A multi-modal modeling, tracing and verifying device and method for cultural relics

Abstract

The application discloses a kind of multi-modal modeling, tracing, verification device and method for cultural relics, belong to cultural relic digital protection technical field;Device includes: fully enclosed constant temperature and humidity collection bin, double-layer rotating platform, multi-modal sensor group, quantum dot mark reading and writing component and main control and data processing component;Method includes: S1: cultural relics multi-modal data acquisition;S2: cultural relics three-dimensional modeling;S3: generate physical characteristic code;S4: quantum dot mark encryption binding;S5: cultural relics digital twin construction;S6: cultural relics return and state intelligent verification;The application solves the current cultural relic digital management data acquisition single, encryption tracing mode rigor degree loss, verification accuracy is not enough.

Description

Technical Field

[0001] This invention belongs to the field of digital preservation technology for cultural relics, specifically relating to a multimodal modeling, tracing, and verification device and method for cultural relics. Background Technology

[0002] Currently, technologies such as 3D laser scanning, close-range photogrammetry, and UHF RFID tags are widely used in the field of digital management of cultural relics. The State Administration of Cultural Heritage has established a complete industry standard system covering data collection accuracy, format specifications, and archive management. Museums at all levels across the country have completed basic digital archiving of millions of cultural relics, and have initially realized the electronicization of basic information on cultural relics and rapid registration of entry and exit from the warehouse.

[0003] However, existing technologies still have core shortcomings: data collection is limited to a single dimension, only able to acquire surface geometry and visible light characteristics, and cannot detect potential damage such as subsurface damage and internal stress accumulation; RFID tags are physically separated from the artifact itself, and even with encryption technology, there is still a risk of them being peeled off, forged, and pasted onto imitations; verification of artifact return relies on manual visual inspection or single image comparison, making it difficult to identify micron-level anomalies and internal structural changes; data is mostly stored in the form of static archives, lacking the ability for dynamic monitoring and damage prediction throughout the entire life cycle, and cannot meet the urgent needs of preventive protection and intelligent management of cultural relics. Summary of the Invention

[0004] This invention provides a multimodal modeling, tracing, and verification device and method for cultural relics. The device enables fully automated, multi-dimensional, and deep physical characteristic acquisition of cultural relics with a single clamping. The modeling method employs a hybrid modeling technique combining 3D Gaussian splashing and DUSt3R to achieve ultra-fast and high-precision reconstruction. Unforgeable tracing is achieved through dual binding of the inherent characteristics of the cultural relic and quantum dot invisible markers. Based on a digital twin platform and twin network AI comparison, the entire life cycle status monitoring and intelligent verification of the cultural relic are completed. This invention solves the problems of current digital management of cultural relics, such as single data collection, lack of rigor in encrypted tracing methods, and insufficient verification accuracy.

[0005] To achieve the above-mentioned technical objectives, the present invention is implemented through the following technical solution: A multimodal modeling, tracing, and verification device for cultural relics includes: The fully enclosed temperature and humidity data acquisition chamber is designed with a fully enclosed light-shielding structure and a built-in temperature and humidity control system to provide a closed temperature and humidity-controlled environment for data acquisition and modeling of cultural relics. A double-layer rotating platform, consisting of an upper platform and a lower platform; The upper platform is designed with a transparent material. The lower platform is located directly below the upper platform, and the upper and lower platforms are coaxial; the upper and lower platforms use two independent rotary drive motors. The multimodal sensor array is configured in at least two regions. The first region includes several sets of the multimodal sensor array, which are evenly spaced on the lower platform. The second region also includes several sets of the multimodal sensor array, which are evenly spaced on the inner wall and top of the fully enclosed constant temperature and humidity acquisition chamber. The multimodal sensor arrays in the first and second regions are used to acquire 360° omnidirectional data features of the cultural relics. A quantum dot tagging and reading / writing component, comprising a quantum dot inkjet device and a fluorescence excitation detection device; The quantum dot inkjet device is used to make invisible marks on cultural relics; the fluorescence excitation detection device is used to read the mark information; The main control and data processing components include a built-in processor and a memory. The memory contains a computer-executable program, and the processor executes the computer-executable program to complete multimodal data processing, cultural relic modeling, tracing, and verification.

[0006] Preferably, the constant temperature and humidity control system includes: Temperature sensor used to monitor the temperature in a fully enclosed constant temperature and humidity acquisition chamber in real time; A humidity sensor is used to monitor the humidity in a fully enclosed, temperature- and humidity-controlled collection chamber in real time. The PLC is used to receive temperature and humidity data, process, analyze and compare the data, and issue control commands based on the comparison results. Temperature control equipment is used to receive control commands from the PLC, respond to them, and adjust the temperature to maintain a constant temperature inside the chamber. A humidity control device is used to receive control commands from the PLC, respond to adjust the humidity, and maintain constant humidity in the chamber. The fully enclosed constant temperature and humidity collection chamber is equipped with an inert gas protection device to create a nitrogen-filled environment inside the chamber, thus protecting the cultural relics and ensuring their stable condition.

[0007] Preferably, the multimodal sensor group includes at least: Terahertz imagers are used to collect the subsurface structure and material distribution characteristics of cultural relics. They emit terahertz waves of 0.1~3THz, which can penetrate 0~5mm into the surface of cultural relics to obtain subsurface information that is difficult to accurately identify with the naked eye and X-rays, such as internal interlayers, cavities, dark cracks, repair traces, and different material layers. They emit no ionizing radiation and are safe and non-destructive to all cultural relics. Polarized light cameras are used to collect data on the stress distribution and material anisotropy of cultural relics; they capture reflected light in different polarization directions to identify areas of stress concentration (potential crack origins) inside the cultural relics, distinguish between the original material and the restoration material (significant differences in polarization characteristics), and detect surface microcracks and material uniformity. Laser speckle interferometer is used to collect nanoscale micro-deformation and vibration characteristics of cultural relics; by measuring the changes in speckle pattern generated by laser irradiation, nanoscale surface micro-deformation can be measured, the trend of loosening, delamination and fine cracking of cultural relic structure can be detected, and the overall structural stability of cultural relic can be evaluated. White light supplementary lighting and cameras are used to collect basic visible light features of cultural relics surfaces; they provide uniform, shadow-free white light illumination, which, together with imaging systems, acquires appearance information such as texture, color, shape, scratches, and stains on the surface of cultural relics, and forms the basis for 3D modeling and surface appearance inspection.

[0008] Another objective of this invention is to provide a multimodal modeling, tracing, and verification method for cultural relics, including: S1: Multimodal data acquisition of cultural relics. The cultural relics are placed on the upper platform of a double-layer rotating platform in a fully enclosed constant temperature and humidity acquisition chamber, maintaining a constant temperature, humidity and inert gas environment inside the chamber; at the same time, the multimodal sensor groups in two areas are activated; the upper platform is started to rotate, and the dual-area multimodal sensor groups synchronously acquire terahertz imaging data, polarized light imaging data, laser speckle interferometry data and white light image data of the entire cultural relics. S2: 3D modeling of cultural relics. A hybrid modeling algorithm combining 3D Gaussian splashing (3DGS) and DUSt3R is used to generate a high-precision 3D model of cultural relics during the acquisition of multimodal feature data of cultural relics in S1. S3: Generate physical feature code, extract the inherent features of the cultural relic, including at least terahertz feature spectrum, polarization stress fingerprint, and speckle interference features, and generate a unique physical feature code for the cultural relic. S4: Quantum dot tag encryption binding. Quantum dot inkjet equipment is used to print quantum dot invisible tags in hidden locations of cultural relics, such as the bottom and inside. The physical feature code generated in S3 is encrypted and bound with the quantum dot tag ID. S5: Construction of a digital twin of cultural relics: All collected data, 3D models, feature codes, and marking information of cultural relics are integrated and entered into the digital twin management platform to complete the initial archiving; all life-cycle operations of cultural relics, including warehousing, warehousing, exhibition, and restoration, are recorded on the digital twin management platform; the digital twin management platform also predicts the development trend of cultural relic diseases based on historical data and AI algorithms and issues early warnings. S6: Intelligent verification of cultural relic return and status. Scan the quantum dot invisible markers on the cultural relic to quickly retrieve the corresponding digital twin and historical benchmark data; the multimodal sensor group automatically collects the current multimodal data of the cultural relic and generates a real-time 3D model and physical feature code; using the twin network AI algorithm, the real-time data and historical benchmark data are compared in multiple dimensions; an automatic verification report is generated, quantifying and marking abnormal areas and their severity; the verification data is updated to the digital twin platform to improve the full life cycle record of the cultural relic.

[0009] Preferably, the specific method for constructing a three-dimensional model of a cultural relic using the hybrid modeling algorithm of 3D Gaussian splash (3DGS) combined with DUSt3R is as follows: S2.1: White light image acquisition, based on white light supplementation in dual regions and the camera acquiring omnidirectional images of the cultural relic at 15° intervals, to acquire at least 24 high-resolution images of the cultural relic; and to perform preprocessing on the images including removing overexposed, underexposed and blurred images, lens distortion correction and white balance calibration, and extracting image feature points (SIFT). S2.2: Fast global initialization of DUSt3R. The 24 unordered images obtained in S2.1 are input into the pre-trained DUSt3R model. The model automatically matches the correspondence between images through the self-attention mechanism and outputs: the rotation matrix R and translation vector t of each image, the camera extrinsic parameters, the globally consistent sparse point cloud, and the depth confidence of each point. Using the known rotation angle and center position of the rotating platform, the point cloud output by DUSt3R is scaled to solve the problem of scale uncertainty in DUSt3R reconstruction; the origin of the coordinate system is aligned with the center of the rotating platform, and the Z-axis is vertically upward to ensure that the model size is consistent with the actual size of the cultural relic, with an error of <0.1mm; Point cloud filtering and optimization: remove outliers with depth confidence below 0.8; remove background point cloud, retaining only the cultural relic area; generate a uniformly distributed initial point cloud to provide high-quality input for 3DGS; S2.3: 3DGS Dense Reconstruction and Parameter Optimization, 3D Gaussian Sphere Initialization: The sparse point cloud generated by DUSt3R is used as the center position of the initial Gaussian sphere; the initial covariance of each Gaussian sphere is set to isotropic, with a size of 1 / 3 of the average distance between adjacent points; the initial color is obtained by interpolating the pixel values ​​of the corresponding image; Differentiable rendering and iterative optimization: Using the differentiable rendering pipeline of 3DGS, a 3D Gaussian sphere is projected onto a 2D image plane to generate a rendered image; the mean square error (MSE) loss between the rendered image and the original acquired image is calculated; the position, covariance, color, and transparency parameters of all Gaussian spheres are iteratively optimized through backpropagation; the Gaussian spheres are automatically split and merged during the optimization process: when the anisotropy of a Gaussian sphere exceeds a threshold, it is split into two Gaussian spheres; when two Gaussian spheres are too close and have similar parameters, they are merged; for transparent cultural relics such as jade and glass, a transmittance parameter is introduced into the 3DGS optimization to accurately simulate the light transmission effect; Enhanced and optimized artifact details: For key detail areas such as inscriptions and patterns on artifacts, the Gaussian sphere density in these areas is automatically increased by 2 to 3 times; Perceptual Loss is introduced to improve the texture details and realism of the model; Iterative optimization is performed 1000 to 2000 times until the loss function converges; S2.4: Model post-processing and accuracy verification, mesh generation: Generate a traditional triangular mesh model, and extract the mesh from the 3D Gaussian sphere using the Poisson surface reconstruction algorithm; perform hole filling and smoothing processing, and remove noise and burrs; Texture mapping optimization: Map the original high-resolution image onto the surface of the 3D model; use the Poisson fusion algorithm to eliminate lighting differences and seams between different images; Accuracy verification: Use a high-precision laser rangefinder to measure the key dimensions of the cultural relic (such as height and diameter); compare them with the corresponding dimensions of the 3D model to ensure an overall accuracy of ±0.05mm; perform local accuracy verification on detailed areas such as inscriptions and patterns to ensure that details are clearly distinguishable.

[0010] Preferably, the physical feature code generation process is as follows: S3.1: Extraction of intrinsic features of different modes 1) Terahertz characteristic spectrum extraction: extracting the molecular vibrational characteristics and subsurface structure distribution features of the cultural relic material; Data preprocessing: The three-dimensional data cubes acquired by the terahertz imager are subjected to the following: temporal domain denoising; reflectivity normalization to eliminate the influence of light source intensity fluctuations; and background subtraction to remove background signals from the acquisition chamber and platform. Spectral feature extraction: For the terahertz spectral curve of each spatial pixel, extract 12 key feature parameters such as peak position, valley position, full width at half maximum (FWHM), slope, and inflection point; generate a 211-dimensional original spectral feature vector, and use PCA to reduce the dimensionality to 32 dimensions, retaining 99.5% of the information; Spatial distribution feature extraction: Material clustering is performed based on spectral features to generate a material distribution map of the artifact surface; statistical features such as the number of material regions, area ratio, boundary shape, and relative position are extracted. Subsurface structure feature extraction: Extract the terahertz reflection intensity distribution at different depths (0.5mm, 1mm, 2mm, 3mm, 5mm); calculate the texture entropy, contrast, and correlation of each depth layer to generate 15-dimensional subsurface structure features; Terahertz feature vector: The above features are concatenated to generate a 63-dimensional terahertz feature vector; 2) Polarization stress fingerprint extraction: Extracting the internal stress distribution and material anisotropy characteristics of cultural relics; Data preprocessing: The images at four polarization angles (0°, 45°, 90°, and 135°) acquired by the polarization camera are subjected to the following: lens distortion correction, illumination non-uniformity correction, and Stokes parameter calculation (S0, S1, S2). Polarization parameter calculation: Calculate the degree of polarization (DOP) and angle of polarization (AOP) for each pixel; generate polarization degree map and polarization angle map; Stress distribution feature extraction: Based on the linear relationship between polarization degree and stress, the stress distribution map of the artifact surface is obtained by inversion; the maximum, minimum, average, and standard deviation of the stress distribution are extracted; the watershed algorithm is used to identify stress concentration areas and extract their location, area, shape, stress gradient and other features. Material anisotropy feature extraction: Calculate the spatial distribution histogram of polarization angles (36 bins); extract features such as anisotropy index and principal polarization direction; Polarization stress fingerprint vector: The above features are concatenated to generate a 62-dimensional polarization stress feature vector; 3) Laser speckle interferometry feature extraction to extract the nanoscale micro-morphology and structural vibration characteristics of the artifact surface; Data preprocessing: The speckle image sequence acquired by the laser speckle interferometer is subjected to the following processes: mean filtering for noise reduction, contrast enhancement, and phase unwrapping. Surface micromorphology feature extraction: Calculate statistical features such as contrast, correlation coefficient, entropy, and energy of speckle patterns; extract surface roughness parameters (Ra, Rq, Rz); generate surface height distribution histogram (32 bins); Vibration characteristic feature extraction: Apply a small sinusoidal excitation (amplitude <1μm) to the cultural relic and collect speckle interferograms at different frequencies; obtain the natural frequency, damping ratio, mode shape and other modal parameters of the cultural relic through Fourier transform; extract the nodal position, amplitude distribution and other features of each mode shape; Speckle interference feature vector: The above features are concatenated to generate a 60-dimensional speckle interference feature vector; S3.2: Cross-modal feature fusion and normalization 1) Feature vector concatenation: The terahertz (63-dimensional), polarization stress (62-dimensional), and laser speckle (60-dimensional) feature vectors are concatenated to generate a 185-dimensional original fused feature vector; 2) Adaptive weighted fusion: Automatically adjusts the weights of each mode according to different types of cultural relics. General weights: terahertz 35%, polarization stress 30%, laser speckle 35%; Bronze artifacts: Terahertz 40%, polarization stress 30%, laser speckle 30%; Ceramics: Terahertz 30%, polarization stress 35%, laser speckle 35%; Jade: Terahertz 35%, polarization stress 25%, laser speckle 40%; 3) Normalization: The Z-score normalization method is used to normalize each dimension of the fused feature vector to a distribution with a mean of 0 and a variance of 1; 4) Dimensional unification: The 185-dimensional feature vector is mapped to a 256-dimensional unified feature space through a fully connected layer, generating a 256-dimensional fused feature vector; S3.3: Robust Physical Feature Generation 1) Robust hash encoding: The SM3 hash algorithm (Chinese national cryptographic standard) is combined with robust hashing technology to convert the 256-dimensional fused feature vector into a 256-bit binary physical feature code; 2) Enhanced stability: A tolerance range of ±5% is set for each dimension of the feature vector to eliminate minor changes caused by environmental factors such as temperature and humidity; a voting mechanism is adopted to perform majority voting on the feature vectors collected multiple times to generate the final stable physical feature code; 3) Uniqueness verification: Calculate the Hamming distance between the physical feature codes of different cultural relics to ensure that the Hamming distance is ≥128 bits (uniqueness probability >99.9999%); collect data from the same cultural relic multiple times to ensure that the Hamming distance of the physical feature codes is ≤8 bits (stability >99.99%). 4) Feature code storage and association: The physical feature code is encrypted and bound to the digital twin ID of the cultural relic; it is stored in the digital twin management platform of the cultural relic and the quantum dot invisible tag.

[0011] Preferably, the specific method for combining the physical feature code with the quantum dot tag ID for encrypted binding is as follows: S4.1: Cultural relic safety grade quantum dot ink and dedicated printing equipment 1) Quantum dot ink preparation: Core-shell structured near-infrared quantum dots are used, with an emission wavelength of 800~900nm. The outer shell is coated with silica, and the core is CdSe / ZnS; 0.1% of polyethylene glycol, a cultural relic preservation agent, is added. 2) Retrofitting of dedicated quantum dot inkjet printing equipment It employs a piezoelectric micron-level printhead with a minimum droplet volume of 1pL and a printing accuracy of ±5μm; it is equipped with a three-axis precision motion platform with a positioning accuracy of ±2μm; it integrates a high-resolution industrial camera to automatically identify the hidden locations of cultural relics and plan the printing path; it has a built-in national cryptographic hardware encryption chip (SM2 / SM3 / SM4), and all encryption calculations are completed within the chip without going through the host operating system; the printhead maintains a distance of 1~2mm from the surface of the cultural relic, enabling completely contactless printing. S4.2: Quantum Dot Invisibility Mark Printing 1) Printing location selection and planning: Prioritize areas that do not affect the display and research of cultural relics, such as the bottom edge, inner recesses, gaps in inscriptions, and the back of the base; avoid areas such as the front of the cultural relic, areas with dense patterns, areas prone to wear, and areas with fragile materials. The system automatically plans the printing path to ensure that the mark size is controlled within 2mm×2mm and does not exceed the hidden area; after confirming the printing position, it adjusts the orientation and angle of the print head and the distance from the cultural relic; 2) High-precision invisible tag printing: The device generates a 128-bit globally unique quantum dot tag ID, including: a 64-bit device number + timestamp to ensure uniqueness; a 32-bit random number to prevent prediction; and a 32-bit CRC checksum for reading and verification. The tag ID is encoded into a QR code format, using the highest level of error correction, H-level; it can still be read correctly even if the tag is 30% damaged. The print head sprays quantum dot ink point by point onto the surface of the artifact according to the planned path, forming a micron-level QR code pattern; after printing, it is dried with hot air. 3) Print quality verification: The device automatically switches to fluorescence detection mode and emits 850nm near-infrared light to excite the quantum dot markers; the camera acquires fluorescence images, automatically identifies and decodes the marker ID; verifies whether the decoded ID is consistent with the original ID, and if they are inconsistent, reprints; takes fluorescence photos and natural light photos of the marker location and stores them in the digital twin archive of the cultural relics. S4.3: Hardware-level encryption binding of physical signature and quantum dot ID 1) Encryption binding condition settings: All encryption operations are completed in the device's built-in hardware encryption chip without connecting to any network; a hash algorithm is used to ensure that the original physical signature cannot be deduced from the encryption result; the quantum dot ID contains the hash value of the physical signature, and the physical signature contains the hash value of the quantum dot ID; any modification to the binding relationship will result in verification failure. 2) Binding steps: a. Data input: Input the 256-bit physical feature code of the cultural relic generated by S3 and the 128-bit quantum dot mark ID printed by S4 into the hardware encryption chip; b. First-level hash operation: The 256-bit physical feature code is hashed using the national cryptographic SM3 algorithm to generate a 256-bit physical feature hash value H1; the 128-bit quantum dot ID is hashed using the national cryptographic SM3 algorithm to generate a 256-bit ID hash value H2. c. Two-way binding operation: Concatenate H1 with the quantum dot ID to generate 384-bit data D1; concatenate H2 with the physical feature code to generate 512-bit data D2; perform SM3 hash operation on D1 and D2 respectively to generate binding credentials H3 (256 bits) and H4 (256 bits); d. Encrypted storage: Using the national cryptographic SM4 algorithm, H3 and H4 are encrypted using the unique master key built into the device; the encrypted binding certificate is written into the encrypted storage area marked by quantum dots; the quantum dot ID, H1, H2, H3, H4 and binding timestamp are encrypted and stored in the cultural relic digital twin management platform. e. Binding Verification: Read the encrypted binding credential from the quantum dot marker, decrypt it to obtain H3 and H4; recalculate H1 and H2, and verify whether they are consistent with the stored values; if the verification is successful, the binding is completed and a binding report is generated.

[0012] Preferably, the method for reading and verifying the binding relationship of the quantum dot teleportation tag is as follows: 1) Quantum dot tag reading Using a dedicated fluorescence excitation detection device, 850nm near-infrared light is emitted to illuminate the marked location of the cultural relic; the device collects the 900nm fluorescence signal emitted by the quantum dots, decodes it to obtain the quantum dot ID and the encrypted binding certificate; decrypts the binding certificate to obtain H3 and H4; 2) Binding relationship verification Retrieve the physical feature hash value H1 corresponding to the quantum dot ID from the digital twin platform; re-collect the current physical features of the cultural relic, generate a real-time physical feature code, and calculate its real-time hash value H1'; verify whether H1 and H1' are consistent, and consider them consistent if the Hamming distance is ≤8 bits; if the verification is successful, the identity of the cultural relic and the binding relationship are confirmed to be valid, otherwise it is judged as abnormal; 3) Handling Abnormal Situations a. Mark reading failed: Adjust the excitation light angle and intensity and try reading again; if it still fails, you can search the digital twin platform through other features of the cultural relic to find the corresponding record. After authorization by two or more administrators, you can reprint the mark near the original location and update the binding relationship. b. Damaged tag: If the tag is partially damaged, try to recover it using the error correction capability of the QR code; if it is completely damaged, reprint the tag after approval, mark the old tag ID as "invalid" and keep the record permanently; c. Updates after cultural relic restoration: After the cultural relic is restored, physical characteristics are re-collected to generate new physical characteristic codes; the binding relationship is updated, while the historical physical characteristic codes and binding records are retained to form a complete cultural relic status change archive.

[0013] Preferably, the specific method for predicting the development trend of cultural relic diseases based on the digital twin management platform is as follows: S5.1: Construction of a Multimodal Temporal Disease Database 1) Core data collection and integration Multimodal detection data: Terahertz subsurface crack data, polarization stress distribution data, laser speckle micro-deformation data, and surface texture data collected in each iteration are aligned by timestamps to form a time sequence; Environmental monitoring data: minute-level continuous monitoring data of environmental conditions such as temperature and humidity, light intensity, carbon dioxide concentration, and vibration acceleration in the storage / exhibition environment of cultural relics; Manually recorded data: Structured data such as cultural relic restoration records, maintenance records, exhibition records, and transportation records; Disease labeling data: Label data including the type, location, severity, and stage of development of diseases, as marked by cultural relics experts; 2) Data preprocessing and standardization Interpolation completion, outlier removal, and smoothing are performed on time-series data; A unified data format and coordinate system were used to map all disease characteristics to the three-dimensional spatial position of the digital twin of the cultural relic; A four-dimensional data index of "cultural relic-part-disease-time" is constructed to support fast retrieval by part and disease type; S5.2: Extraction of Disease-Related Features 1) For each diseased part, extract the temporal feature sequence for more than 6 consecutive months; 2) Calculate derived features such as the rate of change, acceleration, and fluctuation amplitude of the characteristic; 3) Use mutual information to screen key features with a correlation > 0.7 with disease development, and form a disease feature vector; S5.3: Hybrid prediction model construction, adopting a hybrid architecture of "data-driven AI model + physical mechanism model" to balance prediction accuracy and interpretability; 1) Basic Model of Physical Information Neural Network (PINN) The physical properties of cultural relics materials, such as mechanical, thermal, and chemical properties, are embedded into the neural network as prior knowledge; the loss function consists of two parts: data loss (error between predicted and actual observed values) + physical loss (penalty for violating physical laws); It solves the problems of traditional deep learning models requiring a large amount of labeled data and having poor generalization ability, and can still maintain high prediction accuracy under small sample conditions; 2) Set up dedicated prediction models for different diseases a) Crack propagation prediction model Input: Crack history length / depth sequence, stress concentration value sequence, ambient temperature and humidity sequence; Model: PINN combined with Extended Finite Element Method (XFEM); Output: Predictions for crack length, depth, and propagation direction for the next 3 / 6 / 12 months; Key technology: The Paris formula in fracture mechanics is embedded into a neural network as a physical constraint to simulate the crack propagation law under cyclic stress. b. Material aging prediction model Input: Terahertz spectral characteristic sequence, cumulative illumination duration, and cumulative temperature and humidity fluctuation values; Model: Temporal Transformer + Material Aging Dynamics Equations; Output: Predictions of material aging degree and mechanical property degradation rate over the next 1 / 3 / 5 years; Key technologies: Simulating the effect of temperature on aging rate based on the Arrhenius equation, and simulating the degradation effect of light on materials based on the Beer-Lambert law; c. Structural loosening prediction model Inputs: natural frequency sequence, damping ratio sequence, vibration acceleration sequence; Model: LSTM + Modal Analysis Theory; Output: Structural loosening risk level and predicted loosening location for the next 1 / 3 / 6 months; Key technology: Inverting the attenuation law of structural connection stiffness through changes in vibration mode parameters; 3) Multi-model fusion and uncertainty quantification The weighted average method is used to fuse the prediction results of multiple models, and the weights are dynamically adjusted according to the historical prediction accuracy of each model. The Monte Carlo dropout method is introduced to quantify the uncertainty of the prediction results and to give the confidence interval of the prediction values; When the forecast uncertainty exceeds 30%, a manual review process is automatically triggered. S5.4: Tiered Early Warning and Decision Support 1) Risk level classification Based on the predicted severity of the damage and the value of the cultural relics, the early warning is divided into four levels: Level 1 Warning (Red): Fatal damage (such as fracture or collapse) is predicted to occur within 3 months, and protective measures must be taken immediately; Level II Warning (Orange): Severe diseases are predicted to occur within 6 months, requiring special inspection and repair. Level 3 Warning (Yellow): Common diseases are predicted to occur within 12 months, and monitoring frequency needs to be increased; Level 4 warning (blue): There is a potential risk of disease; routine monitoring should be maintained. 2) Early warning triggering mechanism When the predicted disease parameters exceed the threshold of the corresponding level, an early warning will be automatically triggered; Taking into account the historical, artistic, scientific, and fragile value of cultural relics, the early warning threshold is dynamically adjusted. For warning signals that appear repeatedly in the same location, the warning level will be automatically upgraded. 3) Intelligent decision support Based on the knowledge base, targeted protection measures are automatically generated, such as adjusting environmental parameters, limiting exhibition time, and carrying out preventive repairs. The predicted location, development trend, and impact range of damage are displayed in three dimensions on the digital twin of cultural relics; Generate early warning reports, including disease details, prediction results, risk assessments, and protection recommendations, and push them to relevant management personnel; S5.5: Continuous Iterative Optimization of the Model After each collection of artifact condition data and restoration of damage, new data is added to the training set to incrementally update the model; The model's predictive accuracy should be evaluated at least quarterly, and the model parameters should be adjusted based on the evaluation results. An expert feedback mechanism has been established, allowing cultural relics experts to annotate and correct the prediction results, thereby continuously improving the model's performance.

[0014] Preferably, the specific process for verifying the return of cultural relics is as follows: S6.1: Verification Initialization and Data Preparation Rapid identification: Scan the quantum dot invisible markers on cultural relics using a fluorescence excitation detection device, and decode them to obtain a 128-bit marker ID and encrypted binding credential; Baseline data retrieval: Quickly retrieve the artifact's data from the digital twin platform using its tagged ID. The initial multimodal reference data package includes: terahertz cube, polarization stress map, speckle interferogram, and white light 3D model; Historical data and anomaly records from each verification; At least metadata including the material, age, and preservation environment of the artifact; Real-time data acquisition: The integrated device is activated to automatically acquire the current multimodal data of the cultural relics, and the acquisition parameters are completely consistent with the initial archiving. S6.2: Data Preprocessing and 3D Spatial Alignment Preprocessing for real-time acquired single-modal data: Terahertz data: Wavelet denoising, reflectivity normalization, and background subtraction were performed; Polarization data: Complete Stokes parameter calculation, polarization degree / polarization angle diagram generation, and illumination correction; Speckle data: Phase unwrapping, micro-deformation field calculation, and noise filtering were performed; White light data: Complete distortion correction, white balance, and contrast enhancement; Global 3D Spatial Alignment An improved ICP algorithm was used to register the real-time acquired white light point cloud with the reference 3D model, with a registration accuracy of ±0.02mm; The rotation matrix and translation vector obtained from the registration are applied to all modal data to make the real-time data and the reference data completely aligned in the same three-dimensional coordinate system. For each three-dimensional vertex on the surface of the cultural relic, establish a one-to-one correspondence between real-time data and benchmark data; S6.3: Multimodal Siamese Network Feature Extraction and Fusion. A two-branch symmetric Siamese network architecture is adopted, with the two branches sharing weights. Real-time data and benchmark data are input respectively, and the corresponding multimodal fusion feature vectors are output. A modality-specific feature extraction branch is designed with independent feature extraction networks tailored to the characteristics of different modal data. Employing a triple attention mechanism to dynamically fuse multimodal features: Spatial attention: Automatically focuses on key feature areas of cultural relics, such as inscriptions, patterns, and seams, while suppressing background interference; Channel attention: Automatically selects the feature channels that are most valuable to the current alignment task; Modal attention: Automatically adjusts the weights of each modality based on the type of artifact, for example: Jade: Laser speckle (40%) > Terahertz (30%) > Polarized (20%) > White light (10%) Bronze artifacts: Terahertz (35%) > Polarized (30%) > Laser speckle (20%) > White light (15%) Painting and calligraphy: White light (40%) > Terahertz (30%) > Polarized light (20%) > Laser speckle (10%) The final result is a 1024-dimensional global fusion feature vector; Similarity calculation: Calculate the cosine similarity between the global fused feature vectors of real-time data and benchmark data; Simultaneously, the similarity of independent feature vectors for each modality is calculated to obtain a multi-dimensional similarity score; Set a dynamic threshold: It is automatically adjusted based on the material of the cultural relic, the storage environment, and the last verification time, and is usually 0.95~0.98; If the global similarity is greater than or equal to the threshold, it is judged as normal; if it is less than the threshold, it enters the anomaly detection process. S6.4: Anomaly Area Location and Quantitative Analysis Anomaly heatmap generation: Calculate the differences between real-time data and baseline data on feature maps at each level; Upsampling restores the feature differences to the original image space, generating an anomaly heatmap. The intensity of the colors in a heatmap indicates the degree of abnormality: red indicates severe abnormality, yellow indicates moderate abnormality, and blue indicates normal. Abnormal region segmentation and extraction: Threshold segmentation combined with morphological operations is used to extract continuous anomalous regions from anomaly heatmaps; Each abnormal area is reconstructed in three dimensions and mapped to the corresponding position in the digital twin of the cultural relic; Abnormal parameter quantification calculation: The following quantization parameters are automatically calculated for each abnormal region: Geometric parameters: area, volume, maximum depth, perimeter, shape factor; Location parameters: 3D coordinates, location of the cultural relic, and distance from key features; Difference parameters: grayscale difference, spectral difference, stress difference, and deformation compared to the baseline data; Change parameters: the amount and rate of change compared to the last verification; S6.5: Anomaly Classification and Severity Assessment Exception type classification: Train a multi-classifier to classify anomalies into the following types: Physical damage: scratches, bumps, cracks, breaks, missing parts; Material changes: corrosion, oxidation, aging, discoloration; Structural abnormalities: loosening, delamination, deformation, hollowing; Human intervention: traces of repair, cleaning, and replacement; Natural changes: normal aging, minor deformations caused by temperature and humidity; Severity grading assessment: A weighted scoring method was used to comprehensively assess the severity of the anomalies. Level 1 (Critical): The anomaly may cause the artifact to break, collapse, or suffer permanent damage, such as cracks in the main structure or extensive corrosion. Level 2 (Severe): The abnormality has a significant impact on the value of the cultural relic and requires immediate repair, such as damage to important decorative patterns or deep cracks; Level 3 (General): The abnormality has a certain impact on the appearance of the cultural relic and requires repair, such as surface scratches or local stains. Level 4 (Minor): The abnormality does not affect the value and safety of the cultural relic. Only enhanced monitoring is required, such as slight discoloration or minor scratches. Distinguishing between natural changes and human-caused damage A natural aging model for cultural relics is established based on historical data to calculate whether the abnormal rate of change conforms to the laws of natural aging. By combining abnormal morphological characteristics, we can distinguish between natural aging (uniform and slow) and human-caused damage (sudden and irregular). S6.6: Automatically generate verification reports The system automatically generates a standardized digital verification report, which includes the following: Basic information: artifact number, name, material, era, verification time, and verification personnel; Overview of comparison results: global similarity score, similarity score for each modality, and verification conclusion (normal / abnormal); Anomaly Details: 3D visualization annotations of the anomaly areas (highlighted directly on the digital twin model of the cultural relic); type, location, quantification parameters, and severity of each anomaly; magnified views and comparison images of the anomaly areas (baseline vs. real-time); Trend analysis: Compare the results with previous verifications to analyze the development trend of anomalies; Handling suggestions: Based on the anomaly type and severity, automatically generate targeted handling suggestions, such as: Normal: Continue routine monitoring Minor abnormalities: Increase monitoring frequency General anomaly: Arrange a repair plan. Serious / Fatal Abnormality: Immediately cease the exhibition and take emergency protective measures. Report Export: Supports export in PDF and Word formats, and generates a blockchain evidence hash to ensure the report is tamper-proof; S6.7: Digital Twin Updates and Data Accumulation Automatic data updates: The real-time multimodal data, comparison results, anomaly reports, and processing records of this verification will be automatically uploaded to the digital twin management platform; Digital twin iteration: Update the status data of the digital twin of cultural relics and mark abnormal areas; Generate a new version number, retain all historical versions, and support retrospective viewing; Update the health status assessment and risk level of cultural relics; Model self-optimization: The labeled data from this verification was added to the training set to incrementally fine-tune the Siamese network model; Regularly update the anomaly classifier and severity assessment model based on new data; Continuously improve the system's comparison accuracy and anomaly detection capabilities.

[0015] The beneficial effects of this invention are: Breaking through the limitations of traditional methods that can only collect surface information, it can acquire deep physical characteristics of cultural relics such as subsurface structure, internal stress, and micro-deformation under radiation-free and non-contact conditions, comprehensively reflecting the true state of the cultural relics and providing a reliable data foundation for precise protection. By adopting a hybrid modeling method, the modeling speed and reconstruction accuracy are balanced, which significantly shortens the 3D reconstruction cycle of cultural relics and obtains high-fidelity, detailed 3D models to meet the needs of rapid digital archiving of large-scale cultural relics. By generating a unique identifier based on the inherent physical characteristics of the cultural relic and binding it with an invisible quantum marker, the risk of label forgery and replacement of the relic is eliminated from the physical source, thus achieving lifelong reliable traceability of the cultural relic's identity. Establishing digital twins of cultural relics, centrally integrating multimodal data, transfer records, and results of previous tests, forms a dynamic and traceable full lifecycle archive that supports visualization, historical retrospection, and trend analysis. Based on historical data and AI algorithms, it can predict potential damage to cultural relics and provide early warnings, promoting the shift of cultural relic protection from passive restoration to proactive prevention and extending the preservation life of cultural relics. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the device structure of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention.

[0018] In the attached diagram, the structural names represented by each number are as follows: 1- Fully enclosed constant temperature and humidity data acquisition chamber; 2- Upper platform; 3- Lower platform; 4- Multimodal sensor group. Detailed Implementation

[0019] 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.

[0020] Example 1 A multimodal modeling, tracing, and verification device for cultural relics includes: The fully enclosed temperature and humidity data acquisition chamber is designed with a fully enclosed light-shielding structure and a built-in temperature and humidity control system to provide a closed temperature and humidity-controlled environment for data acquisition and modeling of cultural relics. In this embodiment, the constant temperature and humidity control system includes: Temperature sensor used to monitor the temperature in a fully enclosed constant temperature and humidity acquisition chamber in real time; A humidity sensor is used to monitor the humidity in a fully enclosed, temperature- and humidity-controlled collection chamber in real time. The PLC, temperature sensor, and humidity sensor are connected to the PLC's input terminal to receive temperature and humidity data. The PLC's memory has preset constant temperature and humidity thresholds in its execution program. After processing the data, the PLC compares and analyzes the data with the thresholds and issues control commands based on the comparison results. The temperature control device is used to receive control commands from the PLC, respond to the temperature adjustment, and maintain a constant temperature inside the chamber; in this embodiment, the temperature control device can be set as an air conditioner. The humidity control device is used to receive control commands from the PLC, respond to them, and adjust the humidity to maintain constant humidity in the chamber. In this embodiment, the humidity control device can be configured as a combination of a humidifier and a dehumidifying exhaust fan. The fully enclosed temperature and humidity-controlled collection chamber is equipped with an inert gas protection device to create a nitrogen-filled environment inside the chamber, thus protecting the cultural relics and ensuring their stable condition. The double-layer rotating platform consists of an upper platform and a lower platform; the upper platform is designed with transparent material, such as transparent tempered glass or transparent acrylic sheet. The lower platform is located directly below the upper platform, and the upper and lower platforms are set as a coaxial structure. The upper and lower platforms use two independent rotary drive motors. The upper and lower platforms can rotate independently under the control of the drive motors, or the two sets of drive motors can rotate synchronously with the upper and lower platforms at the same frequency. Cultural relics are placed on the upper platform. For cultural relics such as calligraphy and paintings, a scroll hanging rack can be set on the upper platform to assist in hanging. The multimodal sensor array is set up in two regions. The first region includes several multimodal sensor arrays, evenly spaced on the lower platform, used to collect data features from the bottom of the artifact. The second region also includes several multimodal sensor arrays, evenly spaced on the inner wall and top of the fully enclosed temperature and humidity control chamber, used to collect data features from the circumference and top of the artifact. The multimodal sensor arrays in the first and second regions are used to collect 360° omnidirectional data features from the artifact. The multimodal sensor group in this embodiment includes: Terahertz imagers are used to collect the subsurface structure and material distribution characteristics of cultural relics. They emit terahertz waves of 0.1~3THz, which can penetrate 0~5mm into the surface of cultural relics to obtain subsurface information that is difficult to accurately identify with the naked eye and X-rays, such as internal interlayers, cavities, dark cracks, repair traces, and different material layers. They emit no ionizing radiation and are safe and non-destructive to all cultural relics. Polarized light cameras are used to collect data on the stress distribution and material anisotropy of cultural relics; capture reflected light with different polarization directions to identify areas of stress concentration accumulated inside the relics, which may be the origin of potential cracks; distinguish between the original material and the restoration material, as the polarization characteristics produced by the two are significantly different; and detect surface microcracks and material uniformity. Laser speckle interferometer is used to collect nanoscale micro-deformation and vibration characteristics of cultural relics; by measuring the changes in speckle pattern generated by laser irradiation, nanoscale surface micro-deformation can be measured, the trend of loosening, delamination and fine cracking of cultural relic structure can be detected, and the overall structural stability of cultural relic can be evaluated. White light supplementary lighting and cameras are used to collect basic visible light features of cultural relic surfaces; they provide uniform, shadow-free white light illumination, which, together with the imaging system, acquires appearance information such as texture, color, shape, scratches, and stains on the surface of cultural relics, and is the basis for 3D modeling and surface appearance inspection. The quantum dot marking and reading / writing component includes a quantum dot inkjet printer and a fluorescence excitation detection device; the quantum dot inkjet printer is used to make invisible marks on cultural relics; the fluorescence excitation detection device is used to read the marking information. The main control and data processing components include a built-in processor and memory. The memory contains a computer-executable program, and the processor executes the computer-executable program to complete multimodal data processing, cultural relic modeling, tracing, and verification.

[0021] Example 2 A multimodal modeling, provenance tracing, and verification method for cultural relics includes: S1: Multimodal data acquisition of cultural relics. The cultural relics are placed on the upper platform of a double-layer rotating platform in a fully enclosed constant temperature and humidity acquisition chamber, maintaining a constant temperature, humidity and inert gas environment inside the chamber; at the same time, the multimodal sensor groups in two areas are activated; the upper platform is started to rotate, and the dual-area multimodal sensor groups synchronously acquire terahertz imaging data, polarized light imaging data, laser speckle interferometry data and white light image data of the entire cultural relics. S2: 3D modeling of cultural relics. A hybrid modeling algorithm combining 3D Gaussian splashing (3DGS) and DUSt3R is used to generate a high-precision 3D model of the cultural relics during the multimodal feature data acquisition process of S1. The 3D modeling method in this embodiment is as follows: S2.1: White light image acquisition, based on white light supplementation in dual regions and the camera acquiring omnidirectional images of the cultural relic at 15° intervals, to acquire at least 24 high-resolution images of the cultural relic; and to perform preprocessing on the images including removing overexposed, underexposed and blurred images, lens distortion correction and white balance calibration, and extracting image feature points (SIFT). S2.2: Fast global initialization of DUSt3R. The 24 unordered images obtained in S2.1 are input into the pre-trained DUSt3R model. The model automatically matches the correspondence between images through the self-attention mechanism and outputs: the rotation matrix R and translation vector t of each image, the camera extrinsic parameters, the globally consistent sparse point cloud, and the depth confidence of each point. Using the known rotation angle and center position of the rotating platform, the point cloud output by DUSt3R is scaled to solve the problem of scale uncertainty in DUSt3R reconstruction; the origin of the coordinate system is aligned with the center of the rotating platform, and the Z-axis is vertically upward to ensure that the model size is consistent with the actual size of the cultural relic, with an error of <0.1mm; Point cloud filtering and optimization: remove outliers with depth confidence below 0.8; remove background point cloud, retaining only the cultural relic area; generate a uniformly distributed initial point cloud to provide high-quality input for 3DGS; S2.3: 3DGS Dense Reconstruction and Parameter Optimization, 3D Gaussian Sphere Initialization: The sparse point cloud generated by DUSt3R is used as the center position of the initial Gaussian sphere; the initial covariance of each Gaussian sphere is set to isotropic, with a size of 1 / 3 of the average distance between adjacent points; the initial color is obtained by interpolating the pixel values ​​of the corresponding image; Differentiable rendering and iterative optimization: Using the differentiable rendering pipeline of 3DGS, a 3D Gaussian sphere is projected onto a 2D image plane to generate a rendered image; the mean square error (MSE) loss between the rendered image and the original acquired image is calculated; the position, covariance, color, and transparency parameters of all Gaussian spheres are iteratively optimized through backpropagation; the Gaussian spheres are automatically split and merged during the optimization process: when the anisotropy of a Gaussian sphere exceeds a threshold, it is split into two Gaussian spheres; when two Gaussian spheres are too close and have similar parameters, they are merged; for transparent cultural relics such as jade and glass, a transmittance parameter is introduced into the 3DGS optimization to accurately simulate the light transmission effect; Enhanced and optimized artifact details: For key detail areas such as inscriptions and patterns on artifacts, the Gaussian sphere density in these areas is automatically increased by 2 to 3 times; Perceptual Loss is introduced to improve the texture details and realism of the model; Iterative optimization is performed 1000 to 2000 times until the loss function converges; S2.4: Model post-processing and accuracy verification, mesh generation: Generate a traditional triangular mesh model, and extract the mesh from the 3D Gaussian sphere using the Poisson surface reconstruction algorithm; perform hole filling and smoothing processing, and remove noise and burrs; Texture mapping optimization: Map the original high-resolution image onto the surface of the 3D model; use the Poisson fusion algorithm to eliminate lighting differences and seams between different images; Accuracy verification: Use a high-precision laser rangefinder to measure the key dimensions of the cultural relic (such as height and diameter); compare them with the corresponding dimensions of the 3D model to ensure an overall accuracy of ±0.05mm; perform local accuracy verification on detailed areas such as inscriptions and patterns to ensure that details are clearly distinguishable; S3: Generate physical feature code, extract the inherent features of the cultural relic, including at least terahertz feature spectrum, polarization stress fingerprint, and speckle interference features, and generate a unique physical feature code for the cultural relic. The process of generating physical signature codes is as follows: S3.1: Extraction of intrinsic features of different modes 1) Terahertz characteristic spectrum extraction: extracting the molecular vibrational characteristics and subsurface structure distribution features of the cultural relic material; Data preprocessing: The three-dimensional data cubes acquired by the terahertz imager are subjected to the following: temporal domain denoising; reflectivity normalization to eliminate the influence of light source intensity fluctuations; and background subtraction to remove background signals from the acquisition chamber and platform. Spectral feature extraction: For the terahertz spectral curve of each spatial pixel, extract 12 key feature parameters such as peak position, valley position, full width at half maximum (FWHM), slope, and inflection point; generate a 211-dimensional original spectral feature vector, and use PCA to reduce the dimensionality to 32 dimensions, retaining 99.5% of the information; Spatial distribution feature extraction: Material clustering is performed based on spectral features to generate a material distribution map of the artifact surface; statistical features such as the number of material regions, area ratio, boundary shape, and relative position are extracted. Subsurface structure feature extraction: Extract the terahertz reflection intensity distribution at different depths (0.5mm, 1mm, 2mm, 3mm, 5mm); calculate the texture entropy, contrast, and correlation of each depth layer to generate 15-dimensional subsurface structure features; Terahertz feature vector: The above features are concatenated to generate a 63-dimensional terahertz feature vector; 2) Polarization stress fingerprint extraction: Extracting the internal stress distribution and material anisotropy characteristics of cultural relics; Data preprocessing: The images at four polarization angles (0°, 45°, 90°, and 135°) acquired by the polarization camera are subjected to the following: lens distortion correction, illumination non-uniformity correction, and Stokes parameter calculation (S0, S1, S2). Polarization parameter calculation: Calculate the degree of polarization (DOP) and angle of polarization (AOP) for each pixel; generate polarization degree map and polarization angle map; Stress distribution feature extraction: Based on the linear relationship between polarization degree and stress, the stress distribution map of the artifact surface is obtained by inversion; the maximum, minimum, average, and standard deviation of the stress distribution are extracted; the watershed algorithm is used to identify stress concentration areas and extract their location, area, shape, stress gradient and other features. Material anisotropy feature extraction: Calculate the spatial distribution histogram of polarization angles (36 bins); extract features such as anisotropy index and principal polarization direction; Polarization stress fingerprint vector: The above features are concatenated to generate a 62-dimensional polarization stress feature vector; 3) Laser speckle interferometry feature extraction to extract the nanoscale micro-morphology and structural vibration characteristics of the artifact surface; Data preprocessing: The speckle image sequence acquired by the laser speckle interferometer is subjected to the following processes: mean filtering for noise reduction, contrast enhancement, and phase unwrapping. Surface micromorphology feature extraction: Calculate statistical features such as contrast, correlation coefficient, entropy, and energy of speckle patterns; extract surface roughness parameters (Ra, Rq, Rz); generate surface height distribution histogram (32 bins); Vibration characteristic feature extraction: Apply a small sinusoidal excitation (amplitude <1μm) to the cultural relic and collect speckle interferograms at different frequencies; obtain the natural frequency, damping ratio, mode shape and other modal parameters of the cultural relic through Fourier transform; extract the nodal position, amplitude distribution and other features of each mode shape; Speckle interference feature vector: The above features are concatenated to generate a 60-dimensional speckle interference feature vector; S3.2: Cross-modal feature fusion and normalization 1) Feature vector concatenation: The terahertz (63-dimensional), polarization stress (62-dimensional), and laser speckle (60-dimensional) feature vectors are concatenated to generate a 185-dimensional original fused feature vector; 2) Adaptive weighted fusion: Automatically adjusts the weights of each mode according to different types of cultural relics. General weights: terahertz 35%, polarization stress 30%, laser speckle 35%; Bronze artifacts: Terahertz 40%, polarization stress 30%, laser speckle 30%; Ceramics: Terahertz 30%, polarization stress 35%, laser speckle 35%; Jade: Terahertz 35%, polarization stress 25%, laser speckle 40%; 3) Normalization: The Z-score normalization method is used to normalize each dimension of the fused feature vector to a distribution with a mean of 0 and a variance of 1; 4) Dimensional unification: The 185-dimensional feature vector is mapped to a 256-dimensional unified feature space through a fully connected layer, generating a 256-dimensional fused feature vector; S3.3: Robust Physical Feature Generation 1) Robust hash encoding: The SM3 hash algorithm (Chinese national cryptographic standard) is combined with robust hashing technology to convert the 256-dimensional fused feature vector into a 256-bit binary physical feature code; 2) Enhanced stability: A tolerance range of ±5% is set for each dimension of the feature vector to eliminate minor changes caused by environmental factors such as temperature and humidity; a voting mechanism is adopted to perform majority voting on the feature vectors collected multiple times to generate the final stable physical feature code; 3) Uniqueness verification: Calculate the Hamming distance between the physical feature codes of different cultural relics to ensure that the Hamming distance is ≥128 bits (uniqueness probability >99.9999%); collect data from the same cultural relic multiple times to ensure that the Hamming distance of the physical feature codes is ≤8 bits (stability >99.99%). 4) Feature code storage and association: The physical feature code is encrypted and bound to the digital twin ID of the cultural relic; it is stored in the digital twin management platform of the cultural relic and the quantum dot invisible tag.

[0022] S4: Quantum dot tag encryption and binding. Quantum dot inkjet printing equipment is used to print invisible quantum dot tags in concealed locations on the artifact, such as the bottom or inner side. The physical feature code generated in S3 is then encrypted and bound to the quantum dot tag ID. The specific implementation method is as follows: S4.1: Cultural relic safety grade quantum dot ink and dedicated printing equipment 1) Quantum dot ink preparation: Core-shell structured near-infrared quantum dots are used, with an emission wavelength of 800~900nm. The outer shell is coated with silica, and the core is CdSe / ZnS; 0.1% of polyethylene glycol, a cultural relic preservation agent, is added. 2) Retrofitting of dedicated quantum dot inkjet printing equipment It employs a piezoelectric micron-level printhead with a minimum droplet volume of 1pL and a printing accuracy of ±5μm; it is equipped with a three-axis precision motion platform with a positioning accuracy of ±2μm; it integrates a high-resolution industrial camera to automatically identify the hidden locations of cultural relics and plan the printing path; it has a built-in national cryptographic hardware encryption chip (SM2 / SM3 / SM4), and all encryption calculations are completed within the chip without going through the host operating system; the printhead maintains a distance of 1~2mm from the surface of the cultural relic, enabling completely contactless printing. S4.2: Quantum Dot Invisibility Mark Printing 1) Printing location selection and planning: Prioritize areas that do not affect the display and research of cultural relics, such as the bottom edge, inner recesses, gaps in inscriptions, and the back of the base; avoid areas such as the front of the cultural relic, areas with dense patterns, areas prone to wear, and areas with fragile materials. The system automatically plans the printing path to ensure that the mark size is controlled within 2mm×2mm and does not exceed the hidden area; after confirming the printing position, it adjusts the orientation and angle of the print head and the distance from the cultural relic; 2) High-precision invisible tag printing: The device generates a 128-bit globally unique quantum dot tag ID, including: a 64-bit device number + timestamp to ensure uniqueness; a 32-bit random number to prevent prediction; and a 32-bit CRC checksum for reading and verification. The tag ID is encoded into a QR code format, using the highest level of error correction, H-level; it can still be read correctly even if the tag is 30% damaged. The print head sprays quantum dot ink point by point onto the surface of the artifact according to the planned path, forming a micron-level QR code pattern; after printing, it is dried with hot air. 3) Print quality verification: The device automatically switches to fluorescence detection mode and emits 850nm near-infrared light to excite the quantum dot markers; the camera acquires fluorescence images, automatically identifies and decodes the marker ID; verifies whether the decoded ID is consistent with the original ID, and if they are inconsistent, reprints; takes fluorescence photos and natural light photos of the marker location and stores them in the digital twin archive of the cultural relics. S4.3: Hardware-level encryption binding of physical signature and quantum dot ID 1) Encryption binding condition settings: All encryption operations are completed in the device's built-in hardware encryption chip without connecting to any network; a hash algorithm is used to ensure that the original physical signature cannot be deduced from the encryption result; the quantum dot ID contains the hash value of the physical signature, and the physical signature contains the hash value of the quantum dot ID; any modification to the binding relationship will result in verification failure. 2) Binding steps: a. Data input: Input the 256-bit physical feature code of the cultural relic generated by S3 and the 128-bit quantum dot mark ID printed by S4 into the hardware encryption chip; b. First-level hash operation: The 256-bit physical feature code is hashed using the national cryptographic SM3 algorithm to generate a 256-bit physical feature hash value H1; the 128-bit quantum dot ID is hashed using the national cryptographic SM3 algorithm to generate a 256-bit ID hash value H2. c. Two-way binding operation: Concatenate H1 with the quantum dot ID to generate 384-bit data D1; concatenate H2 with the physical feature code to generate 512-bit data D2; perform SM3 hash operation on D1 and D2 respectively to generate binding credentials H3 (256 bits) and H4 (256 bits); d. Encrypted storage: Using the national cryptographic SM4 algorithm, H3 and H4 are encrypted using the unique master key built into the device; the encrypted binding certificate is written into the encrypted storage area marked by quantum dots; the quantum dot ID, H1, H2, H3, H4 and binding timestamp are encrypted and stored in the cultural relic digital twin management platform. e. Binding Verification: Read the encrypted binding credential from the quantum dot tag, decrypt it to obtain H3 and H4; recalculate H1 and H2, and verify whether they are consistent with the stored values; if the verification is successful, the binding is completed and a binding report is generated. After the quantum dot stealth tags are printed and bound to physical features, the quantum dot stealth tags need to be read and the binding relationship verified. 1) Quantum dot tag reading Using a dedicated fluorescence excitation detection device, 850nm near-infrared light is emitted to illuminate the marked location of the cultural relic; the device collects the 900nm fluorescence signal emitted by the quantum dots, decodes it to obtain the quantum dot ID and the encrypted binding certificate; decrypts the binding certificate to obtain H3 and H4; 2) Binding relationship verification Retrieve the physical feature hash value H1 corresponding to the quantum dot ID from the digital twin platform; re-collect the current physical features of the cultural relic, generate a real-time physical feature code, and calculate its real-time hash value H1'; verify whether H1 and H1' are consistent, and consider them consistent if the Hamming distance is ≤8 bits; if the verification is successful, the identity of the cultural relic and the binding relationship are confirmed to be valid, otherwise it is judged as abnormal; 3) Handling Abnormal Situations a. Mark reading failed: Adjust the excitation light angle and intensity and try reading again; if it still fails, you can search the digital twin platform through other features of the cultural relic to find the corresponding record. After authorization by two or more administrators, you can reprint the mark near the original location and update the binding relationship. b. Damaged tag: If the tag is partially damaged, try to recover it using the error correction capability of the QR code; if it is completely damaged, reprint the tag after approval, mark the old tag ID as "invalid" and keep the record permanently; c. Updates after cultural relic restoration: After the cultural relic is restored, physical characteristics are re-collected to generate new physical characteristic codes; the binding relationship is updated, while the historical physical characteristic codes and binding records are retained to form a complete cultural relic status change archive.

[0023] S5: Construction of digital twins of cultural relics: All data collected from cultural relics, 3D models, feature codes, and marking information are integrated and entered into the digital twin management platform to complete the initial archiving; all life-cycle operations of cultural relics, including warehousing, warehousing, exhibition, and restoration, are recorded on the digital twin management platform; the digital twin management platform also predicts the development trend of cultural relic diseases based on historical data and AI algorithms and issues early warnings. The specific methods used by the aforementioned digital twin management platform to predict the development trend of cultural relic diseases are as follows: S5.1: Construction of a Multimodal Temporal Disease Database 1) Core data collection and integration Multimodal detection data: Terahertz subsurface crack data, polarization stress distribution data, laser speckle micro-deformation data, and surface texture data collected in each iteration are aligned by timestamps to form a time sequence; Environmental monitoring data: minute-level continuous monitoring data of environmental conditions such as temperature and humidity, light intensity, carbon dioxide concentration, and vibration acceleration in the storage / exhibition environment of cultural relics; Manually recorded data: Structured data such as cultural relic restoration records, maintenance records, exhibition records, and transportation records; Disease labeling data: Label data including the type, location, severity, and stage of development of diseases, as marked by cultural relics experts; 2) Data preprocessing and standardization Interpolation completion, outlier removal, and smoothing are performed on time-series data; A unified data format and coordinate system were used to map all disease characteristics to the three-dimensional spatial position of the digital twin of the cultural relic; A four-dimensional data index of "cultural relic-part-disease-time" is constructed to support fast retrieval by part and disease type; S5.2: Disease-related feature extraction, as shown in the table below. Disease types Core related features Feature Source Crack propagation Subsurface crack length / depth / direction, stress concentration value, micro-deformation rate Terahertz imager, polarized light camera, laser speckle interferometer Material aging Terahertz spectral characteristic shift, surface roughness variation, color variation Terahertz imager, laser speckle interferometer, white light camera loose structure Natural frequency shift, damping ratio change, vibration mode distortion Laser speckle interferometer Corrosion disease Terahertz reflectivity variation, surface stress distribution variation, and corrosion area Terahertz imager, polarized light camera 1) For each diseased part, extract the temporal feature sequence for more than 6 consecutive months; 2) Calculate derived features such as the rate of change, acceleration, and fluctuation amplitude of the characteristic; 3) Use mutual information to screen key features with a correlation > 0.7 with disease development, and form a disease feature vector; S5.3: Hybrid prediction model construction, adopting a hybrid architecture of "data-driven AI model + physical mechanism model" to balance prediction accuracy and interpretability; 1) Basic Model of Physical Information Neural Network (PINN) The physical properties of cultural relics materials, such as mechanical, thermal, and chemical properties, are embedded into the neural network as prior knowledge; the loss function consists of two parts: data loss (error between predicted and actual observed values) + physical loss (penalty for violating physical laws); It solves the problems of traditional deep learning models requiring a large amount of labeled data and having poor generalization ability, and can still maintain high prediction accuracy under small sample conditions; 2) Set up dedicated prediction models for different diseases a) Crack propagation prediction model Input: Crack history length / depth sequence, stress concentration value sequence, ambient temperature and humidity sequence; Model: PINN combined with Extended Finite Element Method (XFEM); Output: Predictions for crack length, depth, and propagation direction for the next 3 / 6 / 12 months; Key technology: The Paris formula in fracture mechanics is embedded into a neural network as a physical constraint to simulate the crack propagation law under cyclic stress. b. Material aging prediction model Input: Terahertz spectral characteristic sequence, cumulative illumination duration, and cumulative temperature and humidity fluctuation values; Model: Temporal Transformer + Material Aging Dynamics Equations; Output: Predictions of material aging degree and mechanical property degradation rate over the next 1 / 3 / 5 years; Key technologies: Simulating the effect of temperature on aging rate based on the Arrhenius equation, and simulating the degradation effect of light on materials based on the Beer-Lambert law; c. Structural loosening prediction model Inputs: natural frequency sequence, damping ratio sequence, vibration acceleration sequence; Model: LSTM + Modal Analysis Theory; Output: Structural loosening risk level and predicted loosening location for the next 1 / 3 / 6 months; Key technology: Inverting the attenuation law of structural connection stiffness through changes in vibration mode parameters; 3) Multi-model fusion and uncertainty quantification The weighted average method is used to fuse the prediction results of multiple models, and the weights are dynamically adjusted according to the historical prediction accuracy of each model. The Monte Carlo dropout method is introduced to quantify the uncertainty of the prediction results and to give the confidence interval of the prediction values; When the forecast uncertainty exceeds 30%, a manual review process is automatically triggered. S5.4: Tiered Early Warning and Decision Support 1) Risk level classification Based on the predicted severity of the damage and the value of the cultural relics, the early warning is divided into four levels: Level 1 Warning (Red): Fatal damage (such as fracture or collapse) is predicted to occur within 3 months, and protective measures must be taken immediately; Level II Warning (Orange): Severe diseases are predicted to occur within 6 months, requiring special inspection and repair. Level 3 Warning (Yellow): Common diseases are predicted to occur within 12 months, and monitoring frequency needs to be increased; Level 4 warning (blue): There is a potential risk of disease; routine monitoring should be maintained. 2) Early warning triggering mechanism When the predicted disease parameters exceed the threshold of the corresponding level, an early warning will be automatically triggered; Taking into account the historical, artistic, scientific, and fragile value of cultural relics, the early warning threshold is dynamically adjusted. For warning signals that appear repeatedly in the same location, the warning level will be automatically upgraded. 3) Intelligent decision support Based on the knowledge base, targeted protection measures are automatically generated, such as adjusting environmental parameters, limiting exhibition time, and carrying out preventive repairs. The predicted location, development trend, and impact range of damage are displayed in three dimensions on the digital twin of cultural relics; Generate early warning reports, including disease details, prediction results, risk assessments, and protection recommendations, and push them to relevant management personnel; S5.5: Continuous Iterative Optimization of the Model After each collection of artifact condition data and restoration of damage, new data is added to the training set to incrementally update the model; The model's predictive accuracy should be evaluated at least quarterly, and the model parameters should be adjusted based on the evaluation results. An expert feedback mechanism has been established, allowing cultural relics experts to annotate and correct the prediction results, thereby continuously improving the model's performance.

[0024] S6: Intelligent verification of cultural relic return and status. Scanning the quantum dot invisible markers on cultural relics, quickly retrieving the corresponding digital twin and historical benchmark data; multimodal sensor groups automatically collect current multimodal data of cultural relics, generating real-time 3D models and physical feature codes; using twin network AI algorithms, comparing real-time data with historical benchmark data in multiple dimensions; automatically generating verification reports, quantifying and marking abnormal areas and severity; updating the verification data to the digital twin platform, improving the full life cycle record of cultural relics; The specific process for verifying the return of cultural relics is as follows: S6.1: Verification Initialization and Data Preparation Rapid identification: Scan the quantum dot invisible markers on cultural relics using a fluorescence excitation detection device, and decode them to obtain a 128-bit marker ID and encrypted binding credential; Baseline data retrieval: Quickly retrieve the artifact's data from the digital twin platform using its tagged ID. The initial multimodal reference data package includes: terahertz cube, polarization stress map, speckle interferogram, and white light 3D model; Historical data and anomaly records from each verification; At least metadata including the material, age, and preservation environment of the artifact; Real-time data acquisition: The integrated device is activated to automatically acquire the current multimodal data of the cultural relics, and the acquisition parameters are completely consistent with the initial archiving. S6.2: Data Preprocessing and 3D Spatial Alignment Preprocessing for real-time acquired single-modal data: Terahertz data: Wavelet denoising, reflectivity normalization, and background subtraction were performed; Polarization data: Complete Stokes parameter calculation, polarization degree / polarization angle diagram generation, and illumination correction; Speckle data: Phase unwrapping, micro-deformation field calculation, and noise filtering were performed; White light data: Complete distortion correction, white balance, and contrast enhancement; Global 3D Spatial Alignment An improved ICP algorithm was used to register the real-time acquired white light point cloud with the reference 3D model, with a registration accuracy of ±0.02mm; The rotation matrix and translation vector obtained from the registration are applied to all modal data to make the real-time data and the reference data completely aligned in the same three-dimensional coordinate system. For each three-dimensional vertex on the surface of the cultural relic, establish a one-to-one correspondence between real-time data and benchmark data; S6.3: Multimodal Siamese Network Feature Extraction and Fusion. A two-branch symmetric Siamese network architecture is adopted, with the two branches sharing weights. Real-time data and benchmark data are input respectively, and the corresponding multimodal fusion feature vectors are output. A modality-specific feature extraction branch is designed with independent feature extraction networks tailored to the characteristics of different modalities; as shown in the table below: Modal type Network Structure Output feature dimension Extracted core features White Light 3D Model PointNet++ 256 dimensions Geometric shape, surface texture, contour features Terahertz Data Cube 3D-CNN 256 dimensions Subsurface structure, material distribution, internal defects Polarization stress diagram U-Net + Attention 256 dimensions Stress distribution, material anisotropy, stress concentration Laser speckle data Vision Transformer 256 dimensions Micro-deformation field, vibration characteristics, surface micro-morphology Employing a triple attention mechanism to dynamically fuse multimodal features: Spatial attention: Automatically focuses on key feature areas of cultural relics, such as inscriptions, patterns, and seams, while suppressing background interference; Channel attention: Automatically selects the feature channels that are most valuable to the current alignment task; Modal attention: Automatically adjusts the weights of each modality based on the type of artifact, for example: Jade: Laser speckle (40%) > Terahertz (30%) > Polarized (20%) > White light (10%) Bronze artifacts: Terahertz (35%) > Polarized (30%) > Laser speckle (20%) > White light (15%) Painting and calligraphy: White light (40%) > Terahertz (30%) > Polarized light (20%) > Laser speckle (10%) The final result is a 1024-dimensional global fusion feature vector; Similarity calculation: Calculate the cosine similarity between the global fused feature vectors of real-time data and benchmark data; Simultaneously, the similarity of independent feature vectors for each modality is calculated to obtain a multi-dimensional similarity score; Set a dynamic threshold: It is automatically adjusted based on the material of the cultural relic, the storage environment, and the last verification time, and is usually 0.95~0.98; If the global similarity is greater than or equal to the threshold, it is considered normal; if... <阈值，进入异常检测流程；S6.4: Anomaly Area Location and Quantitative Analysis Anomaly heatmap generation: Calculate the differences between real-time data and baseline data on feature maps at each level; Upsampling restores the feature differences to the original image space, generating an anomaly heatmap. The intensity of the colors in a heatmap indicates the degree of abnormality: red indicates severe abnormality, yellow indicates moderate abnormality, and blue indicates normal. Abnormal region segmentation and extraction: Threshold segmentation combined with morphological operations is used to extract continuous anomalous regions from anomaly heatmaps; Each abnormal area is reconstructed in three dimensions and mapped to the corresponding position in the digital twin of the cultural relic; Abnormal parameter quantification calculation: The following quantization parameters are automatically calculated for each abnormal region: Geometric parameters: area, volume, maximum depth, perimeter, shape factor; Location parameters: 3D coordinates, location of the cultural relic, and distance from key features; Difference parameters: grayscale difference, spectral difference, stress difference, and deformation compared to the baseline data; Change parameters: the amount and rate of change compared to the last verification; S6.5: Anomaly Classification and Severity Assessment Exception type classification: Train a multi-classifier to classify anomalies into the following types: Physical damage: scratches, bumps, cracks, breaks, missing parts; Material changes: corrosion, oxidation, aging, discoloration; Structural abnormalities: loosening, delamination, deformation, hollowing; Human intervention: traces of repair, cleaning, and replacement; Natural changes: normal aging, minor deformations caused by temperature and humidity; Severity grading assessment: A weighted scoring method was used to comprehensively assess the severity of the anomalies. Level 1 (Critical): The anomaly may cause the artifact to break, collapse, or suffer permanent damage, such as cracks in the main structure or extensive corrosion. Level 2 (Severe): The abnormality has a significant impact on the value of the cultural relic and requires immediate repair, such as damage to important decorative patterns or deep cracks; Level 3 (General): The abnormality has a certain impact on the appearance of the cultural relic and requires repair, such as surface scratches or local stains. Level 4 (Minor): The abnormality does not affect the value and safety of the cultural relic. Only enhanced monitoring is required, such as slight discoloration or minor scratches. Distinguishing between natural changes and human-caused damage A natural aging model for cultural relics is established based on historical data to calculate whether the abnormal rate of change conforms to the laws of natural aging. By combining abnormal morphological characteristics, we can distinguish between natural aging (uniform and slow) and human-caused damage (sudden and irregular). S6.6: Automatically generate verification reports The system automatically generates a standardized digital verification report, which includes the following: Basic information: artifact number, name, material, era, verification time, and verification personnel; Overview of comparison results: global similarity score, similarity score for each modality, and verification conclusion (normal / abnormal); Anomaly Details: 3D visualization annotations of the anomaly areas, highlighted directly on the digital twin model of the cultural relic; type, location, quantification parameters, and severity of each anomaly; magnified views of the anomaly areas and comparisons between baseline and real-time data; Trend analysis: Compare the results with previous verifications to analyze the development trend of anomalies; Handling suggestions: Based on the anomaly type and severity, automatically generate targeted handling suggestions, such as: Normal: Continue routine monitoring; Minor abnormalities: Increase monitoring frequency; General anomalies: Arrange a repair plan; Serious / fatal abnormality: Immediately stop the exhibition and take emergency protective measures; Report Export: Supports export in PDF and Word formats, and generates a blockchain evidence hash to ensure the report is tamper-proof; S6.7: Digital Twin Updates and Data Accumulation Automatic data updates: The real-time multimodal data, comparison results, anomaly reports, and processing records of this verification will be automatically uploaded to the digital twin management platform; Digital twin iteration: Update the status data of the digital twin of cultural relics and mark abnormal areas; Generate a new version number, retain all historical versions, and support retrospective viewing; Update the health status assessment and risk level of cultural relics; Model self-optimization: The labeled data from this verification was added to the training set to incrementally fine-tune the Siamese network model; Regularly update the anomaly classifier and severity assessment model based on new data; Continuously improve the system's comparison accuracy and anomaly detection capabilities.

Claims

1. A multi-modal modeling, tracing, and verifying device for cultural relics, characterized in that, include: The fully enclosed temperature and humidity data acquisition chamber is designed with a fully enclosed light-shielding structure and a built-in temperature and humidity control system to provide a closed temperature and humidity-controlled environment for data acquisition and modeling of cultural relics. A double-layer rotating platform, consisting of an upper platform and a lower platform; The upper platform is designed with a transparent material. The lower platform is located directly below the upper platform, and the upper and lower platforms are coaxial; the upper and lower platforms use two independent rotary drive motors. The multimodal sensor array is configured in at least two regions. The first region includes several sets of the multimodal sensor array, which are evenly spaced on the lower platform. The second region also includes several sets of the multimodal sensor array, which are evenly spaced on the inner wall and top of the fully enclosed constant temperature and humidity acquisition chamber. The multimodal sensor arrays in the first and second regions are used to acquire 360° omnidirectional data features of the cultural relics. A quantum dot tagging and reading / writing component, comprising a quantum dot inkjet device and a fluorescence excitation detection device; The quantum dot inkjet device is used to make invisible marks on cultural relics; the fluorescence excitation detection device is used to read the mark information; The main control and data processing components include a built-in processor and a memory. The memory contains a computer-executable program, and the processor executes the computer-executable program to complete multimodal data processing, cultural relic modeling, tracing, and verification.

2. The multi-modal modeling, provenance, verification device for cultural heritage artifacts of claim 1, wherein, The constant temperature and humidity control system includes: Temperature sensor used to monitor the temperature in a fully enclosed constant temperature and humidity acquisition chamber in real time; A humidity sensor is used to monitor the humidity in a fully enclosed, temperature- and humidity-controlled collection chamber in real time. The PLC is used to receive temperature and humidity data, process, analyze and compare the data, and issue control commands based on the comparison results. Temperature control equipment is used to receive control commands from the PLC, respond to them, and adjust the temperature to maintain a constant temperature inside the chamber. A humidity control device is used to receive control commands from the PLC, respond to adjust the humidity, and maintain constant humidity in the chamber. The fully enclosed constant temperature and humidity collection chamber is equipped with an inert gas protection device to create a nitrogen-filled environment inside the chamber, thus protecting the cultural relics and ensuring their stable condition.

3. The multi-modal modeling, provenance, verification apparatus for cultural heritage artifacts of claim 1, wherein, The multimodal sensor group includes at least: Terahertz imagers are used to acquire information about the subsurface structure and material distribution characteristics of cultural relics. Polarized light camera is used to collect data on the stress distribution and material anisotropy of cultural relics. Laser speckle interferometer is used to collect nanoscale micro-deformation and vibration characteristics of cultural relics; White light supplementary lighting and a camera are used to collect the basic visible light characteristics of the surface of cultural relics.

4. A multimodal modeling, tracing, and verification method for cultural relics, characterized in that, include: S1: Multimodal data acquisition of cultural relics. The cultural relics are placed on the upper platform of a double-layer rotating platform in a fully enclosed constant temperature and humidity acquisition chamber, maintaining a constant temperature, humidity and inert gas environment inside the chamber; at the same time, the multimodal sensor groups in two areas are activated; the upper platform is started to rotate, and the dual-area multimodal sensor groups synchronously acquire terahertz imaging data, polarized light imaging data, laser speckle interferometry data and white light image data of the entire cultural relics. S2: 3D modeling of cultural relics. A hybrid modeling algorithm combining 3D Gaussian splashing and DUSt3R is used to generate a high-precision 3D model of cultural relics during the acquisition of multimodal feature data of cultural relics in S1. S3: Generate physical feature code, extract the inherent features of the cultural relic, including at least terahertz feature spectrum, polarization stress fingerprint, and speckle interference features, and generate a unique physical feature code for the cultural relic. S4: Quantum dot tag encryption binding. Quantum dot inkjet equipment is used to print quantum dot invisible tags in hidden locations on cultural relics. The physical feature code generated in S3 is encrypted and bound to the quantum dot tag ID. S5: Construction of a digital twin of cultural relics: All collected data, 3D models, feature codes, and marking information of cultural relics are integrated and entered into the digital twin management platform to complete the initial archiving; all operations of cultural relics throughout their entire life cycle are recorded on the digital twin management platform; the digital twin management platform also predicts the development trend of cultural relic diseases based on historical data and AI algorithms, and issues early warnings. S6: Intelligent verification of cultural relic return and status. Scan the quantum dot invisible markers on the cultural relic to quickly retrieve the corresponding digital twin and historical benchmark data; the multimodal sensor group automatically collects the current multimodal data of the cultural relic and generates a real-time 3D model and physical feature code; using the twin network AI algorithm, the real-time data and historical benchmark data are compared in multiple dimensions; an automatic verification report is generated, quantifying and marking abnormal areas and their severity; the verification data is updated to the digital twin platform to improve the full life cycle record of the cultural relic.

5. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The specific method for constructing a 3D model of cultural relics using the hybrid modeling algorithm combining 3D Gaussian splashing and DUSt3R is as follows: S2.1: White light image acquisition, based on white light supplementation in dual regions and the camera acquiring omnidirectional images of the cultural relic at 15° intervals, to obtain at least 24 high-resolution images of the cultural relic; and to perform preprocessing on the images including removing overexposed, underexposed and blurred images, lens distortion correction and white balance calibration, and extracting image feature points. S2.2: Fast global initialization of DUSt3R. The 24 unordered images obtained in S2.1 are input into the pre-trained DUSt3R model. The model automatically matches the correspondence between images through the self-attention mechanism and outputs: the rotation matrix R and translation vector t of each image, the camera extrinsic parameters, the globally consistent sparse point cloud, and the depth confidence of each point. Using the known rotation angle and center position of the rotating platform, scale correction is performed on the point cloud output by DUSt3R; the origin of the coordinate system is aligned with the center of the rotating platform, and the Z-axis is vertically upward to ensure that the model size is consistent with the actual size of the cultural relic; Filter and optimize the point cloud: remove outliers with a depth confidence score below 0.8; Remove background point clouds, retaining only the artifact area; generate a uniformly distributed initial point cloud to provide high-quality input for 3DGS; S2.3: 3DGS Dense Reconstruction and Parameter Optimization, 3D Gaussian Sphere Initialization: The sparse point cloud generated by DUSt3R is used as the center position of the initial Gaussian sphere; the initial covariance of each Gaussian sphere is set to isotropic, with a size of 1 / 3 of the average distance between adjacent points; the initial color is obtained by interpolating the pixel values ​​of the corresponding image; Differentiable rendering and iterative optimization: Using the differentiable rendering pipeline of 3DGS, a 3D Gaussian sphere is projected onto a 2D image plane to generate a rendered image; the mean square error loss between the rendered image and the original acquired image is calculated; the position, covariance, color, and transparency parameters of all Gaussian spheres are iteratively optimized through backpropagation; the Gaussian spheres are automatically split and merged during the optimization process: when the anisotropy of a Gaussian sphere exceeds a threshold, it is split into two Gaussian spheres; when two Gaussian spheres are too close and have similar parameters, they are merged; for transparent cultural relics such as jade and glass, a transmittance parameter is introduced in the 3DGS optimization to accurately simulate the light transmission effect; Enhanced and optimized artifact details: For key detail areas of artifacts, the Gaussian sphere density in those areas is automatically increased by 2 to 3 times; perceptual loss is introduced to improve the texture details and realism of the model; iterative optimization is performed 1000 to 2000 times until the loss function converges; S2.4: Model post-processing and accuracy verification, mesh generation: Generate a traditional triangular mesh model, and extract the mesh from the 3D Gaussian sphere using the Poisson surface reconstruction algorithm; perform hole filling and smoothing processing, and remove noise and burrs; Texture mapping optimization: Map the original high-resolution image onto the surface of the 3D model; use the Poisson fusion algorithm to eliminate lighting differences and seams between different images; Accuracy verification: Use a high-precision laser rangefinder to measure the key dimensions of the cultural relic; compare them with the corresponding dimensions of the 3D model to ensure an overall accuracy of ±0.05mm; perform local accuracy verification on detailed areas to ensure that details are clearly distinguishable.

6. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The physical feature code generation process is as follows: S3.1: Extraction of intrinsic features of different modes 1) Terahertz characteristic spectrum extraction: extracting the molecular vibrational characteristics and subsurface structure distribution features of the cultural relic material; Data preprocessing: The three-dimensional data cubes acquired by the terahertz imager are subjected to the following: temporal domain denoising; reflectivity normalization to eliminate the influence of light source intensity fluctuations; and background subtraction to remove background signals from the acquisition chamber and platform. Spectral feature extraction: Extract key feature parameters from the terahertz spectral curve of each spatial pixel; A 211-dimensional original spectral feature vector was generated, and then PCA was used to reduce the dimensionality to 32 dimensions, retaining 99.5% of the information. Spatial distribution feature extraction: Material clustering is performed based on spectral features to generate a material distribution map of the artifact surface; statistical features of material regions are extracted. Subsurface structure feature extraction: Extract the terahertz reflection intensity distribution at different depths; calculate the texture entropy, contrast, and correlation of each depth layer to generate 15-dimensional subsurface structure features; Terahertz feature vector: The above features are concatenated to generate a 63-dimensional terahertz feature vector; 2) Polarization stress fingerprint extraction: Extracting the internal stress distribution and material anisotropy characteristics of cultural relics; Data preprocessing: The images acquired by the polarization camera at four polarization angles (0°, 45°, 90°, and 135°) are subjected to the following: lens distortion correction, illumination non-uniformity correction, and Stokes parameter calculation. Polarization parameter calculation: Calculate the degree of polarization and polarization angle of each pixel; generate polarization degree map and polarization angle map; Stress distribution feature extraction: Based on the linear relationship between polarization degree and stress, the stress distribution map of the artifact surface is obtained by inversion; the maximum, minimum, average, and standard deviation statistical features of the stress distribution are extracted; the watershed algorithm is used to identify stress concentration areas and extract their location, area, shape, and stress gradient features; Material anisotropy feature extraction: Calculate the spatial distribution histogram of polarization angles; extract anisotropy index and principal polarization direction features; Polarization stress fingerprint vector: The above features are concatenated to generate a 62-dimensional polarization stress feature vector; 3) Laser speckle interferometry feature extraction to extract the nanoscale micro-morphology and structural vibration characteristics of the artifact surface; Data preprocessing: The speckle image sequence acquired by the laser speckle interferometer is subjected to the following processes: mean filtering for noise reduction, contrast enhancement, and phase unwrapping. Surface micromorphology feature extraction: Calculate the statistical characteristics of speckle patterns; Extract surface roughness parameters; Generate a histogram of surface height distribution; Vibration characteristic feature extraction: Apply a small sinusoidal excitation to the cultural relic and collect speckle interferograms at different frequencies; obtain the natural frequency, damping ratio, and mode parameters of the cultural relic through Fourier transform; extract the nodal positions and amplitude distribution characteristics of each mode. Speckle interference feature vector: The above features are concatenated to generate a 60-dimensional speckle interference feature vector; S3.2: Cross-modal feature fusion and normalization 1) Feature vector concatenation: The 63-dimensional terahertz, 62-dimensional polarization stress, and 60-dimensional laser speckle feature vectors are concatenated to generate a 185-dimensional original fused feature vector; 2) Adaptive weighted fusion: Automatically adjusts the weights of each mode according to different types of cultural relics. General weights: terahertz 35%, polarization stress 30%, laser speckle 35%; 3) Normalization: The Z-score normalization method is used to normalize each dimension of the fused feature vector to a distribution with a mean of 0 and a variance of 1; 4) Dimensional unification: The 185-dimensional feature vector is mapped to a 256-dimensional unified feature space through a fully connected layer, generating a 256-dimensional fused feature vector; S3.3: Robust Physical Feature Generation 1) Robust hash encoding: The SM3 hash algorithm (Chinese national cryptographic standard) is combined with robust hashing technology to convert the 256-dimensional fused feature vector into a 256-bit binary physical feature code; 2) Enhanced stability: A tolerance range of ±5% is set for each dimension of the feature vector to eliminate minor changes caused by environmental factors; a voting mechanism is adopted to perform majority voting on the feature vectors collected multiple times to generate the final stable physical feature code; 3) Uniqueness verification: Calculate the Hamming distance between the physical feature codes of different cultural relics to ensure that the Hamming distance is ≥128 bits; collect data from the same cultural relic multiple times to ensure that the Hamming distance between the physical feature codes is ≤8 bits; 4) Feature code storage and association: The physical feature code is encrypted and bound to the digital twin ID of the cultural relic; it is stored in the digital twin management platform of the cultural relic and the quantum dot invisible tag.

7. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The specific implementation method for combining the physical feature code with the quantum dot tag ID for encryption binding is as follows: S4.1: Cultural relic safety grade quantum dot ink and dedicated printing equipment 1) Quantum dot ink preparation: Core-shell structured near-infrared quantum dots are used, with an emission wavelength of 800~900nm. The outer shell is coated with silica, and the core is CdSe / ZnS; 0.1% of polyethylene glycol, a cultural relic preservation agent, is added. 2) Retrofitting of dedicated quantum dot inkjet printing equipment It adopts a piezoelectric micron-level printhead with a minimum droplet volume of 1pL and a printing accuracy of ±5μm; it is equipped with a three-axis precision motion platform; it integrates a high-resolution industrial camera to automatically identify the hidden location of the cultural relic and plan the printing path; it has a built-in national cryptographic hardware encryption chip, and all encryption calculations are completed within the chip without going through the host operating system; the printhead maintains a distance of 1~2mm from the surface of the cultural relic, and the entire process is contactless printing. S4.2: Quantum Dot Invisibility Mark Printing 1) Printing location selection and planning: Prioritize areas that do not affect the display and research of cultural relics, such as the bottom edge, inner recesses, gaps in inscriptions, and the back of the base; avoid areas such as the front of the cultural relic, areas with dense patterns, areas prone to wear, and areas with fragile materials. The system automatically plans the printing path to ensure that the mark size is controlled within 2mm×2mm and does not exceed the hidden area; after confirming the printing position, it adjusts the orientation and angle of the print head and the distance from the cultural relic; 2) High-precision invisible tag printing: The device generates a 128-bit globally unique quantum dot tag ID, including: a 64-bit device number + timestamp to ensure uniqueness; a 32-bit random number to prevent prediction; and a 32-bit CRC checksum for reading and verification. The tag ID is encoded into a QR code format, using the highest level of error correction, H-level error correction. The printhead sprays quantum dot ink point by point onto the surface of the artifact according to the planned path, forming a micron-level QR code pattern. After printing, use hot air to dry; 3) Print quality verification: The device automatically switches to fluorescence detection mode and emits 850nm near-infrared light to excite the quantum dot markers; the camera acquires fluorescence images, automatically identifies and decodes the marker ID; verifies whether the decoded ID is consistent with the original ID, and if they are inconsistent, reprints; takes fluorescence photos and natural light photos of the marker location and stores them in the digital twin archive of the cultural relics. S4.3: Hardware-level encryption binding of physical signature and quantum dot ID 1) Encryption binding condition settings: All encryption operations are completed in the device's built-in hardware encryption chip without connecting to any network; a hash algorithm is used to ensure that the original physical signature cannot be deduced from the encryption result; the quantum dot ID contains the hash value of the physical signature, and the physical signature contains the hash value of the quantum dot ID; any modification to the binding relationship will result in verification failure. 2) Binding steps: a. Data input: Input the 256-bit physical feature code of the cultural relic generated by S3 and the 128-bit quantum dot mark ID printed by S4 into the hardware encryption chip; b. First-level hash operation: The 256-bit physical feature code is hashed using the national cryptographic SM3 algorithm to generate a 256-bit physical feature hash value H1; the 128-bit quantum dot ID is hashed using the national cryptographic SM3 algorithm to generate a 256-bit ID hash value H2. c. Two-way binding operation: Concatenate H1 with the quantum dot ID to generate 384-bit data D1; concatenate H2 with the physical feature code to generate 512-bit data D2; perform SM3 hash operation on D1 and D2 respectively to generate binding credentials H3 and H4; d. Encrypted storage: Using the national cryptographic SM4 algorithm, H3 and H4 are encrypted using the unique master key built into the device; the encrypted binding certificate is written into the encrypted storage area marked by quantum dots; the quantum dot ID, H1, H2, H3, H4 and binding timestamp are encrypted and stored in the cultural relic digital twin management platform. e. Binding Verification: Read the encrypted binding credential from the quantum dot marker, decrypt it to obtain H3 and H4; recalculate H1 and H2, and verify whether they are consistent with the stored values; if the verification is successful, the binding is completed and a binding report is generated.

8. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The method for reading and verifying the binding relationship of quantum dot invisible tags is as follows: 1) Quantum dot tag reading Using a dedicated fluorescence excitation detection device, 850nm near-infrared light is emitted to illuminate the marked location of the cultural relic; the device collects the 900nm fluorescence signal emitted by the quantum dots, decodes it to obtain the quantum dot ID and the encrypted binding certificate; decrypts the binding certificate to obtain H3 and H4; 2) Binding relationship verification Retrieve the physical feature hash value H1 corresponding to the quantum dot ID from the digital twin platform; re-collect the current physical features of the cultural relic, generate a real-time physical feature code, and calculate its real-time hash value H1'; verify whether H1 and H1' are consistent, and consider them consistent if the Hamming distance is ≤8 bits; if the verification is successful, the identity of the cultural relic and the binding relationship are confirmed to be valid, otherwise it is judged as abnormal; 3) Handling Abnormal Situations a. Mark reading failed: Adjust the excitation light angle and intensity, and try reading again; if it still fails, you can search the digital twin platform through other features of the cultural relic to find the corresponding record. After authorization by two or more administrators, reprint the mark near the original location and update the binding relationship. b. Damaged tag: If the tag is partially damaged, try to recover it using the error correction capability of the QR code; if it is completely damaged, reprint the tag after approval, mark the old tag ID as "invalid" and keep the record permanently; c. Updates after cultural relic restoration: After the cultural relic is restored, physical characteristics are re-collected to generate new physical characteristic codes; the binding relationship is updated, while the historical physical characteristic codes and binding records are retained to form a complete cultural relic status change archive.

9. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The specific method for predicting the development trend of cultural relic diseases based on the aforementioned digital twin management platform is as follows: S5.1: Construction of a Multimodal Temporal Disease Database 1) Core data collection and integration Multimodal detection data: Terahertz subsurface crack data, polarization stress distribution data, laser speckle micro-deformation data, and surface texture data collected in each iteration are aligned by timestamps to form a time sequence; Environmental monitoring data: minute-level continuous monitoring environmental data of the cultural relic storage / exhibition environment; Manually recorded data: Structured data including cultural relic restoration records, maintenance records, exhibition records, and transportation records; Disease labeling data: Labeling data of disease type, location, severity, and development stage, marked by cultural relics experts; 2) Data preprocessing and standardization Interpolation completion, outlier removal, and smoothing are performed on time-series data; A unified data format and coordinate system were used to map all disease characteristics to the three-dimensional spatial position of the digital twin of the cultural relic; Construct a four-dimensional data index of "cultural relic-part-disease-time" to support fast retrieval by part and disease type; S5.2: Extraction of Disease-Related Features 1) For each diseased part, extract the temporal feature sequence for more than 6 consecutive months; 2) Calculate the rate of change, acceleration, and fluctuation amplitude of the characteristic to derive the characteristic; 3) Use the mutual information method to screen key features with a correlation > 0.7 with disease development, and form a disease feature vector; S5.3: Hybrid prediction model construction, adopting a hybrid architecture of "data-driven AI model + physical mechanism model", which balances prediction accuracy and interpretability; 1) Basic Model of Physical Information Neural Network The mechanical, thermal, and chemical properties of cultural relics are used as prior knowledge and their physical laws are embedded into neural networks. The loss function consists of two parts: data loss + physical loss; 2) Set up dedicated prediction models for different diseases a) Crack propagation prediction model Input: Crack history length / depth sequence, stress concentration value sequence, ambient temperature and humidity sequence; Model: PINN combined with extended finite element method; Output: Predictions for crack length, depth, and propagation direction for the next 3 / 6 / 12 months; b. Material aging prediction model Input: Terahertz spectral characteristic sequence, cumulative illumination duration, and cumulative temperature and humidity fluctuation values; Model: Temporal Transformer + Material Aging Dynamics Equations; Output: Predictions of material aging degree and mechanical property degradation rate over the next 1 / 3 / 5 years; c. Structural loosening prediction model Inputs: natural frequency sequence, damping ratio sequence, vibration acceleration sequence; Model: LSTM + Modal Analysis Theory; Output: Structural loosening risk level and predicted loosening location for the next 1 / 3 / 6 months; 3) Multi-model fusion and uncertainty quantification The weighted average method is used to fuse the prediction results of multiple models, and the weights are dynamically adjusted according to the historical prediction accuracy of each model. The Monte Carlo dropout method is introduced to quantify the uncertainty of the prediction results and to give the confidence interval of the prediction values; When the forecast uncertainty exceeds 30%, a manual review process will be automatically triggered. S5.4: Tiered Early Warning and Decision Support 1) Risk level classification Based on the predicted severity of the damage and the value of the cultural relics, the early warning is divided into four levels: Level 1 Warning: A potentially fatal disease is predicted to occur within 3 months, requiring immediate protective measures. Level 2 warning: Severe diseases are predicted to occur within 6 months, requiring special inspection and repair. Level 3 warning: General diseases are predicted to occur within 12 months, and monitoring frequency needs to be increased; Level 4 warning: There is a potential risk of disease, and routine monitoring should be maintained; 2) Early warning triggering mechanism When the predicted disease parameters exceed the threshold of the corresponding level, an early warning will be automatically triggered; Taking into account the historical, artistic, scientific, and fragile value of cultural relics, the early warning threshold is dynamically adjusted. For warning signals that appear repeatedly in the same location, the warning level will be automatically upgraded. 3) Intelligent decision support Automatically generate targeted protection measure suggestions based on the knowledge base; The predicted location, development trend, and impact range of damage are displayed in three dimensions on the digital twin of cultural relics; Generate early warning reports, including disease details, prediction results, risk assessments, and protection recommendations, and push them to relevant management personnel; S5.5: Continuous Iterative Optimization of the Model After each collection of artifact condition data and restoration of damage, new data is added to the training set to incrementally update the model; The model's predictive accuracy should be evaluated at least quarterly, and the model parameters should be adjusted based on the evaluation results. An expert feedback mechanism has been established, allowing cultural relics experts to annotate and correct the prediction results, thereby continuously improving the model's performance.

10. The method for multimodal modeling, tracing, and verification of cultural relics according to claim 4, characterized in that, The specific process for verifying the return of the cultural relics is as follows: S6.1: Verification Initialization and Data Preparation Rapid identification: Scan the quantum dot invisible markers on cultural relics using a fluorescence excitation detection device, and decode them to obtain a 128-bit marker ID and encrypted binding credential; Baseline data retrieval: Quickly retrieve the artifact's data from the digital twin platform using its tagged ID. The multimodal baseline data package at the time of initial filing; historical data and anomaly records from each verification; metadata including at least the material, age, and preservation environment of the cultural relic; Real-time data acquisition: The integrated device is activated to automatically acquire the current multimodal data of the cultural relics, and the acquisition parameters are completely consistent with the initial archiving. S6.2: Data Preprocessing and 3D Spatial Alignment Preprocessing for real-time acquired single-modal data: Terahertz data: Wavelet denoising, reflectivity normalization, and background subtraction were performed; Polarization data: Complete Stokes parameter calculation, polarization degree / polarization angle diagram generation, and illumination correction; Speckle data: Phase unwrapping, micro-deformation field calculation, and noise filtering were performed; White light data: Complete distortion correction, white balance, and contrast enhancement; Global 3D Spatial Alignment An improved ICP algorithm was used to register the real-time acquired white light point cloud with the reference 3D model, with a registration accuracy of ±0.02mm; The rotation matrix and translation vector obtained from the registration are applied to all modal data to make the real-time data and the reference data completely aligned in the same three-dimensional coordinate system. For each three-dimensional vertex on the surface of the cultural relic, establish a one-to-one correspondence between real-time data and benchmark data; S6.3: Multimodal Siamese Network Feature Extraction and Fusion. A two-branch symmetric Siamese network architecture is adopted, with the two branches sharing weights. Real-time data and benchmark data are input respectively, and the corresponding multimodal fusion feature vectors are output. A modality-specific feature extraction branch is designed with independent feature extraction networks tailored to the characteristics of different modal data. Employing a triple attention mechanism to dynamically fuse multimodal features: space Attention: Automatically focuses on key feature areas of the artifact, suppressing background interference; Channel attention: Automatically selects the feature channels that are most valuable to the current alignment task; Modal attention: Automatically adjusts the weights of each modality based on the type of artifact; The final result is a 1024-dimensional global fusion feature vector; Similarity calculation: Calculate the cosine similarity between the global fused feature vectors of real-time data and benchmark data; Simultaneously, the similarity of independent feature vectors for each modality is calculated to obtain a multi-dimensional similarity score; Set a dynamic threshold: It is automatically adjusted based on the material of the cultural relic, the storage environment, and the last verification time, and is usually 0.95~0.98; If the global similarity is greater than or equal to the threshold, it is judged as normal; if it is less than the threshold, it enters the anomaly detection process. S6.4: Anomaly Area Location and Quantitative Analysis Anomaly heatmap generation: Calculate the differences between real-time data and baseline data on feature maps at each level; Upsampling restores the feature differences to the original image space, generating an anomaly heatmap. The intensity of the colors in a heatmap indicates the degree of abnormality: red indicates severe abnormality, yellow indicates moderate abnormality, and blue indicates normal. Abnormal region segmentation and extraction: Threshold segmentation combined with morphological operations is used to extract continuous anomalous regions from anomaly heatmaps; Each abnormal area is reconstructed in three dimensions and mapped to the corresponding position in the digital twin of the cultural relic; Abnormal parameter quantification calculation: The following quantification parameters are automatically calculated for each abnormal region: Geometric parameters: area, volume, maximum depth, perimeter, shape factor; Location parameters: 3D coordinates, location of the cultural relic, and distance from key features; Difference parameters: grayscale difference, spectral difference, stress difference, and deformation compared to the baseline data; Change parameters: the amount and rate of change compared to the last verification; S6.5: Anomaly Classification and Severity Assessment Anomaly type classification: Train a multi-classifier to classify anomalies into the following types: Physical damage: scratches, bumps, cracks, breaks, missing parts; Material changes: corrosion, oxidation, aging, discoloration; Structural abnormalities: loosening, delamination, deformation, hollowing; Human intervention: traces of repair, cleaning, and replacement; Natural changes: normal aging, minor deformations caused by temperature and humidity; Severity grading assessment: A weighted scoring method was used to comprehensively assess the severity of the anomalies. Level 1: The anomaly may cause the cultural relic to break, collapse, or be permanently damaged; Level 2: The anomaly significantly impacts the value of the cultural relic and requires immediate restoration; Level 3: The anomaly has a certain impact on the appearance of the cultural relic and requires restoration. Level 4: The anomaly does not affect the value and safety of the cultural relics; only enhanced monitoring is required. Distinguishing between natural changes and human-caused damage A natural aging model for cultural relics is established based on historical data to calculate whether the abnormal rate of change conforms to the laws of natural aging. By combining abnormal morphological characteristics, we can distinguish between natural aging and human-caused damage; S6.6: Automatically generate verification reports The system automatically generates a standardized digital verification report, which includes the following: Basic information: artifact number, name, material, era, verification time, and verification personnel; Overview of comparison results: global similarity score, similarity score for each modality, and verification conclusion; Anomaly Details: 3D visualization annotation of the anomaly area; type, location, quantification parameters, and severity of each anomaly; magnified and comparative views of the anomaly area; Trend analysis: Compare the results with previous verifications to analyze the development trend of anomalies; Handling suggestions: Automatically generate targeted handling suggestions based on the type and severity of the anomaly; Report Export: Supports export in PDF and Word formats, and generates a blockchain evidence hash to ensure the report is tamper-proof; S6.7: Digital Twin Updates and Data Accumulation Automatic data updates: The real-time multimodal data, comparison results, anomaly reports, and processing records of this verification will be automatically uploaded to the digital twin management platform; Digital twin iteration: Update the status data of the digital twin of the cultural relic and mark abnormal areas; generate a new version number, retain all historical versions, and support retrospective viewing; update the health status assessment and risk level of the cultural relic. Model self-optimization: The labeled data from this verification will be added to the training set to incrementally fine-tune the Siamese network model; the anomaly classifier and severity assessment model will be updated regularly based on new data; and the system's comparison accuracy and anomaly identification capabilities will be continuously improved.

Images