Digital cultural intelligent resource library construction method and system

Abstract

The application discloses a kind of digital culture intelligent resource library construction method and system, belong to digital cultural resource management technical field.The method includes: multi-source heterogeneous layered collection (six categories of resources are synchronously collected by multimodal terminal, high-precision detail capture), adaptive heterogeneous resource standardization processing (personalized algorithm, independent DCIR format), double-branch attention intelligent classification association (accurate classification, mining internal correlation), distributed encryption storage and intelligent indexing (safe and traceable, fast retrieval), dynamic self-optimization and intelligent updating (model incremental updating, resource real-time replenishment).The system, device and medium correspond to realize the above method.The present application completely avoids prior art, realizes the high-precision, intelligent, safe and dynamic construction of digital cultural resources, and improves the efficiency of resource protection and utilization.

Description

Technical Field

[0001] This invention relates to the field of digital cultural resource management and intelligent construction technology, and in particular to a method and system for constructing a digital cultural intelligent resource database. Background Technology

[0002] Existing technologies for constructing digital cultural resource databases suffer from the following core defects, all of which are avoided in this invention, and there is no technological overlap: 1. The data collection methods are limited, often using a single device and a single dimension, which makes it impossible to collect digital cultural resources of multiple types and scenarios simultaneously. The collection accuracy is low, and details are seriously lost. In particular, the collection effect is poor for intangible cultural heritage skills, damaged ancient books, and complex cultural relics. Furthermore, there is no spatiotemporal synchronization calibration and redundant backup mechanism, making the collected data easy to lose and unreliable. 2. The standardized processing algorithm is fixed and has not been adapted to the characteristics of different types of digital cultural resources. The resource formats are not uniform, the interoperability is poor, and there is a lack of efficient damage repair and fragment integration technology, which makes it impossible to restore the original appearance of the resources and the standardization qualification rate is low. 3. The classification and association modeling are simple, mostly using traditional machine learning algorithms for single-dimensional classification, failing to explore the inherent relationships between different types of resources, resulting in low classification accuracy, weak association recognition ability, and inability to support custom classification systems, leading to poor adaptability. 4. The storage architecture is simple, mostly adopting centralized storage, with limited storage capacity, unable to achieve dynamic expansion, low resource encryption level, lack of full-process traceability mechanism, inability to guarantee resource security, low indexing efficiency, and slow retrieval response. 5. Without a dynamic self-optimization and intelligent update mechanism, the model parameters are fixed and cannot adapt to new resources in the cultural field and changes in user needs. The resource library is updated late, and low-quality and invalid resources cannot be cleaned up in a timely manner, affecting the integrity and accuracy of the resource library. 6. Poor interactive experience, lack of 3D visualization and immersive display functions, unable to meet the needs of different users, and lack of convenient operation methods such as dialect interaction and voice search, resulting in low accessibility and ease of use of resources; 7. Existing media programs have poor compatibility, cannot support multiple systems and deployment methods, and lack fine-grained permission management, logging, and format conversion functions, thus failing to meet the compliance requirements and actual usage needs of cultural resource management.

[0003] This invention completely avoids the shortcomings of the existing technology mentioned above. It adopts five core technical solutions: original multi-source heterogeneous layered acquisition, adaptive personalized standardization processing, dual-branch attention classification association, distributed encrypted storage and intelligent indexing, and dynamic self-optimization and intelligent updating. At the same time, it is equipped with auxiliary solutions such as intelligent interaction, permission management, and media adaptation to achieve high-precision, intelligent, secure, and dynamic construction of digital cultural intelligent resource library, solving all the pain points of the existing technology. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for constructing a digital cultural intelligent resource library, so as to solve or alleviate one of the technical problems existing in the prior art, and at least provide a beneficial option.

[0005] The technical solution of this invention is implemented as follows: a method and system for constructing a digital cultural intelligent resource database, comprising: Employing an original multimodal acquisition terminal combination, this system covers six major categories of digital cultural resources: intangible cultural heritage skills, ancient books and documents, cultural relics, folk culture, oral history, and cultural sites. Each resource category utilizes dedicated acquisition equipment and parameters to achieve high-precision, detailed acquisition: 8K ultra-high-definition 120fps multi-camera acquisition of dynamic images of intangible cultural heritage, capturing details of techniques in a 360° surround view; 1200dpi multispectral scanning of ancient book rubbings to eliminate the effects of yellowing and damage; dual-modal scanning of cultural relics using LiDAR and structured light, restoring the shape of cultural relics with 0.005mm precision; lossless acquisition of 24bit / 192kHz audio to suppress environmental noise; and BeiDou + GPS dual-mode positioning combined with UAV oblique photography to accurately acquire geographic information of cultural sites. A microsecond-level spatiotemporal synchronization calibration and acquisition self-checking mechanism is constructed, equipped with redundant backups, enabling full-scene adaptability acquisition and generating original resource sets with unique identifiers and spatiotemporal stamps, completely different from existing single-acquisition solutions.

[0006] Further optimization employs a five-layer hierarchical processing architecture, designing personalized processing algorithms for different types of resources: an improved CNN-RNN cascaded processing method for video resources achieves a frame alignment accuracy of 99.8%; an original text recognition and restoration algorithm processes ancient books, achieving a restoration accuracy of 98.5% and supporting 12 ancient fonts; 3D point cloud denoising and texture mapping processes cultural relic data, filling holes with a maximum diameter of 5mm; audio standardization processing improves the signal-to-noise ratio to over 50dB; and geographic information calibration error is ≤0.5m. A resource-individualized adaptation mechanism is established, adding dedicated processing modules for special resources such as damaged ancient books, complex cultural relics, and dialect audio. A self-defined DCIR standardization format is used to address existing format defects, achieving a standardization processing pass rate of over 98% and avoiding the shortcomings of existing fixed standardization algorithms.

[0007] Further optimization utilizes a dual-branch architecture: a multi-label attention Transformer model extracts 32-dimensional resource features, classifying them into 6 major categories, 28 subcategories, and 156 minor categories with an accuracy of 99.2%; a graph neural network constructs a resource association topology graph, uncovering inherent relationships with an association recognition accuracy of 97.8%. By introducing prior knowledge constraints from the cultural domain, a multi-task joint learning framework is constructed, supporting custom classification systems and adjustments to association relationships. Training is conducted using an original dataset, distinguishing it from existing single-classification, unrelated modeling schemes and addressing the problems of classification bias and insufficient association.

[0008] Further optimized, edge-cloud collaborative distributed storage utilizes high-frequency resources at the edge (response latency <50ms) and cloud storage capacity expandable to 100PB. AES-256 encryption and blockchain notarization ensure end-to-end immutability and traceability of resources. A multi-level intelligent index is constructed, supporting multiple retrieval methods with a response time ≤300ms and an accuracy of 98.9%. A dynamic scheduling, triple backup, and hierarchical access control mechanism is established, increasing storage utilization to over 85% and data recovery time ≤1 hour, overcoming the shortcomings of existing centralized storage, such as low security and slow retrieval.

[0009] Further optimization involves constructing a reward function based on reinforcement learning, and performing online incremental training on the processing model and classification association model every 15 days, achieving parameter optimization without retraining. An automatic resource update module is established to monitor newly added cultural resources, completing the entire process of collection, processing, and storage within 24 hours. Resource quality is checked every 30 days to remove low-quality resources, ensuring the integrity and timeliness of the resource database, unlike existing fixed-model, outdated solutions.

[0010] Further optimized, the system includes five core modules, with the addition of an intelligent interaction and visualization module, supporting VR immersive display and voice interactive retrieval; the device adopts a multi-core heterogeneous processor, multi-modal acquisition interface, and dual power backup to ensure continuous operation; the media program supports multiple systems and multiple deployment methods, including functions such as permission management, log recording, and format conversion, adapting to different usage needs and completely avoiding the technical defects of existing systems, devices, and media.

[0011] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions: 1. Multi-source heterogeneous layered acquisition, covering all types and all scenarios of digital cultural resources, with acquisition accuracy of 0.005mm~0.1mm, complete preservation of details, spatiotemporal synchronization error ≤10μs, and acquisition reliability of 100%, solving the problems of existing acquisition methods such as single acquisition, low accuracy, and unreliability; 2. Adaptive and personalized standardization processing: The processing algorithm is adapted to different resource characteristics, and the repair accuracy and standardization pass rate are both over 98%. The DCIR format is independently defined to solve the problems of insufficient standardization and poor interoperability in the existing system. 3. Dual-branch attention-based classification and association, with a classification accuracy of 99.2% and an association recognition accuracy of 97.8%. It supports custom classification systems, solving the problems of inaccurate classification and insufficient association in existing systems. 4. Distributed encrypted storage and intelligent indexing, storage capacity can be dynamically expanded to 100PB, resources are securely encrypted, the whole process is traceable, and the retrieval response time is ≤300ms, solving the problems of limited storage, low security and slow retrieval in existing systems; 5. Dynamic self-optimization and intelligent updating: the model is incrementally updated every 15 days, resources are automatically updated every 24 hours, and quality is checked every 30 days to ensure the timeliness and completeness of the resource library, solving the problem of existing models being fixed and lagging in updates; 6. Intelligent interaction and visualization, supporting VR immersive display and dialect voice search, adapting to different user needs, significantly improving resource accessibility and ease of use, and solving the problem of poor existing interactive experience; 7. The equipment is stable and reliable, with efficient multi-core heterogeneous processors, dual power supply backup, continuous uninterrupted operation, and no data loss; the media program is compatible with multiple systems, has complete functions, meets compliance requirements, and solves the problems of instability and weak media compatibility of existing equipment.

[0012] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in 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.

[0014] Figure 1 This is a flowchart of the overall construction method of the present invention; Figure 2 This is a detailed flowchart of the multi-source heterogeneous hierarchical acquisition method of the present invention; Figure 3 This is a flowchart illustrating the dual-branch attention classification association process of the present invention. Detailed Implementation

[0015] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0016] Example 1: Construction of a Digital Resource Database for Endangered Intangible Cultural Heritage Techniques Acquisition: An 8K ultra-high-definition 120fps 360° multi-camera acquisition terminal and a 24bit / 192kHz lossless audio acquisition terminal were activated to simultaneously acquire dynamic images and operational audio of endangered intangible cultural heritage techniques (such as traditional openwork carving). High-precision crystal oscillators were used to achieve microsecond-level synchronization, and a self-check was performed every 5 seconds. No abnormalities were found during the acquisition process. The images captured carving movements with details down to 0.01mm, and the audio suppressed environmental noise with a signal-to-noise ratio of 52dB. All acquired data were given a unique time stamp and device identifier. Standardization Processing: A dedicated processing module for intangible cultural heritage techniques was activated, employing an improved CNN-RNN cascade model to segment and align the dynamic images, removing 12 blurry frames and retaining 860 core technique action frames, achieving a frame alignment accuracy of 99.8%; the audio underwent noise reduction and normalization, standardized to a 48kHz sampling rate, increasing the signal-to-noise ratio to 53dB; the processed images and audio were converted to DCIR format, and technique-related attribute information was added; Classification and Association: Inputting a dual-branch attention model, extracting semantic, visual, and audio features from the image, constructing a 32-dimensional feature vector, and automatically classifying it into the subcategory "Intangible Cultural Heritage Techniques - Traditional Carving - Openwork Carving", the classification is accurate; through graph neural networks, the association between the technique and the corresponding tool artifacts and the oral history of the inheritors is mined, establishing 8 association records, and the association identification is accurate; Storage and Indexing: Standardized and associated resources are stored on an edge SSD array (high-frequency access) using AES-256 encryption. The entire process of blockchain evidence collection and processing generates a unique hash value. Feature indexes, classification indexes, and association indexes are built, supporting semantic and association retrieval. The retrieval response time is 220ms, and the retrieval accuracy is 99.1%. Hierarchical access permissions are set, allowing researchers to access high-precision resources while ordinary users can access basic display resources. Self-optimization and updating: Based on user access data (researchers search frequently), a reinforcement learning reward function is constructed, and the model parameters are incrementally updated after 15 days, improving the classification accuracy to 99.3%; new inheritance data of this intangible cultural heritage skill is monitored in real time, and the data is collected, processed, and stored within 24 hours to achieve dynamic updates of the resource library; after 30 days, a quality test is conducted, and there are no low-quality or invalid resources, so the resource library is of qualified quality.

[0017] Example 2: Construction of a Digital Resource Database for Damaged Ancient Books Data Acquisition: The 1200dpi multispectral scanning terminal was activated to scan the damaged ancient book rubbings. The full spectrum of 400nm~1000nm was used to eliminate the loss of details caused by yellowing and damage of the rubbings. The scanning accuracy was 1200dpi, and the acquisition frequency was 2 pages / minute. Time and space stamps and device identification were added simultaneously. The self-test module was activated during the acquisition process, and there were no acquisition errors exceeding the standard. Standardized processing: A dedicated processing module for damaged ancient books is activated, employing original text recognition and restoration algorithms to extract text and repair damage on scanned rubbings. Sixteen damaged areas are repaired, with a text recognition accuracy of 98.5%. It supports accurate recognition and unified encoding of ancient clerical script fonts. The restored text is then standardized and converted to DCIR format, with attributes such as the book's date, author, and content summary added. Simultaneously, fragmented pages of ancient books are integrated into complete chapters based on semantic association using a fragment integration module. Classification and Association: Inputting into a dual-branch attention model, extracting semantic and visual features of the text, automatically classifying it into the subcategory "Ancient Books and Documents - Ancient Clerical Script Documents - Eastern Han Dynasty", with accurate classification; using graph neural networks to mine the association between the ancient book and the corresponding historical period, cultural site, and folk culture, establishing 12 association records with an association recognition accuracy of 97.9%; Storage and Indexing: The processed ancient book resources are stored on a cloud-based distributed disk array, using AES-256 encryption and blockchain notarization to achieve full-process traceability; a multi-level intelligent index is built to support text search, date search, and related search, with a search response time of 280ms and a search accuracy of 98.9%; a triple backup strategy is adopted to ensure resource security; Self-optimization and updating: Based on researchers' search feedback and newly added ancient book data, the model parameters are incrementally updated after 15 days, and the accuracy of text restoration is improved to 98.7%; newly discovered damaged ancient books of the same type are monitored in real time, and the collection, processing and storage are completed within 24 hours; after 30 days, quality inspection is carried out, and one low-resolution scan resource is removed to ensure the quality of the resource library.

[0018] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method and system for constructing a digital cultural intelligent resource database, characterized in that: Includes the following steps: S1. Construct a multi-source heterogeneous digital cultural resource hierarchical acquisition architecture. Through an original multimodal acquisition terminal, simultaneously acquire different types and formats of original digital cultural resources. These original digital cultural resources include dynamic images of intangible cultural heritage skills, rubbings of ancient texts, 3D scan data of cultural relics, folk audio and video, oral history recordings, and geographical information of cultural sites. The dynamic images of intangible cultural heritage skills are captured using 8K ultra-high-definition at 120fps, equipped with 360° surround multi-camera synchronous acquisition, achieving 0.01mm-level detail capture of skill movements. The rubbings of ancient texts are scanned using high-resolution linear array scanning (1200dpi resolution), equipped with multispectral imaging technology (400nm~1000nm). (Full spectrum coverage) eliminates the loss of details caused by yellowing and damage of rubbings. The 3D scanning of cultural relics adopts dual-mode scanning of LiDAR + structured light, with a scanning accuracy of 0.005mm and a collection frequency of 1000 points / second. The folk audio adopts 24bit / 192kHz lossless acquisition and is equipped with an environmental noise adaptive suppression module. The oral history recording adopts dual-channel synchronous acquisition and synchronously records the facial expression images of the speaker. The geographical information of cultural sites adopts Beidou + GPS dual-mode positioning, combined with UAV oblique photography (resolution 0.1m), to collect 3D topographic and architectural distribution data of the site. All original resources are added with unique spatiotemporal stamps and acquisition device identifiers to form a multi-source heterogeneous original resource collection. S2. An adaptive heterogeneous resource standardization processing model is established to perform hierarchical standardization processing on the original resources collected in S1. The first layer uses an improved convolutional neural network (CNN)-recurrent neural network (RNN) cascade model to perform action segmentation and frame alignment on intangible cultural heritage dynamic images and folk custom videos, removing blurry and redundant frames and retaining core skill action frames, with a frame alignment accuracy of 99.8%. The second layer uses an original text recognition and restoration algorithm to extract text, repair damage, and standardize fonts on ancient book rubbings, with a restoration accuracy of 98.5%, supporting accurate recognition and unified encoding of 12 ancient fonts such as oracle bone script, bronze script, and clerical script. The third layer uses three-dimensional point... The cloud denoising and texture mapping algorithm performs point cloud denoising, hole filling, and texture optimization on the 3D scan data of cultural relics. The maximum diameter of the filled hole can reach 5mm, and the texture mapping error is ≤0.01mm. The fourth layer uses audio feature extraction and standardization algorithms to denoise and normalize folk audio and oral history recordings, standardizing the audio sampling rate to 48kHz and improving the signal-to-noise ratio to over 50dB. The fifth layer uses a geographic information calibration algorithm to perform coordinate calibration and format unification on the geographic data of cultural sites, converting the geographic information into the WGS84 standard coordinate format with a calibration error of ≤0.5m. Finally, a standardized and interoperable digital cultural resource set is generated. S3. Construct an intelligent classification and association modeling model based on a dual-branch attention mechanism to classify and associate the standardized resources output by S2. The first branch uses a multi-label attention Transformer model to extract semantic, visual, audio, and geographical features of various resources, constructing a 32-dimensional feature vector. It performs intelligent classification according to a classification system of 6 major categories, 28 subcategories, and 156 minor categories, including intangible cultural heritage skills, ancient books and documents, cultural relics, folk culture, oral history, and cultural sites, achieving a classification accuracy of 99.2%. The second branch uses a graph neural network (GNN) to construct a resource association topology graph, mining the inherent associations between different types of resources (such as intangible cultural heritage skills and corresponding cultural relics, ancient books and documents and corresponding historical periods, and cultural sites and corresponding folk customs). Each resource node contains 16-dimensional association features, achieving an association recognition accuracy of 97.8%, generating a resource association set that is clearly classified and closely associated. S4. Establish a distributed encrypted storage and intelligent index architecture to store and index the resource association set output by S3. An edge-cloud collaborative distributed storage architecture is adopted, with SSD arrays at the edge storing frequently accessed resources (response latency <50ms) and a distributed disk array in the cloud (storage capacity dynamically expandable to 100PB). All resources are encrypted end-to-end using the AES-256 encryption algorithm. Combined with blockchain notarization technology, the entire process of resource collection, processing, updating, and access is notarized to ensure that resources are tamper-proof and traceable. Each notarized record contains a unique hash value and timestamp. Simultaneously, a multi-level intelligent index is constructed, including feature index, classification index, association index, and spatiotemporal index, supporting multi-condition combined retrieval, fuzzy retrieval, and semantic retrieval. The retrieval response time is ≤300ms, and the retrieval accuracy reaches 98.9%. S5. Construct a dynamic self-optimization and intelligent update mechanism. Based on user access data, resource call data, and newly added data in the cultural field, construct a reinforcement learning reward function to perform online incremental training on the standardized processing model of S2 and the classification association model of S3. The model parameters are updated every 15 days. At the same time, establish an automatic resource update module to monitor newly added resources in the cultural field in real time (such as newly added intangible cultural heritage skills and newly discovered cultural relics), automatically trigger the collection, processing, classification, and storage process, realize the dynamic update of the resource library, and the update delay is ≤24 hours. Simultaneously, establish a resource quality assessment module to conduct quality testing on the resources in the resource library every 30 days, remove low-quality and invalid resources, and ensure the integrity and accuracy of the resource library.

2. The method and system for constructing a digital cultural intelligent resource database according to claim 1, characterized in that: The multi-source heterogeneous digital cultural resource hierarchical acquisition architecture described in S1 also includes: constructing a spatiotemporal synchronization calibration mechanism for acquisition terminals, using a high-precision crystal oscillator (accuracy ±0.005ppm) to provide a unified clock reference for all acquisition terminals, and achieving microsecond-level synchronization of data acquired by multiple cameras and devices through a timestamp alignment algorithm, with a synchronization error ≤10μs; simultaneously establishing a self-checking and redundant backup module for acquisition terminals, performing a self-check of acquisition accuracy every 5 seconds, and automatically initiating redundant acquisition terminal switching when the acquisition error of any acquisition terminal exceeds a preset threshold (image blur >5%, text scanning error >1.5%, 3D scanning error >0.01mm, audio signal-to-noise ratio <45dB, geolocation error >1m), while simultaneously generating an acquisition anomaly alarm and pushing it to the management personnel terminal to ensure the continuity and reliability of the acquired data; in addition, the acquisition architecture also supports multi-scenario adaptive acquisition, using fixed acquisition terminals for indoor ancient books and cultural relics, portable mobile acquisition terminals for outdoor cultural sites and folk activities, and vehicle-mounted mobile acquisition platforms for endangered intangible cultural heritage techniques, achieving full-scene, no-dead-angle acquisition, with an acquisition coverage rate of 100%.

3. The method and system for constructing a digital cultural intelligent resource database according to claim 1, characterized in that: The adaptive heterogeneous resource standardization processing model described in S2 also includes: establishing a resource-individualized adaptation processing mechanism; dynamically adjusting the threshold values ​​of processing algorithm parameters based on the characteristics of different types of digital cultural resources; specifically, for severely damaged ancient book rubbings, adding a multispectral layered restoration module to restore text and patterns in damaged areas through image overlay of different spectral channels, achieving a restoration success rate of 96.3%; for complex-shaped cultural relics (such as openwork and relief artifacts), adding a 3D point cloud partitioning processing module to apply different scanning and processing parameters to different areas of the artifact, ensuring the complete preservation of artifact details; and for dialect-based folk audio and oral history recordings, adding a dialect recognition and transcription module, supporting 36 The system accurately identifies local dialects and transcribes them into Standard Mandarin, achieving an accuracy rate of 95.7%. For fragmented cultural resources (such as scattered folk images and short oral narratives), a fragment integration module is added. Based on semantic and spatiotemporal associations, fragmented resources are integrated into complete resource packages. Simultaneously, a standardized format is established, converting all resources into the self-defined DCIR format (Digital Cultural Intelligent Resource Format). This format is compatible with existing mainstream formats (JPG, PNG, MP4, WAV, OBJ, etc.) and supports feature embedding and associated information storage, addressing the shortcomings of existing formats in supporting multi-dimensional associated information. After standardization, the reusability of resources is increased to 99.5%.

4. The method and system for constructing a digital cultural intelligent resource database according to claim 1, characterized in that: The intelligent classification and association modeling model based on the dual-branch attention mechanism described in S3 also includes: introducing prior knowledge constraints specific to the digital culture domain, transforming the classification standards, resource association rules, and historical context relationships in the cultural domain into model regularization terms, and embedding them into the loss function of the dual-branch model to avoid classification bias and association errors; simultaneously constructing a multi-task joint learning framework to simultaneously complete four major tasks: resource classification, association recognition, feature extraction, and abnormal resource detection. The model training uses an original digital culture dataset (containing 80,000+ digital culture resources of various types and 30,000+ association relationship data), achieving a classification accuracy of 99.2%, an association recognition accuracy of 97.8%, and an abnormal resource detection accuracy of 99.1% after training; in addition, the model supports a custom classification system, allowing administrators to add or modify classification categories and levels according to actual needs. After the classification system is updated, the model can adaptively adjust within 24 hours without retraining; at the same time, an association visualization module is constructed to present the resource association topology in a visual form, supporting manual adjustment and supplementation of association relationships to ensure the accuracy and completeness of association relationships.

5. The method and system for constructing a digital cultural intelligent resource database according to claim 1, characterized in that: The distributed encrypted storage and intelligent indexing architecture described in S4 also includes: constructing a dynamic storage resource scheduling mechanism that automatically adjusts the storage location of resources at the edge and in the cloud based on user access frequency and resource importance. High-frequency access and high-importance resources (such as national intangible cultural heritage techniques and first-class cultural relic data) are prioritized for storage at the edge, while low-frequency access resources are stored in the cloud. The scheduling latency is ≤100ms, and the storage utilization rate is increased to over 85%. Simultaneously, a resource backup mechanism is established, employing a triple backup strategy of "local backup + cloud backup + off-site backup." Local backup uses a RAID10 array, cloud backup uses multi-node redundant storage, and off-site backup... Utilizing offline storage media, the backup success rate reaches 100%, and data recovery time is ≤1 hour. The intelligent indexing architecture also supports dynamic index optimization, automatically adjusting index weights and optimizing search paths based on user search habits and frequency, further reducing search response time to within 200ms. In addition, the storage architecture supports hierarchical resource permission management, setting different resource access permissions based on user identity (administrators, researchers, and ordinary users). Ordinary users can only access basic resources, researchers can access high-precision resources, and administrators have full operation permissions. Permission verification uses facial recognition + key dual verification to ensure resource security.

6. The method and system for constructing a digital cultural intelligent resource database according to claim 1, characterized in that, include: The multi-source heterogeneous layered acquisition module is used to construct an 8K ultra-high-definition multi-camera acquisition terminal, a high-resolution multispectral scanning terminal, a LiDAR + structured light dual-modal 3D scanning terminal, a lossless audio acquisition terminal, and a BeiDou + GPS dual-mode geographic information acquisition terminal. It simultaneously acquires various types of digital cultural resources, such as intangible cultural heritage skills, ancient books and documents, cultural relics, folk culture, oral history, and cultural sites. It achieves microsecond-level spatiotemporal synchronous acquisition, acquisition accuracy self-checking, redundancy backup, and full-scene adaptation acquisition, generating a set of original resources with unique identifiers and spatiotemporal stamps. The adaptive heterogeneous resource standardization processing module, connected to the multi-source heterogeneous hierarchical acquisition module, is used to perform improved CNN-RNN cascade frame processing, original text recognition and restoration, 3D point cloud noise reduction and texture mapping, audio standardization, and geographic information calibration. It establishes a resource individualization adaptation processing mechanism, supports personalized processing of different types of resources, converts the original resources into the self-defined DCIR standardized format, and generates a standardized and interoperable digital cultural resource set. The restoration accuracy and standardization qualification rate are both over 98%. The dual-branch attention intelligent classification and association module is connected to the adaptive heterogeneous resource standardization processing module. It is used to achieve intelligent resource classification through a multi-label attention Transformer model, construct a resource association topology graph through a graph neural network, explore the inherent relationships between resources, introduce cultural domain prior knowledge constraints and a multi-task joint learning framework, support custom classification system and association relationship adjustment, and generate a resource association set with clear classification and close association. The distributed encrypted storage and intelligent indexing module, connected to the dual-branch attention intelligent classification and association module, is used to construct an edge-cloud collaborative distributed storage architecture. It employs AES-256 encryption and blockchain notarization technology to achieve secure resource storage and end-to-end traceability. It builds a multi-level intelligent index, supports multiple methods for rapid retrieval, and enables dynamic resource scheduling, triple backup, and hierarchical access control to ensure resource security and accessibility. The dynamic self-optimization and intelligent update module, connected to the adaptive heterogeneous resource standardization processing module, the dual-branch attention intelligent classification and association module, and the distributed encrypted storage and intelligent indexing module, respectively, is used to construct a reinforcement learning reward function based on user access data, resource call data, and newly added data in the cultural domain. This enables online incremental updates of the model and establishes an automatic resource update and quality assessment module, achieving dynamic updates and quality control of the resource repository to ensure its integrity, accuracy, and timeliness.

7. The method and system for constructing a digital cultural intelligent resource database according to claim 6, characterized in that, Also includes: The intelligent interaction and visualization module, connected to the distributed encrypted storage and intelligent indexing module, provides personalized interactive interfaces for users with different identities. Administrators can perform operations such as resource collection, processing, classification, and permission management; researchers can perform operations such as resource retrieval, downloading, and correlation analysis; and ordinary users can perform operations such as resource browsing and basic retrieval. Simultaneously, a 3D visualization display module is constructed to support immersive display of 3D models of cultural relics and 3D terrain of cultural sites. Using VR adaptation technology, users can immerse themselves in browsing the details of cultural relics and scenes of cultural sites through VR devices. The display accuracy reaches 0.01mm, and the immersive experience latency is ≤20ms. In addition, the intelligent interaction module also supports voice interaction retrieval. It adopts a voice recognition algorithm and supports voice retrieval in Mandarin and 36 local dialects. The voice recognition accuracy rate reaches 96.2% and the retrieval response time is ≤300ms. Simultaneously, a resource statistics and analysis module is built to provide real-time statistics on the quantity, update status, and access popularity of various resources in the resource library, generating statistical reports to provide decision support for managers.

8. A method and system for constructing a digital cultural intelligent resource database according to claims 1-5, characterized in that, include: Processor, memory, multi-modal acquisition interface, communication interface, display unit, input and output unit; The multi-modal acquisition interface is connected to 8K ultra-high definition multi-camera acquisition terminals, high-resolution multispectral scanning terminals, lidar + structured light dual-modal 3D scanning terminals, lossless audio acquisition terminals, and Beidou + GPS dual-mode geographic information acquisition terminals, supports high-speed acquisition and transmission of multi-source heterogeneous raw resources, with transmission rate ≥ 10Gbps and acquisition delay ≤ 50ms; The communication interface supports three modes of 5G, WiFi6 and optical fiber communication, realizes data interaction and linkage control with edge ends, cloud ends and user terminals, adopts AES-256 algorithm for data transmission encryption, with transmission delay ≤ 100ms and transmission success rate reaching 100%; The display unit adopts a high-resolution touch screen (resolution 3840×2160), which displays resource acquisition status, processing progress, classification results, storage status and retrieval results in real time, supports 3D visual display and report viewing, with screen response time ≤ 5ms; The input and output unit includes a keyboard, a mouse, a stylus, and audio input and output devices, which supports management personnel to perform operation input and issue instructions, and simultaneously supports acousto-optic prompt and voice broadcast for abnormal alarms; the processor adopts a multi-core heterogeneous processor (CPU+GPU+FPGA), with an operation rate ≥ 10TFLOPS, which can process multi-channel acquisition data, model operation, storage scheduling and other tasks at the same time, ensuring the fluency and efficiency of system operation; The device further includes a redundant power module, which supports dual power backup, can automatically switch to the standby power supply after power failure, and the standby power supply has a battery life of ≥ 8 hours, ensuring continuous operation of the device and avoiding data loss.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a digital culture intelligent resource library according to any one of claims 1 to 5 is implemented, and the computer program comprises: a multi-source heterogeneous resource acquisition driver, an adaptive heterogeneous resource standardization processing algorithm program, a dual-branch attention classification association model program, a distributed encrypted storage and intelligent indexing program, a dynamic self-optimization and intelligent updating program, a blockchain evidence storage program, an edge-cloud collaborative scheduling program, and an intelligent interaction and visualization program; All programs are encapsulated with an exclusive encryption algorithm for the digital culture field, combined with hash check technology to ensure that the programs cannot be tampered with or cracked, and protect the intellectual property rights of the algorithm at the same time; the computer program supports multi-system adaptation, can run on mainstream operating systems such as Windows, Linux and Unix, supports three deployment modes: stand-alone deployment, cluster deployment and cloud deployment, with low deployment difficulty and deployment time ≤ 24 hours; In addition, the computer program further comprises a log recording program, which records program operation status, data processing process and user operation behavior in real time, with a log storage time of ≥ 1 year, which facilitates troubleshooting and behavior traceability, and meets the compliance requirements of cultural resource management.

10. The computer-readable storage medium according to claim 9, characterized in that: The computer program also includes: a digital cultural resource quality assessment program, a resource access management program, a dialect recognition and transcription program, a fragmented resource integration program, and a custom classification system editing program; The digital cultural resource quality assessment program uses multi-dimensional assessment indicators (clarity, completeness, accuracy, and relevance) to automatically detect the quality of resources in the resource library, generate quality assessment reports, and support the automatic marking and deletion of low-quality resources. The resource access control program supports user authentication, access allocation, access modification, and access cancellation, enabling fine-grained access control. It supports simultaneous online operation by multiple users, with a concurrent user capacity of ≥1000. The dialect recognition and transcription program supports real-time recognition and Mandarin transcription of 36 local dialects, with a transcription speed of ≥100 characters / minute and a transcription accuracy of 95.7%. The fragmented resource integration program automatically integrates fragmented digital cultural resources into complete resource packages based on semantic features, spatiotemporal features, and correlation features, with an integration success rate of 97.3%. The custom classification system editing program supports administrators in adding, modifying, and deleting classification categories and levels. It also supports the import and export of classification systems. The edited classification system can be synchronized to the classification association model in real time without the need for model retraining, ensuring the flexibility and adaptability of the classification system. In addition, the computer program also supports resource format conversion, which can convert DCIR format to existing mainstream formats to meet the needs of different users, with a format conversion accuracy of 100%.

Images