An intelligent recommendation method for cultural relic display information based on a knowledge graph

Abstract

The application provides a kind of based on knowledge graph's cultural relic display information intelligent recommendation method, comprising: obtaining the entity data of target cultural relic, knowledge relationship data, physical detection data, digital image data, information to be displayed and equipment data, unified cultural relic identity and space coordinate, constructs the cultural relic knowledge graph including cultural relic entity node, knowledge node, evidence node, display information node and knowledge relationship edge, and configures physical evidence condition for physical verifiable knowledge relationship edge;Determine the candidate knowledge relationship edge set and the candidate display information set according to user request, generate evidence determination rule and candidate observation action sequence;According to the content of the interface to be verified recommendation, establish display light path set and display pollution propagation relationship, determine target observation action sequence, interface light emission code and synchronous acquisition timing, collect display frame, dark frame and complementary frame, strip display light reflection component, obtain purified observation data.

Description

Technical Field

[0001] This invention relates to the field of digital display and information recommendation technology for cultural relics, and in particular to an intelligent recommendation method for cultural relic display information based on knowledge graphs. Background Technology

[0002] With the development of digital museum displays, interactive screens, mobile guide terminals, and augmented reality terminals are widely used for displaying cultural relic information. Existing technologies typically build databases or knowledge graphs based on data such as the name of the cultural relic, its age, material, historical events, and related figures, and recommend relevant display content according to user requests.

[0003] However, knowledge about cultural relics originates from various sources, including collection archives, archaeological reports, restoration records, expert research, physical testing, and digital images. These different data sources vary in terms of the relic's identity, creation time, spatial location, and reliability. Existing recommendation methods often focus on knowledge association and content retrieval, lacking verification of the correspondence between knowledge conclusions and physical evidence. This can easily lead to the display of unverified, conflicting, or outdated information.

[0004] Some knowledge relationships among cultural relics require verification based on spectral response, polarization response, surface texture, or local geometry. However, existing systems typically cannot automatically generate illumination bands, incident angles, polarization states, observation poses, and exposure parameters based on the knowledge relationships to be verified. Furthermore, they struggle to balance the safety of cultural relic irradiation with the feasibility of equipment operation. In addition, the display light generated by the display terminal may be reflected by the display case and the surface of the cultural relic before entering the image acquisition device, overlapping with the target detection band. This can cause ghosting, color shifts, or localized highlights in the observation data, affecting the judgment of physical evidence.

[0005] Therefore, existing technologies for recommending information on cultural relics still suffer from problems such as inconsistent correlation of multi-source data, lack of physical evidence to verify knowledge relationships, difficulty in automatically determining observation parameters, and difficulty in effectively separating light pollution from displays, resulting in insufficient authenticity, reliability, and traceability of the recommendation results. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent recommendation method for cultural relic display information based on knowledge graphs, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A knowledge graph-based intelligent recommendation method for cultural relic display information includes the following steps: S1. Obtain the entity data, knowledge relationship data, physical inspection data, digital image data, information to be displayed and equipment data of the target cultural relic, unify the identity and spatial coordinates of the cultural relic, construct a cultural relic knowledge graph containing cultural relic entity nodes, knowledge nodes, evidence nodes, display information nodes and knowledge relationship edges, and configure physical evidence conditions for physically verifiable knowledge relationship edges. S2. Receive user requests, determine the candidate knowledge relationship edge set and candidate display information set based on the request topic and the executability of physical evidence conditions, convert physical evidence conditions into evidence judgment rules, and generate a candidate observation action sequence set. S3. Generate the recommended interface content to be verified and the interface luminescence description data based on the candidate display information set. Combine the spatial relationship between the display terminal, the reflective surface of the display case, the surface of the cultural relic and the image acquisition device to establish the display light path set and obtain the display pollution propagation relationship and the observation pollution evaluation results. S4. Based on the evidence judgment rules, the candidate observation action sequence set and the observation pollution evaluation results, determine the target observation action sequence, interface luminescence coding and synchronous acquisition timing, acquire display frames, dark frames and complementary frames, strip the display light reflection component, and obtain purification observation data and quality label. S5. Based on the evidence judgment rules, extract physical features and judge evidence from the purified observation data, update the recommendation propagation weight and verification status of the knowledge relationship edges, generate a credible recommendation subgraph, and output the recommendation results of the target cultural relic accordingly.

[0008] S1 specifically includes: acquiring the basic attributes, historical research, physical inspection, digital images, and data on the target cultural relic, including data on the objects to be displayed and the equipment; unifying the identity with a unique cultural relic number; completing the marking of the source, time, and spatial coordinates; forming a unified cultural relic dataset and a set of equipment spatial configurations; establishing a multi-agent system composed of knowledge graphs, evidence compilation, display control, observation control, pollution purification, and recommendation decision-making agents; transmitting input, output, and processing status through a shared task status table; constructing a cultural relic knowledge graph based on the unified cultural relic dataset, including cultural relic entity nodes, knowledge nodes, evidence nodes, display information nodes, and knowledge relationship edges; configuring physical evidence conditions and cultural relic security boundaries for physically verifiable knowledge relationship edges; and writing conflicting knowledge relationships and equipment spatial relationships.

[0009] S2 specifically includes: receiving user requests and performing topic mapping; starting from the cultural relic entity node corresponding to the target cultural relic, selecting candidate knowledge relationship edges based on semantic matching, completeness of displayed information, and executability of physical evidence conditions to form a candidate display information set; reading the physical evidence conditions of the candidate knowledge relationship edges, determining the target evidence area, control area, physical features to be measured, necessary features, threshold range, minimum evidence confidence, and interference exclusion requirements to form an evidence judgment rule set; generating parameter combinations based on the evidence judgment rule set, equipment spatial configuration set, and cultural relic safety boundary, eliminating parameter combinations that exceed the irradiation dose, exceed the equipment range, have collision obstruction, or do not meet the spatial resolution and signal-to-noise ratio requirements, and arranging the retained combinations into a candidate observation action sequence set.

[0010] S3 specifically includes: generating recommended interface content to be verified based on the candidate display information set; dividing pixel areas with the same display color, luminous intensity, refresh sequence, and display status into luminous units to form interface luminous description data; tracing the propagation path from the display terminal through the reflective surface of the display case and the surface of the cultural relic to the image acquisition device based on the interface luminous description data and the device space configuration set, retaining the paths that meet the field of view and intensity conditions to form a display light path set; associating the interface luminous description data, the display light path set, the candidate observation action sequence set, and the evidence judgment rule set to determine the contaminated pixel area, the contamination overlap ratio, and the contamination intensity, establishing the display contamination propagation relationship, and forming the observation contamination evaluation result based on the spectral overlap degree, the contamination overlap ratio, and the normalized contamination intensity.

[0011] S4 specifically includes: based on the evidence judgment rules, the candidate observation action sequence set, and the observation pollution evaluation results, jointly evaluating the pollution degree, cumulative irradiation dose, execution time, and expected data quality to determine the target observation action sequence, interface luminescence coding, complementary luminescence coding, and synchronous acquisition timing; controlling each device to acquire display frames, dark frames, and complementary frames according to the synchronous acquisition timing, and removing invalid frames due to occlusion, saturation, changes in ambient light, or excessive device pose deviation to form a multi-frame cultural relic observation dataset; performing spatial registration on the multi-frame cultural relic observation dataset, inverting and stripping the display light reflection component based on the display light path set and display pollution propagation relationship, and forming purified observation data and purified data quality labels based on registration error, residual pollution ratio, signal-to-noise ratio, and effective pixel ratio.

[0012] S5 specifically includes: quality access of purification observation data based on purification data quality identifiers; extraction of physical characteristics of target and control areas according to evidence judgment rules; calculation and correction of evidence confidence; and formation of results indicating evidence satisfaction, insufficient evidence, evidence denial, or inability to verify. Based on the verification results, the recommendation propagation weights and verification status of knowledge relationship edges are updated, original data and verification records are retained, conflicting knowledge relationships are handled, and a credible recommendation subgraph, recommendation status summary, and new graph version are generated. Based on the credible recommendation subgraph, recommendation status summary, request topic, and display information granularity, display information nodes are sorted, target display information is selected, and the final recommendation interface content is generated. When no activation path exists, the status of insufficient evidence, inability to verify by the device, or lack of physically verifiable knowledge relationships is output.

[0013] The beneficial effects of this invention are as follows: This invention improves the accuracy and traceability of multi-source cultural relic data association by unifying the identity, data source, time information, and spatial coordinates of cultural relics, and constructing a knowledge graph that links cultural relic entities, knowledge conclusions, physical evidence, and display information.

[0014] This invention enables the linkage verification of knowledge conclusions and physical object detection processes by configuring physical evidence conditions for physically verifiable knowledge relationships and converting them into evidence judgment rules and candidate observation action sequences. This reduces the likelihood of unverified information being displayed and recommended. By combining the available equipment range, spatial collision avoidance conditions, and cultural relic irradiation safety boundaries to select observation parameters, it is possible to control irradiance, exposure time, and cumulative irradiation dose while meeting detection requirements, thereby improving the safety of the cultural relic observation process.

[0015] This invention establishes the display contamination propagation relationship between the display terminal, the reflective surface of the display case, the surface of the cultural relic, and the image acquisition device. It also utilizes display frames, dark frames, and complementary frames to strip away the display light reflection component, reducing the impact of ghosting, color shifts, and localized highlights on the judgment of physical evidence. By updating the recommendation propagation weights and verification states of knowledge relationship edges based on purification observation data, and outputting display information based on a credible recommendation subgraph, it can distinguish between cases of sufficient evidence, insufficient evidence, refuted evidence, and unverifiable evidence, thereby improving the authenticity, reliability, and interpretability of the recommendation results. Attached Figure Description

[0016] Figure 1 This is a flowchart of the intelligent recommendation method for cultural relic display information based on knowledge graphs according to the present invention. Figure 2 This is a collaborative block diagram of the multi-agent system and the detection hardware of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 and Figure 2 As shown, this embodiment provides an intelligent recommendation method for cultural relic display information based on knowledge graphs, including the following steps: S1. Obtain the entity data, knowledge relationship data, physical inspection data, digital image data, information to be displayed and equipment data of the target cultural relic, unify the identity and spatial coordinates of the cultural relic, construct a cultural relic knowledge graph containing cultural relic entity nodes, knowledge nodes, evidence nodes, display information nodes and knowledge relationship edges, and configure physical evidence conditions for physically verifiable knowledge relationship edges. S2. Receive user requests, determine the candidate knowledge relationship edge set and candidate display information set based on the request topic and the executability of physical evidence conditions, convert physical evidence conditions into evidence judgment rules, and generate a candidate observation action sequence set. S3. Generate the recommended interface content to be verified and the interface luminescence description data based on the candidate display information set. Combine the spatial relationship between the display terminal, the reflective surface of the display case, the surface of the cultural relic and the image acquisition device to establish the display light path set and obtain the display pollution propagation relationship and the observation pollution evaluation results. S4. Based on the evidence judgment rules, the candidate observation action sequence set and the observation pollution evaluation results, determine the target observation action sequence, interface luminescence coding and synchronous acquisition timing, acquire display frames, dark frames and complementary frames, strip the display light reflection component, and obtain purification observation data and quality label. S5. Based on the evidence judgment rules, extract physical features and judge evidence from the purified observation data, update the recommendation propagation weight and verification status of the knowledge relationship edges, generate a credible recommendation subgraph, and output the recommendation results of the target cultural relic accordingly.

[0019] S1 specifically includes the following sub-steps: S110. Obtain basic attribute data, historical research data, physical inspection data, digital image data, information to be displayed, and display inspection equipment data of the target cultural relic, and unify the cultural relic's identity, data source marking, time marking, and spatial coordinates.

[0020] Basic attribute data is retrieved from the museum's collection management system, cultural relic registration archives, and acquisition records, including at least the cultural relic number, name, age, material, size, location in the collection, and storage status. Historical research data is retrieved from audited archaeological excavation reports, restoration archives, research reports, expert annotation records, and publicly published materials, recording the document number, author, creation time, reviewer, and review status. Physical testing data is retrieved from spectrometers, polarization imaging devices, microscopic imaging devices, and 3D scanning devices, recording the testing equipment identification, acquisition time, acquisition area, working band, exposure parameters, and testing results.

[0021] Digital image data is read from the cultural relic digitization acquisition system and includes at least visible light images, multi-band images, three-dimensional models of cultural relics, image resolution, and camera calibration parameters; information to be displayed is read from the museum's digital exhibition system; data on display and testing equipment is read from equipment configuration files, equipment control interfaces, manufacturing archives, and on-site calibration records and includes at least the equipment identification, working range, installation location, and communication address of the display terminal, display case reflective surface, lighting device, polarization adjustment device, image acquisition device, and ambient light sensor.

[0022] The unique artifact identification number in the collection management system is used as the primary key for artifact identity. For data that does not contain a unique artifact identification number, an identity matching score is calculated based on the artifact's name, location in the collection, size, material, and digital image features.

[0023] in, This represents the identity matching score between the i-th data to be merged and the target cultural relic; This represents the matching value of the i-th data to be merged on the j-th identity matching feature. The five identity matching features are, in order, the name of the cultural relic, its location in the collection, its size, its material, and its digital image features. This represents the weight of the j-th identity matching feature, and the sum of all weights is 1.

[0024] Identity matching weights and thresholds are determined based on confirmed data of the same and different cultural relics, ensuring a false merging rate of no more than 1%. In a preferred embodiment, for bronze artifacts, the identity matching features and their weights are configured as follows: artifact name... =0.2, Location of Collections =0.2, Dimensions =0.15, Material =0.15 and digital image features =0.3, and the identity matching threshold is set to 0.85.

[0025] When the identity matching score is not lower than the identity matching threshold, the corresponding data will be associated with the target cultural relic; when it is lower than the identity matching threshold, the corresponding data will be written into the data area to be manually verified. After confirmation by the cultural relic management personnel, the unified identity of the cultural relic will be re-executed. Data that has not been confirmed shall not be written into the unified cultural relic dataset.

[0026] For example, if there are three blue and white porcelain vases with the same name in the collection system, but the spectral analysis file only records the name of the cultural relic, it is necessary to continue to compare the collection location, height, diameter and partial decorative images. It is prohibited to merge them based solely on the name of the cultural relic.

[0027] Set the coordinate system of the 3D model of the cultural relic as the reference coordinate system of the cultural relic, and use no less than 6 non-coplanar calibration points to obtain the rotation matrix and translation vector from each equipment coordinate system to the reference coordinate system of the cultural relic:

[0028] in, Represents the coordinates of a point in the device coordinate system; This indicates the coordinates of a spatial point in the reference coordinate system of the cultural relic; The rotation matrix representing the rotation from the equipment coordinate system to the artifact's reference coordinate system; This represents the translation vector from the equipment coordinate system to the cultural relic reference coordinate system.

[0029] If the image reprojection error is no higher than 1.5px, the device translation error is no higher than the preset translation error threshold, and the device rotation error is no higher than the preset rotation error threshold, the calibration result is retained; otherwise, recalibration is performed. The time stamp includes the data formation time, data acquisition time, and data import time. The historical period of the cultural relic is recorded separately as knowledge content. The processed data is used to form a unified cultural relic dataset and a device spatial configuration set, and version identifiers are set for both, serving as inputs for S120 and S130.

[0030] S120. Establish a multi-agent system for processing a unified cultural relic dataset and equipment spatial configuration set. The multi-agent system includes a knowledge graph agent, an evidence compilation agent, a display control agent, an observation control agent, a pollution purification agent, and a recommendation decision agent. Each agent is a software processing unit with fixed input data types, processing rules, output data types, and message interfaces.

[0031] The knowledge graph agent is used to construct, maintain, and update the cultural relic knowledge graph; the evidence compilation agent is used to convert physically verifiable knowledge relationship edges into physical evidence conditions and evidence judgment rules; the display control agent is used to generate the recommended interface content to be verified, interface luminescence description data, interface luminescence encoding, complementary luminescence encoding, and the final recommended interface content; the observation control agent is used to generate candidate observation action sequences and target observation action sequences; the pollution purification agent is used to establish the display light path set and display pollution propagation relationship, and generate purification observation data and purification data quality labels; the recommendation decision agent is used to generate the candidate knowledge relationship edge set, the candidate display information set, and the display information recommendation results.

[0032] Establish a shared task status table, recording at least the task identifier, artifact number, user request identifier, knowledge relationship edge identifier, current processing agent, current processing status, input data address, output data address, exception code, graph version identifier, device configuration version identifier, creation time, and completion time. The current processing status includes pending, processing, processing completed, processing failed, awaiting manual verification, topic pending confirmation, and currently unable to be verified.

[0033] The next agent reads the output only after the previous agent has updated the corresponding task status to indicate that the task is complete. When output parameters from different agents conflict, they are processed in the following order: cultural relic safety constraints, physical evidence acquisition constraints, and interface display constraints.

[0034] For example, the evidence compilation agent determines that the target detection band is 620nm to 700nm, while the display control agent intends to output a red highlighted box. The pollution purification agent writes the band conflict into the shared task status table, and the display control agent re-determines the interface color or arranges the red highlighted box to be output outside the exposure time slot of the image acquisition device. This forms a multi-agent collaborative processing environment that can be invoked by S130 to S530.

[0035] S130. The knowledge graph agent establishes cultural relic entity nodes, knowledge nodes, evidence nodes, and display information nodes based on the unified cultural relic dataset. Cultural relic entity nodes must at least record the cultural relic number, basic attributes, the address of the cultural relic's 3D model, data version, and the cultural relic's security boundary; knowledge nodes must at least record the knowledge content, knowledge type, data source, formation time, review status, and version identifier; evidence nodes must at least record the evidence type, target evidence area, detection data address, detection equipment identifier, acquisition parameters, acquisition time, evidence confidence level, and validity status; display information nodes must at least record the display text, display format, applicable display information granularity, interface resource address, and review status.

[0036] The safety boundaries for cultural relics include at least the permitted illumination bands, maximum irradiance per unit time, maximum duration of a single exposure, and upper limit of cumulative irradiance dose. The relevant data comes from cultural relic protection archives, material sensitivity level records, and the review results of protection personnel.

[0037] Knowledge relationship edges are established based on the correspondence between nodes. Each knowledge relationship edge records at least the starting node identifier, ending node identifier, relationship type, original edge weight, recommendation propagation weight, verification status, physical evidence conditions, and version identifier. The initial recommendation propagation weight is equal to the original edge weight.

[0038] Knowledge relationships are categorized into physically verifiable knowledge relationships, data-verifiable knowledge relationships, and interpretive knowledge relationships. Only physically verifiable knowledge relationships are included in the active observation process from S210 to S520. For knowledge nodes that cannot be simultaneously established, conflicting knowledge relationships are created, and conflict relationship identifiers, conflict node identifiers, and conflict resolution rules are recorded.

[0039] Physical evidence conditions refer to the executable detection constraints for determining whether a physically verifiable knowledge relationship has a physical basis. These constraints include at least the target evidence region, control region, physical feature to be tested, necessary feature, negation judgment feature, observation method, allowable range of equipment parameters, positive sample threshold range, negative sample threshold range, minimum evidence confidence, negation confidence, minimum spatial resolution, minimum signal-to-noise ratio, allowable saturation pixel ratio, interference exclusion requirements, and evidence invalidation conditions.

[0040] The physical evidence conditions are generated based on the reviewed test report, test data of standard samples of the same material, historical test data of the target cultural relic, equipment calibration data, and the target evidence area confirmed by experts. Each threshold is bound to the applicable equipment model, working band, exposure conditions, calibration date, and rule version.

[0041] For example, for the knowledge relationship edge that "the target area has been repainted later", the target evidence area is set as the designated surface area under the cultural relic reference coordinate system, the adjacent area that has been confirmed to belong to the original painting is set as the control area, and the difference in 365nm ultraviolet excitation fluorescence and the difference in cross-polarization reflection are set as necessary features; only when both of the necessary features meet the positive sample threshold is the evidence satisfied result allowed.

[0042] The device nodes and their spatial relationships are written into the cultural relic knowledge graph to form a cultural relic knowledge graph that includes physical evidence conditions, cultural relic security boundaries, conflict knowledge relationships, and device spatial relationships. The version identifier is written into the shared task status table as a unified data foundation for S210 to S530.

[0043] S2 specifically includes the following sub-steps: S210. The recommendation decision-making agent receives the user request and determines the set of candidate knowledge relationship edges. The user request comes from the museum's fixed interactive screen, mobile guide terminal, augmented reality display terminal, voice explanation terminal, or in-museum digital exhibition system, and includes at least the request identifier, target cultural relic identifier, original request content, request input method, display terminal identifier, display information granularity, and request time.

[0044] The granularity of the displayed information is a level parameter that limits the level of detail of the recommended content, including overview level, explanatory level, and evidence level. When the user does not select, it is determined according to the single display capacity of the display terminal and the navigation mode. Specifically, when the single display content is less than 100 characters, it is determined as overview level; when the display is 100 to 300 characters, it is determined as explanatory level; and when the display of detection data, evidence images, and evidence sources is allowed, it is determined as evidence level.

[0045] The recommendation decision-making agent invokes a request topic dictionary reviewed by cultural relics professionals to perform word segmentation, synonym merging, and topic mapping on the original request content. For example, "repainting," "post-painting coloring," and "repair pigments" are uniformly mapped to the topic "post-painting repainting." If topic mapping cannot be completed, the task status is set to "topic pending confirmation," and candidate topics are provided through the display terminal. If the user confirms, S210 is re-executed; otherwise, the active observation task is terminated.

[0046] The semantic matching value between the requested topic and the e-th knowledge relation edge is calculated using the following formula:

[0047] in, Represents the semantic matching value of the e-th knowledge relation edge; This represents the set of requested keywords after the synonym merging is complete. This represents the set of words consisting of the knowledge node name, relationship type, and displayed information keywords corresponding to the e-th knowledge relationship edge; the vertical line indicates the number of words in the set.

[0048] Starting from the cultural relic entity node corresponding to the target cultural relic, retrieve physically verifiable knowledge relationship edges that are not blocked and have physical evidence conditions, and calculate candidate scores:

[0049] in, This represents the candidate score for the e-th knowledge relation edge; Indicates semantic matching value; This represents the ratio of the number of fields already filled in the corresponding displayed information node to the total number of fields required for the current displayed information granularity; This represents the executable value for the physical evidence conditions. It is set to 1 if the current device covers the required operating band, polarization state, illumination incident angle, and observation pose, and otherwise to 0. , and This represents the corresponding weight, and the sum of the three is 1.

[0050] The weights and filtering thresholds are calibrated based on no fewer than 100 historical requests verified by professionals. In a preferred embodiment, semantic matching weights are used. =0.4, Weight of completeness of displayed information =0.3, physical evidence condition enforceability weight =0.3, and the candidate score screening threshold is set to 0.75.

[0051] Knowledge relationship edges whose candidate scores reach the screening threshold and whose executable value is 1 are written into the candidate knowledge relationship edge set, and the corresponding candidate display information is extracted to form a candidate display information set. If there are no knowledge relationship edges that meet the conditions, "The current request does not have a physically verifiable knowledge relationship" is recorded in the shared task status table. The candidate knowledge relationship edge set and the candidate display information set are written into the shared task status table as inputs to S220, S310, S510, and S530.

[0052] S220. The evidence compilation agent converts the physical evidence conditions corresponding to the candidate knowledge relation edges into a set of evidence judgment rules. Evidence judgment rules are structured rules that directly guide the generation of observation parameters and the determination of verification results. Each evidence judgment rule is bound to a unique candidate knowledge relation edge identifier and includes at least the target evidence region, control region, physical feature to be tested, necessary features, negation judgment features, observation method, feature calculation method, positive sample threshold range, negative sample threshold range, minimum evidence confidence, negation confidence, minimum spatial resolution, minimum signal-to-noise ratio, allowed saturated pixel ratio, interference exclusion requirements, and rule version identifier.

[0053] The target evidence area is read from the physical evidence conditions in S130; the control area is selected first from adjacent areas with the same material and storage environment as the target evidence area and no features to be verified are found. If it cannot be obtained, a standard sample area of ​​the same material is used.

[0054] The physical characteristics of the target region and the control region are calculated using the following formula:

[0055] in, Represents the normalized difference value of the physical characteristic to be measured; The average measurement value representing the target evidence area; This represents the average measurement value of the control area; This indicates a fixed positive number to prevent the denominator from being 0, and takes the minimum quantization value of the corresponding detection device.

[0056] Display light interference, ambient light variations, and display case reflections are converted into identifiable interference elimination requirements. The rate of change in ambient light intensity is calculated using the following formula:

[0057] in, Indicates the rate of change of ambient light intensity; and These represent the ambient light sensor measurements at the beginning and end of the observation period, respectively. This indicates a fixed positive number to prevent the denominator from being zero.

[0058] When the target detection band overlaps with the main emission band of the display terminal, time coding or complementary frame acquisition requirements are set; when the ambient light intensity change rate exceeds the ambient light change threshold, evidence judgment is not performed; when the overlap ratio between the target evidence area and the predicted specular reflection area exceeds a preset threshold, the observation pose or illumination incident angle is required to be adjusted. The set of evidence judgment rules is written into the shared task state table as input to S230, S410, and S510.

[0059] S230: The observation and control agent generates a set of candidate observation action sequences based on the evidence judgment rule set, the equipment spatial configuration set, and the cultural relic safety boundary. Equipment parameters are derived from the equipment manufacturer's configuration file, the currently available parameters returned by the equipment control interface, the on-site calibration records generated in S110, and the equipment self-test results. Illumination device parameters include at least the output band, irradiance range, incident angle adjustment range, and switching time; polarization adjustment device parameters include at least the polarization angle range and minimum adjustment step size; image acquisition device parameters include at least the focal length, exposure time, gain, achievable observation pose, spatial resolution, and effective range.

[0060] The target detection band, polarization state, minimum spatial resolution, minimum signal-to-noise ratio, target evidence area, and control area are determined according to the evidence judgment rules. Then, parameter combinations are generated based on the minimum adjustment step size of the equipment. The observation pose is generated based on the center point of the target evidence area and the surface normal vector, so that the optical axis of the image acquisition device points to the center point of the target evidence area, and the target evidence area is completely within the acquisition field of view, the effective number of pixels meets the feature calculation requirements, and the equipment movement path does not intersect with the cultural relics, supports, or display cases.

[0061] Exposure time and gain are determined by preview frames to ensure that the average gray level of the target evidence area is between 20% and 80% of the effective range of the image acquisition device, and the saturated pixel ratio is not higher than 5%.

[0062] Calculate the cumulative radiation dose for each parameter combination:

[0063] in, This represents the cumulative radiation dose generated by the a-th candidate observation action sequence, in J / m³. 2 ; This represents the irradiance of the target artifact surface during the r-th observation, expressed in W / m². 2 ; The duration of the r-th observation is expressed in seconds; n represents the number of observations contained in the candidate observation action sequence.

[0064] Parameter combinations that exceed the cultural relic safety boundary, exceed the current usable range of the equipment, involve spatial collisions or occlusions, or fail to meet the minimum spatial resolution or minimum signal-to-noise ratio are sequentially eliminated. The remaining parameter combinations are then arranged in the order of equipment positioning, lighting settings, polarization settings, exposure settings, and image acquisition to form a candidate observation action sequence. When all parameter combinations are eliminated, the corresponding candidate knowledge relationship edges are marked as currently unobservable, and the reason for the inability to execute is recorded in the recommended status summary.

[0065] For example, if physical evidence requires 365nm ultraviolet excitation, but the field lighting only supports 450nm to 700nm, blue light cannot be substituted for ultraviolet light. The set of candidate observation action sequences is written into the shared task state table as input to S330 and S410.

[0066] S3 specifically includes the following sub-steps: S310. The display control agent generates the recommended interface content to be verified based on the candidate display information set, and forms the interface illumination description data. The recommended interface content to be verified refers to the interface content used to display pollution predictions before the physical verification of candidate knowledge relationships is completed, and is not output as the final recommendation result. The display text, associated evidence area, and display information granularity are read from the candidate display information set; the interface template is read from the digital exhibition system in the museum; the display terminal resolution, pixel size, maximum brightness, refresh rate, and display spectrum calibration table are read from the device space configuration set.

[0067] The smallest consecutive pixel area on the display terminal that has the same display color, light intensity, refresh timing, and display status is defined as the light-emitting unit. Initial segmentation is performed according to text strokes, indicator arrows, local highlighted areas, background areas, and digital restoration layers; adjacent pixels whose color component difference, brightness difference, and refresh phase difference do not exceed the corresponding segmentation threshold are merged into the same light-emitting unit.

[0068] The color component difference threshold, brightness difference threshold, and refresh phase difference threshold are determined based on the repeated output error of the display terminal under the same display command. Each light-emitting unit records at least the light-emitting unit identifier, pixel coordinate range, display color, display brightness, light emission spectrum distribution, refresh sequence, candidate knowledge relationship edge identifier, and target evidence area identifier.

[0069] The display brightness is calculated based on the digital drive value and brightness calibration curve, while the luminous spectrum distribution is determined by the display spectrum calibration table collected by a spectrometer after the display terminal is installed. The display spectrum calibration table records the spectral power distribution of the red, green, and blue display channels under multiple digital drive values. Brightness and spectrum calibration must be performed again when the display terminal location changes, the display panel is replaced, or the calibration validity period expires.

[0070] For example, the main pixels of the red highlighted border, the anti-aliased edges, and the semi-transparent background have different actual luminous intensities and should be divided into separate luminous units; the entire border should not be treated as a single luminous unit. The recommended interface content to be verified and the interface luminous description data are written into the shared task state table as input to S320, S330, and S410.

[0071] S320: The pollution purification intelligent agent establishes a set of display light paths based on the interface luminescence description data and the equipment spatial configuration set. The equipment spatial configuration set provides the spatial position and normal vector of the luminescent unit in the cultural relic's reference coordinate system, the three-dimensional geometric model, thickness, refractive index, and spectral reflectivity of the display case's reflective surface, the three-dimensional model and surface normal vector of the target cultural relic, and the optical center position, optical axis direction, focal length, field of view, and pixel mapping relationship of the image acquisition device. The display case material parameters are derived from the display case manufacturing archives and on-site reflectivity calibration, and the surface reflectivity of the target cultural relic is derived from historical physical object testing data; when corresponding data is lacking, the measured reflectivity of a standard sample of the same material is used.

[0072] For each light-emitting unit, sampling points are selected at fixed pixel intervals. Using these sampling points as the starting point of the displayed light, the propagation paths are sequentially traced: the light directly enters the image acquisition device from the display case's reflective surface; it enters the image acquisition device after reflection from the target artifact's surface; and it enters the image acquisition device after continuous reflection from both the display case's reflective surface and the target artifact's surface. Double-layered or laminated glass is used to establish the front and rear surfaces, respectively.

[0073] The spectral intensity of the u-th emitting unit reaching the image acquisition device via the h-th propagation path is calculated using the following formula:

[0074] in, Indicates wavelength as The spectral intensity reaching the image acquisition device via the h-th propagation path; This represents the spectral intensity of the u-th luminescent unit; This represents the equivalent spectral reflectance coefficient of the reflecting surface and the target artifact surface traversed by the h-th propagation path; The geometric propagation coefficient is determined by the propagation distance, incident angle, reflection angle, transmission loss, and the field-of-view response of the image acquisition device. This indicates the wavelength, measured in nm.

[0075] For each propagation path, check whether it is obstructed by the display case frame, artifact support, or other objects; whether the reflection direction enters the effective field of view of the image acquisition device; whether the reflection point is located on the actual surface; and whether the spectral intensity after propagation is higher than the minimum detectable intensity of the image acquisition device. Only propagation paths that meet the conditions are written into the display optical path set. The display optical path set is written into the shared task status table as input to S330 and S430.

[0076] S330: The pollution purification intelligent agent associates the interface luminescence description data, the display light path set, the candidate observation action sequence set, and the evidence judgment rule set to establish the display pollution propagation relationship and form the observation pollution evaluation result.

[0077] The pollution propagation relationship is a directed correspondence between luminescent units, propagation paths, target evidence regions, image acquisition pixel regions, and candidate observation action sequences. Each relationship records at least the luminescent unit identifier, candidate observation action sequence identifier, propagation path identifier, pollution band range, predicted pollution intensity, target evidence region identifier, pollution pixel region, pollution overlap ratio, and relationship validity status.

[0078] For each candidate observation action sequence, the effective propagation path is projected onto the imaging plane of the image acquisition device according to its observation pose and image acquisition parameters to obtain the predicted contaminated pixel region and calculate the contamination overlap ratio:

[0079] in, This represents the contamination overlap ratio of the u-th luminescent unit under the a-th candidate observation action sequence; Indicates the predicted contaminated pixel region; The vertical line represents the image pixel region corresponding to the target evidence region; the vertical line represents the number of pixels contained in the region.

[0080] Based on the degree of spectral overlap between the emission spectral distribution and the target detection band, the contamination overlap ratio, and the normalized contamination intensity, the contamination evaluation value of the candidate observation action sequence is calculated:

[0081] in, This represents the pollution assessment value of the a-th candidate observation action sequence; Indicates the number of light-emitting units; This indicates the degree of spectral overlap between the u-th emitting unit and the target detection band of the a-th candidate observation action sequence; Indicates the percentage of pollution overlap; This indicates the normalized contamination intensity of the light reaching the image acquisition device via the effective propagation path.

[0082] Based on calibration data from the no-display, normal-display, and highlighted-display states, the observed contamination assessment results are categorized into no contamination, coded and separable contamination, and unacceptable contamination. Candidate observation action sequences for unacceptable contamination are marked as requiring adjustments to the observation pose, illumination incident angle, or interface color. The displayed contamination propagation relationships and observed contamination assessment results are written into a shared task status table as input to S410 and S430.

[0083] S4 specifically includes the following sub-steps: S410. The observation control agent and the display control agent jointly determine the target observation action sequence, interface illumination code, and synchronous acquisition timing. The target observation action sequence refers to the actual execution sequence selected from the candidate observation action sequence set under the premise of satisfying the evidence judgment rules, cultural relic safety boundaries, the current available range of the equipment, and spatial collision avoidance constraints; the interface illumination code refers to the combination of display color, digital drive value, illumination time slot, and refresh phase configured for each illumination unit.

[0084] Calculate the joint evaluation value for each candidate observation action sequence:

[0085] in, This represents the joint evaluation value of the a-th candidate observation action sequence; This represents the normalized pollution assessment value; This represents the normalized cumulative radiation dose; Indicates the normalized estimated execution time; Indicates the expected quality of normalized data; to This represents the corresponding weight, and the sum of all weights is 1.

[0086] Each parameter is linearly normalized based on its minimum and maximum values ​​within the same candidate observation task; when the maximum and minimum values ​​are equal, the corresponding normalized value is set to 0. The candidate observation action sequence with the smallest joint evaluation value that satisfies the cultural relic safety boundary is determined as the target observation action sequence.

[0087] In a preferred embodiment that balances the safety of cultural relics with the quality of observation, the aforementioned weights are specifically configured as follows: weight of normalized pollution assessment value. =0.3, normalized cumulative irradiation dose weight =0.4, Normalized Expected Execution Time Weight =0.1, Normalized expected data quality weight =0.2.

[0088] When the overlap of the contamination spectra of two light-emitting units is not lower than the spectral merging threshold, the cross-union ratio of the contamination pixel regions is not lower than the region merging threshold, and the relative difference in geometric propagation coefficients is not higher than the propagation difference threshold, they are assigned to the same contamination coding group. When there are several independent contamination coding groups, generate no less than The interface emits light in a linearly independent manner. The dark frame digital drive value is set to 0, and the complementary emittance code is selected from the display spectrum calibration table and limited to the linear response range of the display terminal.

[0089] Based on the joint calibration results of the display terminal refresh delay, lighting device response time, and image acquisition device trigger delay, a synchronous acquisition sequence is generated, including device parameter switching, stabilization waiting, exposure start, and exposure end times. The target observation action sequence, interface emission code, complementary emission code, and synchronous acquisition sequence are written into the shared task status table as input to S420.

[0090] S420: The display control agent and the observation control agent perform multi-frame synchronous observation according to the synchronous acquisition sequence. All devices are connected to the same hardware trigger controller; when hardware triggering cannot be used, a unified clock source and timestamp correction are used, and the maximum synchronization error does not exceed 10% of the shortest exposure time.

[0091] Specifically, using the rising edge signal output by the hardware trigger controller as the time base, and based on the pre-determined inherent response delay of each device, before the reference time... Send display drive commands to the display terminal, send trigger signals to the lighting device and polarization adjustment device at the reference time, and after the reference time... Send an exposure trigger signal to the image acquisition device; wherein, This indicates that the display terminal drives the time difference in advance. This indicates the time difference between the image acquisition delay and the trigger time. Greater than or equal to the shortest waiting time required for the lighting and polarization devices to reach a stable state.

[0092] During the display frame, the display terminal outputs the recommended interface content to be verified according to the interface light emission code; during the dark frame, only the digital drive value of the light emission unit corresponding to the recommended interface content to be verified is set to 0, and the parameters of the lighting device, polarization adjustment device and image acquisition device remain unchanged; during the complementary frame, the display terminal outputs according to the complementary light emission code, and the parameters of other devices remain consistent with the display frame.

[0093] After each switch of display drive value, illumination band, illumination incident angle, or polarization angle, wait for the device output to enter the calibrated allowable error range before starting exposure. For each frame type, continuously acquire at least 3 frames, and for valid repeating frames, take the pixel median value to form the corresponding observation frame. During acquisition, simultaneously record the candidate knowledge relationship edge identifier, target observation action sequence identifier, luminescence code identifier, acquisition cycle number, image acquisition device identifier, actual exposure time, actual gain, device trigger timestamp, ambient light sensor measurement value, actual device pose, and data version identifier.

[0094] If the saturated pixel ratio of the target evidence area exceeds 5%, is obscured by visitors or other objects, the rate of change in ambient light intensity exceeds the threshold in the evidence judgment rules, the actual exposure time deviation exceeds 5%, or the equipment translation error exceeds the translation error threshold or the equipment rotation error exceeds the rotation error threshold, the corresponding frame will be marked as invalid. If there are fewer than 2 valid frames of any frame type, the corresponding acquisition cycle will be re-acquired. The display frames, dark frames, and complementary frames that pass the integrity check, along with their acquisition parameters, will form a multi-frame cultural relic observation dataset, which will be written into the shared task status table as input to S430.

[0095] S430: The pollution purification agent reads multiple frames of cultural relic observation datasets, displays the optical path set and the pollution propagation relationship, performs spatial registration of the multiple frames of cultural relic observation data, displays the light reflection component inversion and data quality evaluation.

[0096] First, based on the 3D model of the cultural relic, the internal and external parameters of the image acquisition device, and the actual pose of the device in each frame, the target evidence area is projected onto each observation frame. Local 2D geometric transformation is applied to planar areas, and curved areas are reprojected based on the 3D model of the cultural relic. Subpixel offset is corrected using stable textures around the target evidence area that are unaffected by display light. If the registration error exceeds 1px, the data is marked as requiring re-observation.

[0097] Establish an observation model for the observations of the p-th image pixel in the f-th coded frame:

[0098] in, Indicates the actual observed value; This represents the inherent response of the target artifact under current lighting and polarization conditions; Indicates the ambient light baseline; Indicates the number of contamination coding groups; This represents the actual propagation coefficient from the c-th contamination coding group to the p-th image pixel; This represents the known digital drive value of the c-th contaminated coding group in the f-th coded frame; This indicates the noise level during data acquisition.

[0099] The equivalent spectral reflectance and geometric propagation coefficient in the display optical path set are used as the initial values ​​for the actual propagation coefficient. The pollution propagation relationship is displayed to provide the range of polluted pixels. Multiple sets of linearly independent coded frames provide the solution equations. The actual propagation coefficient and intrinsic response are calculated using the least squares method. The ambient light sensor measurements are converted into the ambient light baseline change of each coded frame according to the pre-calibrated mapping relationship between ambient light intensity and image pixel response.

[0100] The light reflection component is stripped from the display frame observation data. Pixels below the black level of the image acquisition device are truncated to black level. Saturated pixels and large invalid areas are not included in subsequent physical feature calculations. Spatial registration error, residual contamination ratio, signal-to-noise ratio, effective pixel ratio, and ambient light stability are calculated to form a clean data quality label.

[0101] When the registration error does not exceed 1px, the residual contamination ratio does not exceed 5%, the signal-to-noise ratio is not lower than the equipment calibration lower limit, and the effective pixel ratio is not lower than 90%, the quality level is qualified. If any indicator exceeds the limit but can be corrected by adjusting the interface luminescence code or the target observation action sequence, the quality level requires re-observation. If the quality level is still unqualified after 3 consecutive re-observations, the quality level is invalid. The purification observation data and purification data quality identifier are written into the shared task status table as the observation data input for S510.

[0102] S5 specifically includes the following sub-steps: S510: The evidence compilation agent performs quality access, physical feature extraction, and knowledge relationship edge verification on the cleaned observation data. When the cleaned data quality is marked as qualified, the cleaned observation data enters the physical feature extraction stage; when the quality level requires re-observation, it returns to S410 to redetermine the target observation action sequence or interface luminous encoding; when the quality level is invalid, an unverifiable result is generated, and no evidence denial result should be generated.

[0103] For each candidate knowledge relationship edge, according to the target evidence area, control area, observation method and feature calculation method in the evidence judgment rules, the target band reflection difference, polarization response difference, texture gradient change, color component change or local geometric change are extracted from the purified observation data, and the original measurement value, normalized value, unit of measurement, equipment identification and acquisition cycle number are recorded.

[0104] The confidence level of the modified evidence for the candidate knowledge relation edge e is calculated using the following formula:

[0105] in, This represents the revised evidence confidence level of the e-th candidate knowledge relation edge; This represents the data quality coefficient obtained by normalizing spatial registration error, residual contamination ratio, signal-to-noise ratio, and effective pixel ratio. Indicates the number of physical features to be measured; This represents the weight of the z-th physical feature, and the sum of all weights is 1; This represents the degree of matching between the z-th physical feature of the e-th candidate knowledge relation edge and the corresponding threshold range.

[0106] When the cleaned data quality is qualified, all necessary features fall within the positive sample threshold range, and the revised evidence confidence level is not lower than the minimum evidence confidence level, an evidence satisfaction result is formed; when all negative judgment features fall within the negative sample threshold range and the revised evidence confidence level reaches the negative confidence level, an evidence negation result is formed; when only some necessary features are satisfied, the conclusions of different physical features conflict, or the revised evidence confidence level is insufficient, an evidence insufficiency result is formed; when the cleaned data quality is invalid, an unverifiable result is formed.

[0107] For example, if the difference in ultraviolet fluorescence reaches the threshold for the supplementary sample, and the difference in cross-polarization reflectance is within the overlap range between the original and supplementary colored samples, an insufficient evidence result should be formed. The physical characteristic values, the revised confidence level of evidence, and the verification results are combined to form a set of verification results for the knowledge relationship edge, which serves as the input to S520.

[0108] S520. The knowledge graph agent updates the cultural relic knowledge graph based on the set of knowledge relation edge verification results, and generates a reliable recommendation subgraph and a recommendation status summary. Each knowledge relation edge records the original edge weight, recommendation propagation weight, verification status, and verification record. The verification record includes at least the observation task identifier, cleaned observation data identifier, physical feature value, corrected evidence confidence level, verification time, device version, and rule version.

[0109] The recommendation propagation weight of the e-th knowledge relation edge is updated using the following formula:

[0110] in, This indicates the updated recommendation propagation weight; This indicates the recommendation propagation weight before the update; Indicates the revised confidence level of the evidence; Indicates the fusion coefficient of evidence; This represents the weighting factor when there is insufficient evidence, and 0 is less than 0. Less than 1. In a preferred embodiment, to prevent accidental abrupt changes caused by a single observation, the evidence fusion coefficient is... The weighting factor is set to 0.3; this is to accelerate the elimination of knowledge without supporting evidence. Set to 0.5.

[0111] When the evidence is sufficient, the verification status is set to active; when the evidence is insufficient, it is set to await further observation; when the evidence is negative, it is set to blocked; when verification is impossible, the original verification status is maintained, but the corresponding knowledge relationship edge is not included in the current trusted recommendation subgraph. Blocking means setting the recommendation propagation weight to 0 in the current graph version and the current recommendation task, stopping the recommendation propagation of the knowledge relationship edge to the display information node, but retaining the knowledge relationship edge, original data, and verification records.

[0112] Before updating the graph, a snapshot of the current graph version is created, and the recommendation propagation weight, verification status, evidence nodes, and verification records are written as a single transaction; if any field fails to be written, the update is revoked. Using the cultural relic entity node corresponding to the target cultural relic as the root node, only knowledge relationship edges with an active verification status, a recommendation propagation weight greater than the propagation threshold, and the ability to reach candidate display information nodes, along with their associated nodes, are retained to form a trusted recommendation subgraph.

[0113] For conflicting knowledge conclusions, if one knowledge relation edge satisfies the evidence while another refutes it, only the path where the evidence satisfies is retained; if both are insufficient, only the "current evidence insufficient" explanation node is retained. A recommendation state summary is generated synchronously, recording the corresponding states: no candidate knowledge relation edges, device currently unexecutable, invalid cleanup observation data, insufficient evidence, refuted evidence, or presence of an active path. The credible recommendation subgraph, recommendation state summary, and new graph version identifier are written into the shared task state table as input to S530.

[0114] S530. The recommendation decision-making agent generates the recommendation results for the target cultural relic's display information based on the credible recommendation subgraph, recommendation status summary, request topic, and display information granularity. For the display information nodes in the credible recommendation subgraph, they are sorted in the following order: the credibility of the recommendation path evidence, the semantic matching value with the request topic, the completeness of the display information fields, the degree of conformity with the display information granularity, and the length of the recommendation path.

[0115] The credibility of the recommended path evidence is determined by the lowest recommendation propagation weight within that path, to prevent low-credibility knowledge relationships from being obscured by other high-weight knowledge relationships. When the same information node is supported by multiple independent activation paths, the path with the highest credibility of the recommended path evidence is selected as the primary recommended path, and the remaining paths are used as supplementary evidence paths.

[0116] The target information is selected for display based on the sorting results. The number of selections is determined by the single display capacity of the display terminal and the granularity of the displayed information. The overview level includes at least the name of the cultural relic, its age, and verified conclusions; the description level further includes the material, craftsmanship, target evidence area, and relationship description; the evidence level further includes the observation method, physical characteristic results, revised evidence confidence level, and evidence source.

[0117] The display control agent regenerates the final recommended interface content based on the target display information, the display terminal configuration file, and the interface template of the museum's digital exhibition system, instead of directly using the unverified recommended interface content generated by S310. The final recommended interface content is bound to at least the user request identifier, the target cultural relic identifier, the trusted recommended sub-map version identifier, the display information node identifier, the main recommendation path, the confidence level of the correction evidence, and the generation time.

[0118] When there is no activation path to the displayed information node in the trusted recommendation subgraph, the recommendation decision agent outputs the corresponding status in the recommendation status summary: "Insufficient physical evidence," "Verification cannot be completed under current device conditions," "Invalid purification observation data," or "No physically verifiable knowledge relationship exists in the current request." It is not allowed to forcibly select a deterministic conclusion from the original candidate display information set.

[0119] For example, if the "post-drawing" knowledge relationship edge only forms insufficient evidence, the output should be "This observation did not form sufficient physical evidence to confirm the post-drawing," and the observation method and the status of waiting for re-observation should be displayed. A positive conclusion should not be output. After the display terminal completes the output, the final recommended interface content identifier, actual display time, output status, and map version identifier should be written into the shared task status table to form the display information recommendation result for the target cultural relic.

[0120] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0121] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. 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 implementation should not be considered beyond the scope of this application.

[0122] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0123] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent recommendation of cultural relic display information based on knowledge graphs, characterized in that, Includes the following steps: S1. Obtain the entity data, knowledge relationship data, physical inspection data, digital image data, information to be displayed and equipment data of the target cultural relic, unify the identity and spatial coordinates of the cultural relic, construct a cultural relic knowledge graph containing cultural relic entity nodes, knowledge nodes, evidence nodes, display information nodes and knowledge relationship edges, and configure physical evidence conditions for physically verifiable knowledge relationship edges. S2. Receive user requests, determine the candidate knowledge relationship edge set and candidate display information set based on the request topic and the executability of physical evidence conditions, convert physical evidence conditions into evidence judgment rules, and generate a candidate observation action sequence set. S3. Generate the recommended interface content to be verified and the interface luminescence description data based on the candidate display information set. Combine the spatial relationship between the display terminal, the reflective surface of the display case, the surface of the cultural relic and the image acquisition device to establish the display light path set and obtain the display pollution propagation relationship and the observation pollution evaluation results. S4. Based on the evidence judgment rules, the candidate observation action sequence set, and the observation pollution evaluation results, determine the target observation action sequence, interface luminescence code, and synchronous acquisition timing. Acquire display frames, dark frames, and complementary frames, strip the display light reflection component, and obtain purification observation data and quality labels.

2. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 1, characterized in that, Also includes: S5. Based on the evidence judgment rules, extract physical features and judge evidence from the purified observation data, update the recommendation propagation weight and verification status of the knowledge relationship edges, generate a credible recommendation subgraph, and output the recommendation results of the target cultural relic accordingly.

3. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 1, characterized in that, S1 specifically includes: Acquire the basic attributes, historical research, physical inspection, digital images, data on the objects to be displayed and equipment of the target cultural relics, unify their identity with a unique cultural relic number, complete the marking of origin, time and spatial coordinates, and form a unified cultural relic dataset and equipment spatial configuration set; Establish a multi-agent system consisting of knowledge graph, evidence compilation, display control, observation control, pollution purification, and recommendation decision-making agents, and transmit input, output, and processing status through a shared task status table.

4. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 3, characterized in that, It also includes: constructing a cultural relic knowledge graph based on a unified cultural relic dataset, which includes cultural relic entity nodes, knowledge nodes, evidence nodes, display information nodes and knowledge relationship edges; configuring physical evidence conditions and cultural relic security boundaries for physically verifiable knowledge relationship edges; and writing conflicting knowledge relationships and device spatial relationships.

5. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 1, characterized in that, S2 specifically includes: Receive user requests and perform topic mapping. Starting from the cultural relic entity node corresponding to the target cultural relic, candidate knowledge relationship edges are selected by combining semantic matching, completeness of displayed information and executability of physical evidence conditions to form a candidate display information set. Read the physical evidence conditions of candidate knowledge relationship edges, determine the target evidence area, control area, physical features to be tested, necessary features, threshold range, minimum evidence confidence and interference exclusion requirements, and form a set of evidence judgment rules; Based on the evidence judgment rule set, equipment spatial configuration set, and cultural relic safety boundary parameter combination, parameter combinations that exceed the irradiation dose, exceed the equipment range, have collision obstruction, or do not meet the spatial resolution and signal-to-noise ratio requirements are eliminated, and the retained combinations are arranged into a candidate observation action sequence set.

6. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 1, characterized in that, S3 specifically includes: Based on the candidate display information set, generate the recommended interface content to be verified, and divide the pixel areas with the same display color, light intensity, refresh time and display status into light units to form interface light description data; Based on the interface luminescence description data and the device space configuration set, the propagation path of the display terminal through the display case reflective surface, the surface of the cultural relic and to the image acquisition device is tracked, and the path that meets the field of view and intensity conditions is retained to form a display light path set.

7. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 6, characterized in that, Also includes: By associating interface luminescence description data, display optical path set, candidate observation action sequence set, and evidence judgment rule set, the polluted pixel area, pollution overlap ratio, and pollution intensity are determined, the display pollution propagation relationship is established, and the observation pollution evaluation result is formed based on the degree of spectral overlap, pollution overlap ratio, and normalized pollution intensity.

8. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 1, characterized in that, S4 specifically includes: Based on the evidence judgment rules, the candidate observation action sequence set and the observation pollution assessment results, the pollution degree, cumulative irradiation dose, execution time and expected data quality are jointly evaluated to determine the target observation action sequence, interface luminescence coding, complementary luminescence coding and synchronous acquisition timing. According to the synchronous acquisition timing control, each device acquires and displays frames, dark frames and complementary frames, and removes invalid frames that are blocked, saturated, have changes in ambient light or exceed the limits of device pose deviation, to form a multi-frame cultural relic observation dataset. Spatial registration is performed on multi-frame cultural relic observation datasets. Based on the display light path set and the display contamination propagation relationship, the display light reflection component is inverted and stripped. Purified observation data and purified data quality labels are formed according to the registration error, residual contamination ratio, signal-to-noise ratio and effective pixel ratio.

9. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 2, characterized in that, S5 specifically includes: The purification observation data is quality-controlled according to the purification data quality label. The physical characteristics of the target and control areas are extracted according to the evidence judgment rules. The confidence of the corrected evidence is calculated and the results of evidence satisfaction, insufficient evidence, evidence denial or unverifiable are formed. Based on the verification results, update the recommendation propagation weights and verification status of knowledge relationship edges, retain the original data and verification records, handle conflicting knowledge relationships, and generate a reliable recommendation subgraph, a recommendation status summary, and a new graph version.

10. The intelligent recommendation method for cultural relic display information based on knowledge graphs according to claim 9, characterized in that, Also includes: Based on the trusted recommendation subgraph, recommendation status summary, request topic and display information granularity, the display information nodes are sorted, the target display information is selected and the final recommendation interface content is generated. If no activation path exists, output "insufficient evidence", "device cannot be verified", or "no physically verifiable knowledge relationship status".

Images