the disease units by comparing the differences between the boundary contours of key surface points in the three-dimensional models of adjacent periods.

[0063] Among them, key points on the surface refer to key locations that can reflect the morphological characteristics of the surface of the site. Their boundary contours are closed lines formed by connecting these key points, which can clearly delineate the range of different areas on the surface of the site.

[0064] In one specific implementation, close-up photogrammetry is first used to acquire image data of the target rammed earth site at different time points, and the data is processed to generate multi-stage high-precision 3D models. Then, an edge detection algorithm is used to extract the boundary contours of key surface points in each stage of the model. This algorithm can automatically identify areas with abrupt changes in grayscale values ​​in the model, thereby locking down key surface points and connecting them to form boundary contours. Finally, the boundary contours of models from adjacent periods are superimposed and compared. When the difference in the position, shape, or range of the contour exceeds a preset threshold, the difference area is identified as the diseased unit.

[0065] For example, taking a rammed earth site as the monitoring object, image data at different time points were acquired using close-up photogrammetry, and then processed by 3D modeling software to generate multiple 3D models. After extracting the boundary contours of key points on the surface of each model using an edge detection algorithm, the contours of adjacent models were compared. It was found that the contour of a certain area of ​​the site shifted by 3 cm in the length direction and expanded by 2 cm in the width direction, exceeding the preset difference threshold of 2 cm. This area was classified as a crack-type disease unit. The above example is only one example of this application. In practical applications, the difference threshold can be adjusted according to needs, and this application does not limit it.

[0066] S1012. Calculate the volume value and shape parameter of each disease unit in the three-dimensional model at each period, arrange the volume value and shape parameter corresponding to each disease unit into a sequence in chronological order, calculate the change range by the difference between adjacent elements in the sequence, and combine the change range into a change vector.

[0067] Here, volume refers to the size of the space occupied by the disease unit in three-dimensional space, and shape parameters include data that describe the shape characteristics of the disease unit, such as its length, width, and depth. Variation range refers to the difference in volume or shape parameters of disease units between adjacent periods, used to quantify the degree of change of the disease unit.

[0068] In one specific implementation, firstly, spatial geometric calculation methods are used to calculate the volume and shape parameters of each disease unit in the 3D model at each stage. The volume value can be automatically calculated by selecting the disease unit entity using the volume measurement function of the 3D modeling software; the shape parameters are obtained by acquiring key dimension data of the boundary contour using the distance measurement function of the software. Next, the volume, length, width, depth, and other data of the same disease unit are arranged in chronological order to form their respective time series. Then, the variation range is obtained by calculating the difference between data from adjacent periods. Finally, the variation ranges corresponding to each parameter are integrated to form the variation vector of the disease unit.

[0069] For example, for the previously divided crack-type disease units, the data of each stage are calculated using 3D modeling software: the volume of the early stage is 0.02 cubic meters, the length is 1.2 meters, the width is 0.05 meters, and the depth is 0.3 meters; the volume of the middle stage is 0.03 cubic meters, the length is 1.25 meters, the width is 0.08 meters, and the depth is 0.449 meters; and the volume of the late stage is 0.04 cubic meters, the length is 1.3 meters, the width is 0.1 meters, and the depth is 0.627 meters.

[0070] After arranging the parameter sequences in chronological order, the magnitude of change is calculated: the volume change from the early to the middle stage is 0.03 - 0.02 = 0.01 cubic meters, the length change is 1.25 - 1.2 = 0.05 meters, the width change is 0.08 - 0.05 = 0.03 meters, and the depth change is 0.449 - 0.3 = 0.149 meters, forming a change vector [0.01 cubic meters, 0.05 meters, 0.03 meters, 0.149 meters]; the change vector from the middle to the late stage is obtained similarly [0.01 cubic meters, 0.05 meters, 0.02 meters, 0.178 meters]. The above example is only one example of this application. In practical applications, shape parameters can be added or reduced according to the type of disease, and this application does not limit this.

[0071] This application solves the problem that traditional methods are unable to intuitively capture the dynamic evolution of site diseases by accurately dividing disease units and quantifying their changes over time. Compared with manual surveys, it not only improves the accuracy and efficiency of data acquisition, but also provides objective and systematic basic data for subsequent disease analysis, demonstrating the advantages of digital technology in cultural relic protection.

[0072] S102. Determine the displacement trajectory corresponding to the key points on the surface of the target rammed earth site, and generate deformation data of the key points on the surface based on the displacement trajectory.

[0073] The displacement trajectory refers to the path formed by the continuous change of the coordinate positions of key points on the surface at different time points, which can intuitively show the movement process of the key points. Deformation data is a set formed by integrating the displacement-related features of key points on the surface, used to reveal the movement patterns and change patterns of the key points within a set time period. The preset period refers to the pre-set time range for monitoring changes in key points on the surface, which can be flexibly adjusted according to monitoring needs.

[0074] Optionally, step S102 may specifically include the following steps:

[0075] S1021. Perform continuous three-dimensional laser scanning on the key points of the surface of the target rammed earth site to obtain point cloud sequences at multiple time points.

[0076] Among them, the point cloud sequence is a dataset containing the three-dimensional coordinate information of key points on the surface at multiple time points, which is collected by a three-dimensional laser scanning device. The point cloud data at each time point can reflect the instantaneous morphology of the surface of the site.

[0077] In one specific implementation, firstly, scanning stations are strategically deployed around the target rammed earth site to ensure comprehensive coverage of all key surface points. Then, a 3D laser scanning device is used to continuously scan the key points on the site surface at fixed time intervals within a preset cycle. During the scanning process, the device quickly acquires the 3D coordinate information of each key point, and the point cloud data collected at different times are organized chronologically to form a complete point cloud sequence.

[0078] S1022. For each key surface point in the key surface points, extract its coordinate position at different time points from the point cloud sequence, and calculate the displacement trajectory through the continuous change of coordinate position.

[0079] Among them, key surface points refer to the core points on the surface that are sensitive to the deformation of the site and can reflect the trend of disease changes. Their displacement changes can be directly related to the development status of the site's diseases.

[0080] In one specific implementation, key surface points are first selected from the key surface points, based on criteria such as the point's sensitivity to deformation and whether it is located in the core area of ​​the defect. Then, a point cloud registration algorithm is used to unify the point cloud data from different time points in the point cloud sequence to the same coordinate system, ensuring the accuracy of coordinate comparison. Next, for each key surface point, its three-dimensional coordinates are extracted from the point cloud data at each time point, and the coordinate information is arranged in chronological order. By connecting the coordinate positions of adjacent time points, the displacement trajectory of the key surface point is formed.

[0081] In practical applications, 20 key surface points were selected from the surface of a rammed earth site, with 10 located around cracks and 10 located in erosion areas. After unifying the coordinate system using a point cloud registration algorithm, the coordinates of one of the key surface points were extracted: the coordinates of the first time point were (10.2m, 8.5m, 3.1m), the coordinates of the second time point were (10.3m, 8.6m, 3.0m), and the coordinates of the third time point were (10.4m, 8.7m, 2.9m). Connecting these coordinates in chronological order formed a displacement trajectory that subsided in a southeast direction. The above example is only one example of this application. In practical applications, the number and criteria for selecting key surface points can be adjusted according to the type of damage to the site, and this application does not limit this.

[0082] S1023. Summarize the displacement trajectories of all surface points, calculate the average velocity and direction changes of the displacement trajectories in the local area, and combine the average velocity and direction changes of all surface points into deformation data of key surface points.

[0083] Here, average velocity refers to the average displacement distance of surface points within a local area per unit time, quantifying the rate of displacement. Directional change refers to the change in the direction of the surface point displacement trajectory over time, reflecting the stability of the displacement. A local area refers to several sub-regions divided according to the surface morphology and distribution of damage at the site; surface points within each region exhibit similar deformation characteristics.

[0084] In this embodiment, the area where the key surface points are located is first divided into several local regions based on the spatial structure and disease distribution of the site. Then, the displacement trajectories of all surface points within each local region are summarized, and the displacement distance and direction information of each trajectory are extracted. The instantaneous velocity of each surface point is calculated based on the displacement distance and the corresponding time interval, and the average of all instantaneous velocities is taken as the average velocity of the region. By comparing the displacement directions at adjacent time points, the frequency and angle of direction changes are statistically analyzed to obtain direction change data. Finally, the average velocity and direction changes of all local regions are integrated to form the deformation data of the key surface points.

[0085] For example, the surface of a rammed earth site is divided into 5 local regions, each containing 30 surface points. After summing the displacement trajectories of surface points within one region, the instantaneous velocity of each surface point is calculated: the instantaneous velocity of point 1 is 0.02 mm / d, the instantaneous velocity of point 2 is 0.03 mm / d, and so on, until point 30, which has an instantaneous velocity of 0.01 mm / d. The average velocity of this region is (0.02 + 0.03 + ... + 0.01) / 30 = 0.02 mm / d. By comparing the displacement directions of adjacent time points, it is found that the direction of the surface points within this region changes twice, with a maximum change angle of 15°. The average velocity and direction changes of the 5 regions are integrated to form the deformation data of key points on the surface of the site.

[0086] This application acquires point cloud data through continuous scanning, accurately tracks the displacement trajectory of key surface points, and generates deformation data. This solves the problem that traditional methods are unable to accurately capture subtle deformations on the surface of archaeological sites. Compared with manual observation, it not only improves the accuracy and efficiency of deformation monitoring, but also objectively reflects the movement pattern of the surface of archaeological sites, providing a reliable basis for risk prediction in cultural relic protection and demonstrating the advantages of digital monitoring technology.

[0087] S103. Associate the change vector with the deformation data at the corresponding spatiotemporal location to form the composite attribute of the node, and construct a spatiotemporal graph based on the composite attribute. In the spatiotemporal graph, use convolution operation to mine the spatiotemporal relationship between disease units to output feature data.

[0088] In this step, the composite attribute of a node refers to the data set formed by integrating the change vector and deformation data at the same spatiotemporal location. A spatiotemporal diagram is a graphical structure that simultaneously contains spatial dimensional relationships and temporal dimensional sequences. The spatial dimension reflects the positional relationships of nodes, while the temporal dimension reflects the changes in node attributes over time. Spatiotemporal relationships include the spatial interactions and temporal order of influence between disease units, and feature data is structured data formed by refining these relationships.

[0089] Specifically, such as Figure 2 As shown, diseased units were pre-determined from the multi-phase three-dimensional models of the target rammed earth site. and disease unit It should be noted that the results of the above two disease units are only examples, and multiple disease units can actually be identified. Through steps S101~S102, the disease units are respectively... and disease unit Change vector [ size, [Morphology], deformation data of key points on the surface of the site. [ Locally, [Direction], according to time series [ …] and spatial location are associated to ultimately form the composite attributes of each node. For example, the composite attribute of node N1 is Among them, the change vector of the diseased unit Includes changes in volume and shape parameters. Deformation data of key points on the site surface. Key points The changes in average speed and direction in the trajectory of motion.

[0090] Optionally, step S103, "constructing a spatiotemporal graph based on composite attributes, and mining the spatiotemporal relationships between disease units through convolution operations in the spatiotemporal graph to output feature data," may specifically include the following steps:

[0091] S1031. Use key points on the surface of the target rammed earth site as nodes to construct a spatiotemporal map.

[0092] Among them, nodes are the basic building blocks of the spatiotemporal graph. Each node corresponds to a key point on the surface of the rammed earth site and carries the composite attribute information corresponding to that key point.

[0093] In one specific implementation, key points on the surface of the target rammed earth site are first screened, including key points within the identified diseased unit area and key points around the diseased unit that may be affected. These surface key points are then used as nodes, and a spatiotemporal graph is constructed based on the identified nodes, such as... Figure 3 As shown, the vertical solid lines in the spacetime diagram represent the same point in time (e.g., ...). The spatial adjacency relationships between different nodes (N1, N2, N3) are represented by the horizontal dashed lines, while the horizontal dashed lines represent the same node at different points in time. arrive The state evolution of a node is based on a spatiotemporal graph constructed from multiple nodes. In this graph, vertical solid lines represent points in time (e.g., ...). The spatial adjacency relationships between different nodes (N1, N2, N3) are represented by the horizontal dashed lines, while the horizontal dashed lines represent the same node at different points in time. arrive The state evolution of ).

[0094] S1032. By performing convolution operations on the composite attributes of nodes based on the spatial and temporal connections of the spatiotemporal graph, the spatial transmission relationship and temporal causal order between disease units are mined, and feature data is output.

[0095] Among them, the feature data is used to characterize the spatial transmission relationship and temporal causal order between disease units.

[0096] Specifically, step S1032 may include the following process: in the spatiotemporal diagram, for each spatial connection, the composite attributes of adjacent nodes are weighted and summed; in the spatiotemporal diagram, for each temporal connection, the composite attributes of the same node at different time points are weighted and summed; the weighted summation results of spatial connections and temporal connections are fused to obtain the updated attributes of each node; the transmission relationship and causal order between disease units are identified through the updated attributes; and the transmission relationship and causal order are combined into feature data.

[0097] In the above steps, spatial connection refers to the relationship between geographically adjacent nodes in the spatiotemporal diagram, used to reflect the spatial location correlation of nodes. Temporal connection refers to the relationship between the same node at different points in time, used to reflect the temporal evolution of node attributes. Spatial transmission relationship refers to the spatial impact of changes in one disease unit on other surrounding disease units, while temporal causal order refers to the temporal order in which changes in different disease units affect each other.

[0098] Specifically, based on the constructed spatiotemporal graph, the model performs the core "convolution operation," that is, simultaneously performing "spatial convolution" on the spatial neighbor information of the nodes. And performing "temporal convolution" on the node's own historical state ( ).

[0099] First, spatial convolution is performed based on the spatiotemporal graph, targeting each time point in the spatiotemporal graph. The spatial connections of the nodes below retrieve the composite attributes of the adjacent nodes at the same point in time for each node. Then, assign corresponding spatial weights to the composite attribute A of each adjacent node. The spatial weights are used to reflect the degree of influence of different neighbors on the current node. Finally, the spatial weights are summed with the composite attribute A of the neighboring nodes to obtain the spatial convolution result of the node, i.e. For example, in At time N, the spatial convolution of N2 integrates the combined properties of N1 and N3. ;arrive At time N, the spatial convolution of N1 will be associated with the spatial convolution of N2. This covers the spatial association information of all nodes.

[0100] Temporal convolution, on the other hand, covers the temporal connections of all nodes: based on the temporal connections in the spatiotemporal graph, for example, N1 in... The horizontal dashed line between them retrieves the composite attributes of each node at an earlier time point. And assign composite attributes to each historical time point. Assign corresponding time weights W t The time weight is used to reflect the degree of influence of different historical states on the current node. The composite attributes of each historical time point are... Multiply by the corresponding time weight Summation after the summation, i.e. .for example, It will be related to itself A, It will be related to itself , A, It is a coefficient used to quantify the degree of influence of states at different historical moments.

[0101] After each node completes spatial and temporal convolutions, a fusion operation is performed, combining the characteristics of the two convolution results. First, the spatial and temporal convolution results are standardized separately to ensure they are of the same order of magnitude. Then, based on the model's preset fusion ratio or by adaptively adjusting the weights during training, the two results are weighted and summed to ultimately generate the updated attributes for each node. The correspondence is as follows: ( ),in, Represents a node At any moment The updated attributes include both "spatial information of surrounding nodes at the current time" and "time evolution information of itself from the past to the present", which are node attributes that integrate the entire spatiotemporal context. This indicates a fusion operation.

[0102] Next, by analyzing all update attributes rich in spatiotemporal information... The system can decode and output the final feature data. From a spatial perspective, this is achieved by comparing different nodes. The changes are related, for example, of After the change, of The resulting corresponding fluctuations allow for clear identification. This spatial transmission relationship intuitively reflects the path of diseased units' influence in terms of location; from a temporal perspective, by tracing the same node... exist The order of evolution, for example of Change precedes That will make it clear The temporal causal sequence accurately reflects the temporal impact of disease units.

[0103] Finally, the system combines these two relationships to form the final feature data. (Transmission, order) comprehensively and accurately characterizes the spatiotemporal correlation patterns of disease units.

[0104] This application constructs composite attributes by associating change vectors and deformation data, and mines the spatiotemporal relationships of disease units by relying on spatiotemporal graphs and convolution operations, thus solving the problem that traditional methods are difficult to integrate multi-dimensional data and cannot accurately identify the correlation patterns between diseases.

[0105] S104. Map the feature data to the vector space of the pre-constructed disease knowledge graph.

[0106] The disease knowledge graph is a structured knowledge base that integrates information related to various diseases of rammed earth sites, including disease types, evolution processes, and influencing factors. The vector space is a multi-dimensional data space formed by converting the nodes and edges in the knowledge graph into numerical forms. It encapsulates disease evolution patterns and causal relationships, with each dimension corresponding to a disease-related feature. Disease evolution patterns refer to the typical path of disease change from occurrence to development, while causal relationships refer to the association between the factors that cause the occurrence or development of the disease and the disease itself.

[0107] Specifically, such as Figure 4As shown, for example, the obtained feature data includes spatial transmission intensity, temporal order, and weight vector, where from arrive The spatial conduction intensity at a point is strong. Click The spatial conduction intensity is medium; the time sequence is from... arrive For speed, from arrive For slow; weight vector The calculation is performed dimensionally. Then, based on these feature data, the numerical weight of each dimension is calculated, and a vector space for the disease knowledge graph is constructed. The calculated numerical weights are numerically matched with the corresponding dimensions in the vector space to achieve the mapping of feature data. For example, the evolutionary mode corresponding to the F-mapping point is determined by calculating the Euclidean distance between the F-mapping point and multiple evolutionary modes. The specific implementation process is as follows:

[0108] Optionally, step S104 may specifically include the following steps:

[0109] S1041. Obtain the pre-constructed disease knowledge graph, extract the numerical representations of all nodes and edges from the disease knowledge graph, and organize the numerical representations of all nodes and edges into a vector space.

[0110] In this knowledge graph, nodes are the basic units that carry specific information, including entities such as disease types and influencing factors. Edges are the lines connecting nodes, used to represent the relationships between them, such as the association between diseases and their causes, or the evolutionary relationships between diseases. Numerical representation transforms the semantic information of nodes and edges into computable numerical forms, enabling the analysis of abstract relationships through data computation.

[0111] In one specific implementation, a pre-constructed disease knowledge graph is first obtained. This graph is constructed by collecting a large amount of data on disease cases of rammed earth sites, sorting out information such as disease types, such as erosion expansion, collapse evolution, weathering and erosion, evolution paths, and influencing factors, and adopting a graph structure.

[0112] Subsequently, using a word embedding algorithm, each node and edge in the knowledge graph is transformed into a fixed-dimensional numerical vector. Nodes correspond to disease entities in different evolutionary patterns within the graph, and edges correspond to the associations between nodes and F-mapping points. For example, ... Figure 4 As shown, the "Evolution Mode: Erosion Expansion" node is transformed into a numerical vector consisting of dimensions x and y. Similarly, the "Evolution Mode: Collapse Evolution" and "Evolution Mode: Weathering and Erosion" nodes are also transformed into vectors of the same dimensions. Furthermore, the edges between these disease nodes and the core mapping entity "F Mapping Point" representing the disease association in the diagram are shown below. It is also transformed into a numerical vector of the corresponding dimension, thereby realizing the quantification of semantic information.

[0113] Finally, the numerical vectors of all nodes and edges are arranged in a unified two-dimensional format, i.e., dimensions x and y, to form... Figure 4 The vector space shown contains the relationship between disease evolution patterns and their causes. It shows the positions of different disease nodes in the vector space representing erosion expansion, collapse evolution, and weathering, as well as the edges between them and the F-mapping points. This is precisely the visual representation of numerical vectors in two-dimensional space.

[0114] S1042. For the feature data, calculate the numerical weight of each dimension, and realize the mapping of the feature data by matching the numerical weight of the feature data with the corresponding dimension in the vector space.

[0115] The feature data includes the spatial transmission relationships, temporal causal order, and corresponding weight information of disease units. Numerical weights are weight vectors obtained by quantifying the information in each dimension of the feature data, used to reflect the degree of influence of different information. Numerical matching compares the numerical weights of the feature data with the dimensions of the disease knowledge graph vector space to find the corresponding mapping positions within the space.

[0116] In this embodiment, the feature data F is first analyzed by decomposing its information dimensions: the strength of spatial transmission relationships and the rate of temporal causal order are extracted first, for example... For strong hierarchical spatial transmission relationships, For intermediate-level spatial transmission relationships; the rate of temporal causal order. For speed, It is slow.

[0117] Secondly, this information is quantified using weighting methods, such as the analytic hierarchy process (AHP). The spatial transmission intensity is then converted into strong, medium, and strong categories. The numerical value, the rate of change over time, is converted from fast to slow. The numerical values ​​are then integrated to obtain the numerical weight vector of the feature data F, such as a two-dimensional vector [0.65, 0.5], corresponding to the x and y dimensions of the vector space.

[0118] Next, numerical matching is performed: first, the dimensional rules of the disease knowledge graph vector space are determined, such as... Figure 4 In the vector space, the x-dimensional coordinate corresponds to spatial correlation, and the y-dimensional coordinate corresponds to the rate of change over time. Then, using methods such as Euclidean distance, the similarity between the numerical weight vector of the feature data F and the vectors of all nodes in the vector space is calculated. For example, to calculate the similarity between a two-dimensional vector […]. ] and erosion expansion node vector [ ], Collapse Evolution Node Vector [ The distance to the weight vector is used to find the position in space that is closest to the weight vector; this position is the feature data. corresponding Mapping point.

[0119] Finally, with With the mapping point as the core, the edges connected to it in the associated vector space, such as... Figure 4 In By using the association relationships corresponding to the edges, we can retrieve... The disease evolution pattern matched by the mapping point, for example, through By associating it with erosion expansion and combining it with the attribute information of nodes in the vector space, we can further obtain the causal relationship corresponding to this evolutionary pattern.

[0120] In practical applications, taking the feature data F in the figure as an example, it includes "spatial transmission: strong..." ,middle "Time sequence: fast" ,slow First, calculate the numerical weights: quantize strong conduction to 0.8 and medium conduction to 0.5, taking the average to obtain a spatial dimension value of 0.65; quantize rapid change to 0.9 and slow change to 0.3, taking the average to obtain a time dimension value of 0.5, thus obtaining the final weight vector. .

[0121] Subsequently, numerical matching is performed: the x-dimensional vector space corresponds to the "spatial correlation" and the y-dimensional vector corresponds to the "time rate of change," and the weight vector is calculated. With the spatial erosion expansion node vector The Euclidean distance is , and the collapse evolution node vector The distance is The nearest matching position is the F-mapped point in the graph. Finally, the edges connected through the F-mapped point... The corresponding evolutionary pattern was found to be "erosion expansion". At the same time, through the attributes of the "erosion expansion" node in the vector space, the cause of this pattern was found to be "long-term rainwater infiltration".

[0122] This application, through a combination of information dimension decomposition, quantified weighting, and vector space matching techniques, breaks through the limitations of the separation of spatial and temporal information in traditional disease analysis, and solves the problem of the difficulty in accurately linking the evolution patterns and causes of diseases in rammed earth archaeological sites. It transforms discrete disease characteristics into structured information that can be linked into a knowledge graph, upgrading the implicit relationships of disease evolution from empirical inference to data-driven precise retrieval, thus improving the accuracy and efficiency of matching.

[0123] S105. In the vector space, find multiple disease evolution patterns that satisfy the preset conditions in terms of semantic distance from the feature data, and analyze the disease units corresponding to the feature data based on the attributes and causal relationships of the multiple disease evolution patterns to determine the semantic labels corresponding to the disease units.

[0124] Semantic distance refers to the degree of numerical difference between feature data and disease evolution patterns in vector space, used to measure their similarity. This application can use the Euclidean distance between feature data and multiple disease evolution patterns in vector space as semantic distance. Preset conditions refer to pre-defined semantic distance judgment criteria used to filter out disease evolution patterns that match the feature data. Semantic labels are summary identifiers of core information such as the type, cause, and risk level of disease units, facilitating rapid understanding of the key attributes of disease units.

[0125] Optionally, step S105 may specifically include the following steps:

[0126] S1051. In the vector space, calculate the numerical difference between the feature data and each disease evolution mode, and select multiple disease evolution modes whose numerical differences meet the preset threshold.

[0127] In one specific implementation, the numerical vectors corresponding to all preset disease evolution modes in the vector space are first obtained. Then, the numerical difference between the feature data vector and each evolution mode vector is calculated using the Euclidean distance formula, as shown in equation (1):

[0128] (1)

[0129] in, Indicates numerical differences, Represents the feature data vector of the first The numerical value of dimension, The vector representing the disease evolution pattern. The numerical value of dimension, This represents the vector dimension. This formula comprehensively reflects the overall differences across multiple dimensions.

[0130] Finally, the calculated numerical differences are compared with a preset threshold, and multiple disease evolution patterns with numerical differences less than or equal to the threshold are selected.

[0131] S1052. For the disease units corresponding to the feature data, extract the common attributes and related causal relationships of multiple disease evolution patterns. By comparing the attributes of the disease units with the common attributes and causal relationships item by item, pattern matching is achieved, and semantic labels are selected and assigned to the disease units based on the matching results.

[0132] Common attributes refer to the features shared by multiple selected disease evolution patterns, such as the stage characteristics of disease development and spatial distribution patterns. Related causal relationships refer to the information on factors associated with these evolution patterns that lead to the occurrence and development of the disease. Pattern matching involves comparing the actual attributes of disease units with the common attributes and causal relationships of the evolution patterns to confirm the degree of fit.

[0133] Specifically, firstly, common attributes are extracted from multiple selected disease evolution patterns. By statistically analyzing the overlap of attributes across patterns, core shared features are determined. Simultaneously, the causal relationships between these patterns are extracted, identifying frequently occurring causal types and their strengths. Next, the actual attributes of disease units are listed, including disease variation characteristics and spatial relationships, and compared item by item with the common attributes and causal relationships of the evolution patterns, recording the number and importance of matching items. Finally, the matching level is determined based on the degree of fit, and corresponding semantic tags are selected from a pre-defined semantic tag library based on the matching level and assigned to the disease unit.

[0134] For example, such as Figure 4 As shown, within the vector space, the Euclidean distance between the feature data and each disease evolution mode is calculated, and multiple disease evolution modes that meet the preset conditions are selected, such as erosion expansion and collapse evolution. Subsequently, common attributes are extracted from these two modes, identifying core shared characteristics such as "gradual expansion of the disease range" and "a development trend from the surface to the depths." Simultaneously, relevant causal relationships are extracted, revealing that rain erosion and weathering are frequently associated causative factors for both modes, significantly promoting disease development, while the correlation between earthquakes and these two modes is weak. Next, the actual attributes of the disease units corresponding to the feature data are listed, including continuous expansion of the disease area, initial surface damage followed by inward extension, and significant influence from environmental factors.

[0135] Next, the actual attributes of the diseased units are compared item by item with the extracted common attributes and causal relationships. This confirms that the expansion characteristics of the diseased units highly match the common attributes of the two types of patterns, and that their actual impact from rain erosion and weathering corresponds to the causal relationships. Based on this degree of fit, a matching level is determined, and corresponding semantic tags are selected from a pre-set semantic tag library. Finally, the diseased unit is labeled "erosion-collapse composite type." The specific format of the semantic tag may include [type, causal factor, risk level]. The above example is merely one example of this application. In practical applications, the content of the semantic tag library can be enriched according to the type of disease; this application does not limit this.

[0136] This application determines semantic labels by quantitatively screening and matching disease evolution patterns, and combining common attributes and causal relationship analysis. This solves the problem that traditional disease labeling can only describe surface phenomena and lacks in-depth information. Compared with manual labeling, it not only improves the accuracy and standardization of labels, but also gives disease units more comprehensive attribute information, demonstrating the advantages of data-driven and knowledge integration in disease analysis.

[0137] S106. Generate valid clues based on semantic labels and the spatiotemporal relationships contained in the feature data.

[0138] In this step, effective clues are structured guiding information formed by integrating core information, evolutionary patterns, and potential risks of diseased units, providing a clear basis for cultural relic protection decisions. Potential risk points refer to areas inferred from spatiotemporal relationships and evolutionary patterns that may be affected by current diseases or may subsequently worsen.

[0139] Optionally, step S106 may specifically include the following steps:

[0140] S1061. Extract spatiotemporal relationships from feature data and combine spatiotemporal relationships with semantic labels.

[0141] The extraction of spatiotemporal relationships refers to separating key information such as the spatial location associations and temporal order of influence between disease units from feature data. The combination process involves associating and integrating spatiotemporal relationships with information such as disease type, causes, and risk levels contained in semantic tags to form more comprehensive disease description information.

[0142] In this embodiment, firstly, a data parsing algorithm is used to extract spatiotemporal relationships from feature data, clarifying the spatial distribution associations and temporal evolution order of disease units. Then, association mapping rules are established to match the extracted spatiotemporal relationships with the various attributes of semantic tags. Rule validation ensures the rationality of the association, ultimately forming a combined information containing disease attributes and spatiotemporal features.

[0143] In practical applications, the semantic label of a certain disease unit is "Type: stress-driven crack; Cause: stress change due to rainwater infiltration + local erosion; Risk level: high". The spatiotemporal relationship extracted from the feature data is "spatially adjacent to the upper erosion zone, temporally crack expansion lags behind the deformation of the erosion zone". According to the association mapping rule, the two are combined to form the combined information of "high-risk stress-driven crack, caused by stress change due to rainwater infiltration and local erosion, spatially adjacent to the upper erosion zone, temporally crack expansion lags behind the deformation of the erosion zone". The above example is only one example of this application. In practical applications, the combination rule can be adjusted according to the information complexity, and this application does not limit this.

[0144] S1062. For each disease unit, identify the corresponding evolutionary pattern based on the combination results, infer potential risk points through spatiotemporal relationships, and integrate disease units, evolutionary patterns, and potential risk points into effective clues.

[0145] Among these, evolutionary pattern recognition refers to matching typical development paths corresponding to disease units based on combined information. Potential risk point inference combines spatiotemporal relationships and evolutionary patterns to predict areas where disease expansion or the occurrence of new diseases may occur.

[0146] In one specific implementation, the combined information is first compared with a pre-defined evolutionary pattern library to identify the disease evolutionary pattern that best matches the combined information. Then, based on spatial correlations and temporal trends in spatiotemporal relationships, combined with the development patterns of the evolutionary patterns, a trend prediction algorithm is used to infer potential risk points, clarifying their locations and possible deterioration trends. Finally, the core information of the disease unit, the identified evolutionary patterns, and the inferred potential risk points are structurally integrated to form a clear and effective set of clues.

[0147] For example, based on the combined information formed in S1061, after comparison with the evolutionary pattern library, the corresponding evolutionary pattern was identified as "the continuous development of the erosion zone leads to stress concentration in the surrounding area, which in turn triggers crack propagation and extends to the weak area." Based on the spatial correlation of "the crack is adjacent to the upper erosion zone" and the temporal trend of "the crack continues to propagate," a trend prediction algorithm was used to infer that the potential risk point is "the historical repair area 1 meter southeast of the crack," which may be affected by crack propagation.

[0148] Integrating this information to form a valid clue: "High-risk stress-driven fractures are caused by stress changes resulting from rainwater infiltration and localized erosion. Spatially, they are adjacent to the upper erosion zone, and temporally, the fracture expansion lags behind the deformation of the erosion zone. Currently, they follow the evolutionary pattern of 'erosion-induced fracture expansion'. The potential risk point is the historical repair area 1 meter southeast of the fracture. It is recommended to prioritize the reinforcement of this fracture and the associated erosion zone, while strengthening the monitoring of the historical repair area."

[0149] This application integrates semantic tags and spatiotemporal relationships to accurately identify evolutionary patterns and infer potential risk points, solving the problems of scattered and untargeted clues in traditional cultural relic protection. Compared with manual sorting, it not only improves the systematicness and accuracy of clues, but also provides clear guidance for protection actions, demonstrating the advantages of data integration and trend prediction in cultural relic protection decision-making.

[0150] Optionally, after generating valid clues, the method further includes: prioritizing the protection resources of the target rammed earth site based on the potential risk points in the valid clues, and dynamically optimizing the resource allocation by comparing and adjusting the resource requirements of high-risk points with those of low-risk points.

[0151] Specifically, the process begins by extracting all potential risk points from valid leads, clarifying key information such as the scope of harm, rate of deterioration, and consequences of each risk point. A risk assessment indicator system is then established, covering dimensions such as risk level, urgency of restoration, and conservation value. The analytic hierarchy process (AHP) is used to calculate a comprehensive score for each risk point. Based on the comprehensive scores, the risk points are prioritized, with high-scoring risk points designated as priority protection targets.

[0152] Next, the resource requirements for each priority risk point are calculated, including the required resource type, quantity, and usage cycle. Finally, by comparing the resource requirements of high-risk and low-risk points, limited resources are prioritized for allocation to high-priority risk points, while some flexible resources are reserved to cope with emergencies. A dynamic monitoring mechanism is established to adjust the allocation plan in real time based on changes in risk points and resource usage effectiveness, thereby achieving dynamic optimization of resource allocation.

[0153] In practical application, the effective clues of a certain rammed earth site contain three potential risk points: Risk Point A: Stress-driven fissures, high risk, potentially causing local collapse; Risk Point B: Slight erosion, medium risk, slow development; Risk Point C: Small erosion zone, low risk, limited impact. After establishing a risk assessment index system, the comprehensive score was calculated: Risk Point A scored 85 points, Risk Point B scored 60 points, and Risk Point C scored 35 points, with the priority ranking as A > B > C.

[0154] Next, the resource requirements are calculated: Risk point A requires 5 tons of grouting material, 4 professional construction workers, and 2 monitoring devices, with a cycle of 15 days; Risk point B requires 2 tons of repair material and 2 construction workers, with a cycle of 10 days; Risk point C requires 1 ton of simple filling material and 1 construction worker, with a cycle of 5 days. When allocating resources, priority is given to meeting all resource requirements for risk point A, then allocating the necessary resources to risk point B, and finally, based on the remaining resources, meeting the needs of risk point C.

[0155] After 10 days of operation, monitoring revealed that the rate of deterioration of risk point B accelerated, and the comprehensive score rose to 75 points. Subsequently, resource allocation was adjusted, and one construction worker and 0.5 tons of materials were transferred from the reserved resources of risk point C to support risk point B, thereby achieving dynamic optimization of resources.

[0156] Figure 5 This is a schematic diagram illustrating a specific implementation of a tag-based cultural relic protection resource data governance system provided in this application. (Refer to...) Figure 5 The system may include:

[0157] The first determining module 51 is used to determine the diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site, and to construct a change vector for each diseased unit. The change vector is used to reflect the changes in the size and shape of the diseased unit over time. The module also determines the displacement trajectory corresponding to the surface key points of the target rammed earth site, and generates deformation data of the surface key points based on the displacement trajectory. The deformation data is used to reveal the movement pattern of the surface key points within a preset period.

[0158] Output module 52 is used to associate the change vector with the deformation data at the corresponding spatiotemporal location to form the composite attribute of the node, and to construct a spatiotemporal graph based on the composite attribute. In the spatiotemporal graph, convolution operation is used to mine the spatiotemporal relationship between disease units to output feature data.

[0159] The mapping module 53 is used to map feature data to the vector space of a pre-constructed disease knowledge graph. The vector space encapsulates the disease evolution pattern and the relationship between causes.

[0160] The second determining module 54 is used to find multiple disease evolution patterns in the vector space that satisfy the preset conditions of semantic distance with the feature data, and to analyze the disease units corresponding to the feature data according to the attributes and causal relationships of the multiple disease evolution patterns in order to determine the semantic labels corresponding to the disease units.

[0161] The generation module 55 is used to generate valid clues based on semantic labels and the spatiotemporal relationships contained in the feature data.

[0162] This application provides a tag-based cultural relic protection resource data governance system to implement the aforementioned tag-based cultural relic protection resource data governance method. Therefore, the specific implementation of the tag-based cultural relic protection resource data governance system can be found in the embodiment section of the tag-based cultural relic protection resource data governance method above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0163] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described tag-based cultural relic protection resource data governance methods.

[0164] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described tag-based cultural relic protection resource data governance methods.

[0165] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0166] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the tag-based cultural relic protection resource data governance method.

[0167] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0168] The foregoing has provided a detailed description of a tag-based method and system for managing cultural relic protection resource data. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A tag-based method for managing cultural relic conservation resource data, characterized in that, include: The diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site are identified, and a change vector is constructed for each diseased unit. The change vector is used to reflect the changes in the size and shape of the diseased unit over time. The displacement trajectory corresponding to the surface key points of the target rammed earth site is determined, and deformation data of the surface key points is generated based on the displacement trajectory. The deformation data is used to reveal the movement pattern of the surface key points within a preset period. The change vector is associated with the deformation data at the corresponding spatiotemporal location to form a composite attribute of the node, and a spatiotemporal graph is constructed based on the composite attribute. In the spatiotemporal graph, the spatiotemporal relationship between disease units is mined through convolution operation to output feature data. The feature data is mapped to the vector space of a pre-constructed disease knowledge graph, which encapsulates the disease evolution pattern and the relationship between causes. Within the vector space, multiple disease evolution patterns that satisfy a preset condition in semantic distance from the feature data are searched. Based on the attributes of the multiple disease evolution patterns and the causal relationship, the disease units corresponding to the feature data are analyzed to determine the semantic labels corresponding to the disease units. Based on the semantic tags and the spatiotemporal relationships contained in the feature data, valid clues are generated.

2. The method according to claim 1, characterized in that, A spatiotemporal graph is constructed based on the aforementioned composite attributes. Convolutional operations are then used to mine the spatiotemporal relationships between diseased units within this graph, outputting feature data, including: Using key points on the surface of the target rammed earth site as nodes, a spatiotemporal map is constructed; By performing convolution operations on the spatial and temporal connections of the spatiotemporal graph based on the composite attributes of the nodes, the spatial transmission relationship and temporal causal order between the disease units are mined, and feature data is output. The feature data is used to characterize the spatial transmission relationship and temporal causal order between the disease units.

3. The method according to claim 2, characterized in that, By performing convolution operations on the spatial and temporal connections of the spatiotemporal graph based on the composite attributes of the nodes, the spatial transmission relationships and temporal causal order among the disease units are mined, and feature data is output, including: In the spatiotemporal graph, for each spatial connection, the composite attributes of adjacent nodes are weighted and summed. In the spatiotemporal graph, for each temporal connection, the composite attributes of the same node at different time points are weighted and summed. The weighted summation results of spatial connections and time connections are fused to obtain the updated attributes of each node. The transmission relationships and causal order between disease units are identified through the updated attributes, and the transmission relationships and causal order are combined into feature data.

4. The method according to claim 1, characterized in that, Mapping the feature data to the vector space of a pre-constructed disease knowledge graph includes: Obtain a pre-constructed disease knowledge graph, extract the numerical representations of all nodes and edges from the disease knowledge graph, and organize the numerical representations of all nodes and edges into a vector space. For the feature data, the numerical weight of each dimension is calculated, and the feature data is mapped by matching the numerical weight of the feature data with the corresponding dimension in the vector space.

5. The method according to claim 1, characterized in that, Within the vector space, multiple disease evolution patterns that satisfy a preset condition in semantic distance to the feature data are searched. Based on the attributes of these multiple disease evolution patterns and the causal relationships, the disease units corresponding to the feature data are analyzed to determine the semantic labels corresponding to the disease units, including: Within the vector space, the numerical difference between the feature data and each disease evolution mode is calculated, and multiple disease evolution modes whose numerical differences satisfy a preset threshold are selected. For the disease units corresponding to the feature data, common attributes and related causal relationships of the multiple disease evolution patterns are extracted. Pattern matching is achieved by comparing the attributes of the disease units with the common attributes and causal relationships item by item, and semantic labels are selected and assigned to the disease units based on the matching results.

6. The method according to claim 1, characterized in that, Based on the semantic tags and the spatiotemporal relationships contained in the feature data, valid clues are generated, including: Spatiotemporal relationships are extracted from the feature data, and the spatiotemporal relationships are combined with the semantic tags; For each diseased unit, the corresponding evolutionary pattern is identified based on the combined results, potential risk points are inferred through the spatiotemporal relationship, and the diseased unit, the evolutionary pattern, and the potential risk points are integrated into effective clues.

7. The method according to claim 1, characterized in that, Identify the diseased units in the multi-phase 3D model corresponding to the target rammed earth site, and construct a change vector for each diseased unit, including: A multi-phase three-dimensional model of the target rammed earth site is obtained, and the boundary contours of key surface points are extracted from the multi-phase three-dimensional model. The disease units are divided by comparing the differences between the boundary contours of key surface points in the three-dimensional models of adjacent periods. Calculate the volume value and shape parameters of each disease unit in the three-dimensional model at each period, arrange the volume value and shape parameters corresponding to each disease unit into a sequence in chronological order, calculate the change amplitude by the difference between adjacent elements in the sequence, and combine the change amplitude into a change vector.

8. The method according to claim 1, characterized in that, Determine the displacement trajectories corresponding to key points on the surface of the target rammed earth site, and generate deformation data of the key points on the surface based on the displacement trajectories, including: Continuous three-dimensional laser scanning was performed on key points on the surface of the target rammed earth site to obtain point cloud sequences at multiple time points; For each key surface point in the key surface points, the coordinate positions of the key points at different time points are extracted from the point cloud sequence, and the displacement trajectory is calculated by the continuous change of the coordinate positions. The displacement trajectories of all surface points are summarized, the average velocity and direction changes of the displacement trajectories in the local area are calculated, and the average velocity and direction changes of all surface points are combined into the deformation data of the key surface points.

9. The method according to claim 1, characterized in that, After generating valid clues, the process also includes: prioritizing the protection resources of the target rammed earth site based on the potential risk points in the valid clues, and dynamically optimizing the resource allocation by comparing and adjusting the resource requirements of high-risk points with those of low-risk points.

10. A tag-based data governance system for cultural relic protection resources, characterized in that, include: The first determining module is used to determine the diseased units in the multi-phase three-dimensional model corresponding to the target rammed earth site, and to construct a change vector for each diseased unit. The change vector is used to reflect the changes in the size and shape of the diseased unit over time. The displacement trajectory corresponding to the surface key points of the target rammed earth site is determined, and deformation data of the surface key points is generated based on the displacement trajectory. The deformation data is used to reveal the movement pattern of the surface key points within a preset period. The output module is used to associate the change vector with the deformation data at the corresponding spatiotemporal location to form a composite attribute of the node, and to construct a spatiotemporal graph based on the composite attribute. In the spatiotemporal graph, convolution operation is used to mine the spatiotemporal relationship between disease units to output feature data. The mapping module is used to map the feature data to the vector space of a pre-constructed disease knowledge graph, wherein the vector space encapsulates the disease evolution pattern and the relationship between causes. The second determining module is used to find multiple disease evolution patterns that satisfy a preset condition in the vector space and have a semantic distance to the feature data, and to analyze the disease units corresponding to the feature data according to the attributes of the multiple disease evolution patterns and the causal relationship, so as to determine the semantic label corresponding to the disease unit. The generation module is used to generate valid clues based on the semantic tags and the spatiotemporal relationships contained in the feature data.

Images