Cultural heritage propagation system and method based on multi-modal analysis and block chain excitation

Abstract

The invention discloses a cultural heritage propagation system and method based on multi-modal analysis and block chain incentive, comprising environmental perception, cultural decoding, value incentive and dynamic propagation, and according to the cultural heritage propagation system and method based on multi-modal analysis and block chain incentive, through the combination of a double-flow neural network and a cross attention mechanism, the problem of misrecognition under complex light and shadow is solved, and the accuracy of the cultural heritage propagation is improved. The symbol matching accuracy rate of 92.3% on a test set (only 78.5% in a traditional method) is achieved, and millisecond-level real-time reasoning is supported. A quantitative model of culture depth, scene analysis and propagation influence is created for the first time, contribution values are calculated in real time through on-chain Oracle, high-quality content rewards are improved by three times, and Ethereum and coin-safety chain double-standard NFT cross-chain circulation is supported. The prediction model fusing the Markov chain and the reinforcement learning realizes the propagation path accuracy of 89.2% in the trial, the calculation efficiency is improved by 3 times (the delay is reduced from 120 ms to 40 ms), and the topic life cycle is prolonged to 5 days.

Description

Technical Field

[0001] This invention relates to the field of cultural dissemination technology, and in particular to a multimodal analysis and blockchain-incentivized cultural heritage dissemination system and method. Background Technology

[0002] Current cultural heritage dissemination mainly includes, 1. User-uploaded content: Tourists upload images, videos or text related to cultural heritage through the platform. The content is stored in a static format.

[0003] 2. Cultural Heritage Database: It only stores the basic attributes of cultural relics (such as age and material), and lacks related data such as lighting parameters and dynamic narratives.

[0004] 3. Manual review and tagging: Rely on manual review of content compliance and manually add simple tags (such as "Buddha statue" or "building").

[0005] 4. Fixed push strategy: Content distribution is based on time or popularity ranking, without dynamic adjustment based on user behavior or environmental parameters.

[0006] Therefore, 1. User-generated content (UGC) is only used as material storage and lacks in-depth correlation analysis between lighting effects (such as changes in light and shadow, color temperature fluctuations) and cultural symbols (such as historical symbols, cultural heritage, and historical allusions). (For example, the correlation between light leakage trajectories and religious symbols).

[0007] 2. User participation relies on a single points reward, which fails to stimulate the creation of high-quality content (such as the lack of distinction between "ordinary check-in" and "in-depth cultural interpretation").

[0008] 3. The content distribution strategy is fixed and cannot be optimized for dissemination based on real-time data (such as changes in lighting conditions and user dwell time). Summary of the Invention

[0009] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is to provide a multimodal analysis and blockchain-incentivized cultural heritage dissemination system and method, which solves the problems of existing technologies not considering light effects, weak cultural connections, single incentive methods, fixed push strategies, and difficulty in adapting to complex environmental changes.

[0010] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A multimodal parsing and blockchain-incentivized cultural heritage dissemination system includes, The data acquisition module is used to collect scene data, lighting effect data, and user behavior data of cultural heritage scenes, forming a multi-dimensional data matrix of "lighting effect-behavior-space", and establishing a light and shadow culture knowledge graph database that maps and associates the multi-dimensional data matrix of the cultural heritage with its cultural symbols. The multimodal AI parsing module is used to extract scene data, lighting effect data, and user behavior data from user-uploaded UGC content. Based on the multidimensional data matrix of the cultural heritage and the association mapping relationship between cultural symbols in the lighting knowledge graph database, dynamic narrative content is generated, and sentiment analysis is performed based on user comments on UGC content. The blockchain incentive module is used to evaluate the value of UGC content and award rewards according to a tiered system based on a pre-defined reward contract. The adaptive propagation module is used to dynamically adjust the propagation path based on user behavior and environmental parameters, and dynamically generate contextualized tags.

[0011] As an optimization, the multimodal AI parsing module includes, The light effect feature extraction module adopts an improved dual-stream convolutional neural network. The physical parameter stream processes temporal light effect data through a 1D convolutional layer, while the visual feature stream extracts spatial light distribution features in the image based on ResNet-50. The two are semantically aligned through a cross-attention mechanism. The dynamic narrative generation module combines 3D-ResNet and LSTM networks to analyze the light effect gradation process in UGC content and automatically generate explanatory text that conforms to the context of cultural heritage. The cultural cognition assessment module uses the BERT model to perform sentiment analysis and knowledge density calculation on user comments, constructs a "scene perception-cultural understanding" correlation map, and identifies implicit correlations not recorded in the literature.

[0012] As an optimization, the blockchain incentive module includes, The multi-dimensional contribution quantification module constructs an evaluation system from three dimensions: cultural depth, scene resolution, and dissemination influence. It obtains data in real time through an on-chain oracle and calculates the contribution value. The NFT tiered reward module allows users to unlock rare digital collectibles by reaching a certain contribution threshold. The metadata of these collectibles is stored in IPFS and ownership is confirmed anonymously through zero-knowledge proofs.

[0013] As an optimization, the adaptive propagation module includes, The state-space modeling module encodes user behavior and environmental parameters into a joint state vector and dynamically adjusts the content distribution weights through the Q-Learning algorithm. The semantic tag dynamic generation module generates contextualized push tags based on real-time data, guiding social media platforms to conduct precise topic operations.

[0014] Based on the aforementioned dissemination system, this invention also provides a multimodal analysis and blockchain-incentivized method for cultural heritage dissemination, comprising the following steps: 1) Environmental perception; Collect scene data, lighting effect data, and user behavior data of cultural heritage to form a multi-dimensional data matrix of "lighting effect-behavior-space", and establish a light and scene culture knowledge graph database that maps and associates the multi-dimensional data matrix of cultural heritage with its cultural symbols; 2) Cultural Decoding; After users upload UGC content, the system analyzes the cultural heritage scene data, lighting effect data, and user behavior data in the UGC content. Based on the mapping relationship between the multidimensional data matrix of the cultural heritage and cultural symbols in the Light and Scene Culture Knowledge Graph Database, the system automatically generates a dynamic narrative and performs sentiment analysis based on user comments on the UGC content. 3) Value-based incentives; The value of user-uploaded UGC content is assessed, and tiered rewards are given according to a pre-set reward contract. 4) Dynamic propagation; The propagation path is dynamically adjusted based on user behavior and environmental parameters, and contextualized tags are dynamically generated.

[0015] As an optimization, the scene data includes the name, geography, and local weather of the cultural heritage site; the lighting effect data includes light intensity, color temperature, and incident angle; and the user behavior data includes the user's gaze area and dwell time.

[0016] As an optimization, in step 2), the parsing of UGC content adopts a dual-stream convolutional neural network. The physical parameter stream processes the temporal light effect data through a 1D convolutional layer, while the visual feature stream extracts the spatial light distribution features in the image based on ResNet-50. The two are semantically aligned through a cross-attention mechanism to realize the mapping association between specific light effects and cultural symbols. The dynamic narrative includes cultural interpretation content generated by analyzing the gradual changes in light effects using a 3D-ResNet+LSTM model; The sentiment analysis includes using the BERT model to analyze the sentiment tendency and cultural knowledge density in user comments, constructing a dynamic cultural cognition map, and discovering hidden connections.

[0017] As an optimization, in step 3), the value assessment of UGC content includes calculating the contribution value from three dimensions: cultural depth, accuracy of scene interpretation, and dissemination influence, and rewarding it according to the preset reward contract based on the contribution value.

[0018] As an optimization, in step 4), the user behavior state includes activity level, content preference, and social relationship strength, and the state categories are divided by DBSCAN clustering; the environmental parameters include light intensity, weather forecast, and geographical location, which together with the user behavior state form a joint state vector, and the state transition probability is calculated using a Markov chain prediction model, and the recommendation weight is adjusted according to the real-time click rate using a Q-Learning model to maximize the dissemination benefits.

[0019] Compared with the prior art, this application has the following advantages: This invention, 1. The combination of light effects and culture in the precise mapping of light effects and culture solves the problem of misidentification under complex light and shadow conditions. It achieves a symbol matching accuracy of 92.3% on the test set (compared to only 78.5% for traditional methods) and supports millisecond-level real-time inference.

[0020] 2. A three-dimensional dynamic incentive mechanism, pioneering a quantitative model for cultural depth, landscape analysis, and dissemination influence, calculates contribution value in real time through an on-chain oracle, increasing rewards for high-quality content by 3 times, and supports cross-chain circulation of NFTs on both Ethereum and Binance Chain.

[0021] 3. The adaptive propagation engine, which integrates Markov chains and reinforcement learning prediction models, achieved a propagation path accuracy of 89.2% in the pilot test, improved computational efficiency by 3 times (reduced latency from 120ms to 40ms), and extended the topic lifecycle to 5 days. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the technical principle of the present invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings.

[0024] For specific implementation: see [link / reference] Figure 1 , An example of a multimodal parsing and blockchain-incentivized cultural heritage dissemination system includes: The data acquisition module is used to collect scene data, lighting effect data, and user behavior data of cultural heritage sites, forming a multi-dimensional data matrix of "lighting effect-behavior-space". It also establishes a lighting culture knowledge graph database that maps and associates this multi-dimensional data matrix with its cultural symbols. Specifically, existing technologies largely rely on static content uploaded by users (such as photos and text), lacking dynamic perception of lighting environment parameters. This invention integrates a miniature lighting sensor (sampling rate 1kHz) to collect physical parameters such as light intensity (LUX), color temperature (CCT), and incident angle in real time within the cultural heritage site, and records them synchronously with user behavior data (gazing hotspots, dwell time), forming a multi-dimensional data matrix of "lighting effect-behavior-space". For example, deploying a sensor network in the Thousand-Hand Guanyin Hall of the Dazu Rock Carvings can capture the light leakage paths formed by sunlight passing through the protected building at different times of day, providing a data foundation for subsequent cultural narrative generation. The process and method of establishing the lighting culture knowledge graph database are existing technologies and will not be elaborated further.

[0025] To address the issues of low efficiency and weak cultural relevance in existing technologies, this invention designs a hybrid parsing architecture that integrates computer vision and natural language processing: a multimodal AI parsing module, which extracts scene data, lighting effect data, and user behavior data from user-uploaded UGC content, generates dynamic narrative content based on the multidimensional data matrix of the cultural heritage and the association mapping relationship between cultural symbols in the lighting knowledge graph database, and performs sentiment analysis based on user comments on the UGC content.

[0026] Specifically, the multimodal AI parsing module includes, An improved dual-stream convolutional neural network (TS-CNN) is adopted, in which the physical parameter stream processes temporal light effect data through 1D convolutional layers, while the visual feature stream extracts spatial light distribution features in the image based on ResNet-50. The two are semantically aligned through a cross-attention mechanism. The dynamic narrative generation module combines 3D-ResNet and LSTM networks to analyze the gradual change of light effects in the video stream (such as the silhouette change at sunrise in Yuanjue Cave) and automatically generate explanatory text that conforms to the cultural heritage context (such as 'backlit silhouettes symbolize the threshold of history, echoing the cultural symbol of local historical heritage'). The cultural cognition assessment module uses the BERT model to perform sentiment analysis and knowledge density calculation on user comments, constructs a "light and scene perception-cultural understanding" correlation map, and identifies implicit correlations not recorded in the literature (such as "the emotional resonance threshold between the intensity of reflection during the rainy season and historical and cultural imagery").

[0027] Traditional UGC platforms generally suffer from drawbacks such as a single incentive method and a lack of copyright protection. This invention proposes a dynamic contribution evaluation model based on smart contracts: a blockchain incentive module used to evaluate the value of UGC content and award rewards in tiers according to a pre-set reward contract.

[0028] Specifically, the blockchain incentive module includes, The multi-dimensional contribution quantification module constructs an evaluation system from three dimensions: cultural depth (frequency of Buddhist classic references), scene resolution (AI matching confidence), and dissemination influence (cross-platform forwarding volume). It obtains data in real time through an on-chain oracle and calculates the contribution value.

[0029] The NFT tiered reward module unlocks rare digital collectibles (such as the "Historical Site Dynamic Light Effect Badge") by reaching a certain contribution threshold. The metadata of these collectibles is stored in IPFS and anonymized through zero-knowledge proofs. This mechanism, piloted at the Dazu Rock Carvings, increased the proportion of high-quality content from 12% to 41%.

[0030] Existing technologies rely on fixed push strategies, making it difficult to adapt to complex environmental changes. This patent introduces a fusion model of Markov chains and reinforcement learning: an adaptive propagation module, used to dynamically adjust the propagation path based on user behavior and environmental parameters, and dynamically generate contextualized tags.

[0031] Specifically, the adaptive propagation module includes, The state space modeling module encodes user behavior (activity level, content preference) and environmental parameters (lighting conditions, weather conditions) into a joint state vector, and dynamically adjusts the content distribution weights through the Q-Learning algorithm.

[0032] The semantic tag dynamic generation module generates contextualized push tags based on real-time data (such as "Sacred Relic Pool Reflection - LUX value 152 - Reflecting historical and cultural symbols") to guide social media platforms in precise topic operation. Tests show that this strategy extends the average lifespan of topics from 48 hours to 120 hours.

[0033] Based on the aforementioned dissemination system, this invention also provides a multimodal analysis and blockchain-incentivized method for cultural heritage dissemination, comprising the following steps: 1) Environmental perception; Scene data, lighting effect data, and user behavior data of cultural heritage are collected to form a multi-dimensional data matrix of "lighting effect-behavior-space". A knowledge graph database of light and scene culture is established to map and associate the multi-dimensional data matrix of cultural heritage with its cultural symbols. The scene data includes the name, geography, and local weather of the cultural heritage. The lighting effect data includes light intensity, color temperature, and incident angle. The user behavior data includes the user's gaze area and dwell time.

[0034] 2) Cultural Decoding; After users upload UGC content, the system analyzes the cultural heritage scene data, lighting effect data, and user behavior data within the UGC content. Based on the mapping relationship between the multidimensional data matrix of the cultural heritage and cultural symbols in the Light Scene Culture Knowledge Graph Database, a dynamic narrative is automatically generated, and sentiment analysis is performed based on user comments on the UGC content. The parsing of UGC content uses a dual-stream convolutional neural network. The physical parameter stream processes temporal lighting effect data through 1D convolutional layers, while the visual feature stream extracts spatial light distribution features from the image based on ResNet-50. The two streams are semantically aligned through a cross-attention mechanism to recognize the mapping relationship between specific lighting effects and cultural symbols. The dynamic narrative includes cultural interpretation content generated by analyzing the gradual changes in light effects using a 3D-ResNet+LSTM model; The sentiment analysis includes using the BERT model to analyze the sentiment tendency and cultural knowledge density in user comments, constructing a dynamic cultural cognition map, and discovering hidden connections.

[0035] 3) Value-based incentives; The system assesses the value of user-uploaded UGC content and awards it in tiers according to a pre-defined reward contract. The value assessment of UGC content includes calculating contribution value from three dimensions: cultural depth, accuracy of scene interpretation, and dissemination influence. Based on the contribution value, rewards are awarded in tiers according to the pre-defined reward contract.

[0036] 4) Dynamic propagation; The propagation path is dynamically adjusted based on user behavior and environmental parameters, and contextualized tags are dynamically generated. User behavior includes activity level, content preference, and social relationship strength, which are categorized using DBSCAN clustering. Environmental parameters include light intensity, weather forecast, and geographic location, which, together with user behavior, form a joint state vector. A Markov chain prediction model is used to calculate state transition probabilities, and a Q-Learning model is used to adjust recommendation weights based on real-time click-through rates to maximize propagation benefits.

[0037] In summary, the core of this invention lies in constructing an intelligent, closed-loop cultural heritage dissemination system. By integrating environmental perception, artificial intelligence, blockchain incentives, and dynamic dissemination optimization technologies, it addresses the problems of superficial content value mining, rigid incentive mechanisms, and singular dissemination paths in existing cultural heritage dissemination.

[0038] Overall System Workflow From the moment visitors step into a cultural heritage site (such as the Dazu Rock Carvings), the system begins a "symphony of light and culture": 1. Environmental Perception: Capturing the Language of Light. A high-precision network of light sensors deployed at cultural heritage sites acts like countless "digital eyes," recording light intensity, color temperature, and angle thousands of times per second. For example, when sunlight shines on the Thousand-Hand Guanyin statue at a 45-degree angle, the sensors not only capture the warm yellow temperature of 2700K but also simultaneously track visitors' gaze hotspots and dwell times. This data is linked in real-time with user-uploaded images and videos, forming a three-dimensional data stream of "light effect-behavior-space," providing a realistic physical world mirror for subsequent analysis.

[0039] 2. Cultural Decoding: The AI-powered system for cracking the code of light and shadow employs a dual-stream neural network architecture. One branch analyzes the temporal changes in light effects (such as the rhythm of light and shadow at dawn and dusk), while the other branch analyzes visual details in the image (such as the light and shadow transitions in the gilding of a Buddha statue). Through a cross-attention mechanism, the two branches interact like archaeologists comparing documents with physical objects, precisely linking light and shadow with cultural symbols. For example, when the system detects a light intensity of 86 lux, the light effect created by the reflection on the statue's hands triggers the AI ​​to generate corresponding cultural explanations. Simultaneously, the BERT model analyzes the frequency of Buddhist classic references in user comments, constructing a dynamic cultural cognitive map and discovering hidden connections (such as the intensity of reflection during the rainy season and the emotional resonance threshold of the "Nirvana imagery").

[0040] 3. Value Incentives: After AI evaluation, user-uploaded content is evaluated using blockchain-based quantified content creation. The system calculates contribution value based on three dimensions: cultural depth (e.g., number of citations of historical and cultural classics), accuracy of scene analysis (AI matching rate ≥ 85%), and dissemination influence (cross-platform reposting volume). Smart contracts automatically distribute rewards: first-time uploaders receive a "Light and Shadow Explorer" NFT badge; users accumulating 1000 points unlock the "Historical Site Light and Shadow Master" dynamic badge, with its metadata permanently stored via IPFS and anonymously established using zero-knowledge proof technology. Pilot data shows that this mechanism increased the proportion of high-quality content from 12% to 41%.

[0041] 4. Dynamic Dissemination: The system, like a shrewd strategist, adapts its push notification strategy based on real-time environmental changes. A Markov chain model predicts dissemination paths: if rain is forecast for the next day, it pushes out an analysis of the light and shadow effects of an indoor grotto in advance; if a user lingers in front of a statue for more than 5 minutes, it immediately pushes the story of the "mirror image" via AR glasses. A reinforcement learning model dynamically optimizes weights based on user click-through rates, extending the lifespan of content related to "stone carving reflections and the imagery of Nirvana" from 48 hours to 120 hours, making it a cultural hotspot on social media.

[0042] More specifically, in this invention, I. Multimodal AI Analysis Engine: The decoding technology architecture from light and shadow to cultural narrative adopts a hybrid model of "dual-stream neural network + semantic alignment" to achieve a deep connection between physical lighting effects and cultural heritage.

[0043] Implementation details 1. Data Acquisition Layer Deploy a network of miniature light sensors to collect light intensity (LUX), color temperature (CCT), and incident angle (XYZ three-dimensional coordinates) in real time at a sampling rate of 1kHz.

[0044] Synchronously bind user behavior data (gaze area, dwell time) to form a spatiotemporal matrix of "light effect-behavior-space".

[0045] Example: When a user takes a picture of the Thousand-Hand Guanyin, the system records the sunlight color temperature (2700K), the angle of incidence (45°), the duration of the shot (5 minutes and 32 seconds), and the coordinates of the shooting location.

[0046] 2. Feature Extraction Layer Two-stream convolutional neural network (TS-CNN) Physical parameter stream: processes time-series data of light effects (such as color temperature change curves) and extracts periodic features through 1D convolution.

[0047] Visual Feature Stream: Processes image / video frames, extracts spatial features through ResNet-50, and the spatial attention module focuses on key areas (such as the gold plating and reflection on Guanyin's hands).

[0048] Cross-attention mechanism: dynamically align dual-stream features to identify the association between specific lighting effects and cultural symbols (such as the cultural explanatory text corresponding to the reflection mapping of Guanyin's hand under 86LUX lighting).

[0049] 3. Dynamic Narrative Generation 3D-ResNet+LSTM model: Analyze the gradual changes in light effects (such as the changes in sunrise silhouette) and generate cultural interpretations (such as "backlit silhouettes symbolize the threshold of history").

[0050] BERT Cultural Cognition Analysis: Analyzing the sentiment and cultural knowledge density (such as the frequency of historical and cultural references) in user comments to construct a cognitive map.

[0051] Technological advantages Real-time performance: Inference latency <50ms, supporting real-time interaction on mobile devices.

[0052] Robustness: With multi-sensor fusion and adversarial training, symbol matching accuracy reached 92.3% under complex lighting conditions.

[0053] II. Blockchain-enabled dynamic incentive mechanism: quantifying contributions and stimulating creativity. The technical architecture is based on a multi-dimensional contribution evaluation model of smart contracts, combined with NFT and cross-chain technology to achieve transparent incentives.

[0054] Implementation details 1. Contribution Measurement Model Cultural Depth Index: The RoBERTa model counts the number of times Buddhist classics are cited in user texts (10 points are added for each citation).

[0055] Scene resolution: AI matching confidence level ≥ 85% is judged as high-quality content.

[0056] Dissemination influence: Calculate the number of reposts across platforms and the user's PageRank value.

[0057] Formula: Total score = 0.6 × cultural depth + 0.3 × visual resolution + 0.1 × dissemination influence.

[0058] 2. Tiered Rewards for NFTs Basic NFT: Earn the "Light and Shadow Explorer" badge upon your first content upload, with metadata stored in IPFS.

[0059] Rare NFT: Unlock the "Master of Dazu Rock Carvings" badge by accumulating 1000 points, supporting cross-chain transfer between ERC-721 and BEP-721.

[0060] Privacy Protection: zk-SNARK zero-knowledge proofs enable anonymous rights confirmation.

[0061] Technological advantages Transparent and trustworthy: On-chain oracle verifies data in real time, eliminating the need for human intervention.

[0062] Incentive effectiveness: The proportion of high-quality content in the pilot program increased to 41%.

[0063] III. Adaptive Propagation Optimization Engine: Dynamic Strategy-Driven Precise Reach Technology Architecture integrates Markov chains and reinforcement learning prediction models to dynamically adjust propagation strategies.

[0064] Implementation details 1. State-space modeling User behavior status: Define activity level, content preference, and social relationship strength, and classify status categories using DBSCAN clustering.

[0065] Environmental parameter status: Integrate light intensity, weather forecast, and geographical location to construct a joint state vector.

[0066] 2. Dynamic Strategy Generation Markov chain prediction: Calculate state transition probabilities (e.g., the probability of pushing indoor scenes increases by 60% on rainy days).

[0067] Q-Learning optimization: Adjust recommendation weights based on real-time click-through rates to maximize dissemination benefits.

[0068] Example: If a user stays in the AR glasses for more than 5 minutes, the AR glasses will push the story of the "image in the mirror" and real-time lighting parameters.

[0069] 3. Semantic Tag Generation Dynamically generate contextualized tags (such as "Sacred Relic Pool Reflection - LUX Value 152 - Reflecting Historical and Cultural Symbols") to guide social media topic management.

[0070] Technological advantages Prediction accuracy: The propagation path accuracy is 89.2%, which is 30% higher than traditional methods.

[0071] Computational efficiency: Latency reduced from 120ms to 40ms.

[0072] The core technologies of each module in this invention are: 1. Dynamic mapping algorithm for light effects and cultural symbols Core technology: Multimodal feature fusion architecture: Two-Stream CNN was used to process the physical parameters of light effects (illuminance, color temperature, and angle of incidence) and the visual features of cultural heritage (image texture and spatial layout).

[0073] Physical parameter stream: The input is real-time data from the optical sensor (LUX value, CIE chromaticity coordinates), and the temporal features are extracted through a 1D convolutional layer. The output is a 128-dimensional feature vector.

[0074] Visual feature flow: The input is a high-resolution image / video frame. Multi-scale features are extracted using a ResNet-50 backbone network. The Spatial Attention Module focuses on key regions (such as the details of the hands of the Thousand-Hand Guanyin).

[0075] Feature fusion layer: Input dual-stream features into the cross-attention mechanism to generate a dynamic weight matrix, thereby achieving semantic alignment between light effect parameters and cultural symbols.

[0076] Training of dynamic mapping models: Dataset Construction: Collect multimodal data (light effect parameters + cultural annotations) from heritage sites such as Dazu Rock Carvings, and construct an annotated dataset (example annotation: {"light effect parameters": [86lx, 2700K, 45°], "cultural symbols": "historical site - light and shadow symbol", "related classics": "cultural heritage related literature"}).

[0077] Loss function design: Contrastive loss is adopted to maximize the similarity of positive samples (different lighting effects of the same cultural symbols) and minimize the similarity of negative samples (similar lighting effects of different symbols).

[0078] Optimization strategy: Use the AdamW optimizer and combine it with Curriculum Learning to gradually increase the lighting complexity (from a single light source to a multi-light source mixed scene).

[0079] Innovative breakthrough: Cross-modal alignment accuracy: Achieves 92.3% symbol matching accuracy on the test set (the highest existing technology is 78.5%), supporting real-time inference (latency <50ms).

[0080] Adaptive lighting modeling: By pre-training a lighting transmission model using a physics engine (such as the Unreal Engine optics module), robust analysis of complex ambient light (such as diffuse reflection in cloudy weather) is achieved.

[0081] 2. Contribution Measurement Model Empowered by Blockchain Core technology: Multi-dimensional contribution evaluation system: Content quality dimensions: Cultural Depth Index: Analyzes the density of cultural concepts (such as the number of historical and cultural allusions cited) in user-generated text using NLP models (such as RoBERTa).

[0082] Light and shadow resolution: Based on the confidence level of light effect-symbol matching output by the AI ​​engine (if the matching probability is >85%, it is judged as high-value content).

[0083] Dimensions of dissemination influence: Social network centrality: Calculates a user's PageRank value in the propagation network to measure their ability to spread content.

[0084] Cross-platform coverage: Statistics on the number of reposts and interaction rates of content on platforms such as WeChat, Douyin, and Instagram are used to generate a weighted score for dissemination power.

[0085] Smart contract design: Dynamic reward allocation: Design an on-chain Oracle to obtain contribution evaluation data output by the AI ​​engine in real time, and calculate user points using the following formula: Score = α × Cultural Depth Index + β × Dissemination Influence Index + γ × Content Timeliness Factor (where α = 0.6, β = 0.3, γ = 0.1, and the weights can be dynamically adjusted). NFT graded casting: Basic NFT: Users can obtain this upon their first content upload (such as the "Light and Shadow Explorer" badge).

[0086] Rare NFTs: These are unlocked after accumulating a certain number of points (e.g., 1000 points) (e.g., the "Master of Dazu Rock Carvings" dynamic badge).

[0087] IPFS proof-of-existence mechanism: User-generated content (images / videos) is stored in IPFS, a CID hash value is generated and written to a smart contract to ensure that the data is immutable.

[0088] Privacy is protected through zero-knowledge proofs (zk-SNARK), allowing users to anonymously contribute content while verifying copyright ownership.

[0089] Innovative breakthrough: Contribution quantification granularity: Achieving a leap from "extensive points" to "multi-dimensional dynamic evaluation", with a 300% increase in rewards for high-quality content.

[0090] Cross-chain interoperability: Supports dual-standard NFTs of ERC-721 (Ethereum) and BEP-721 (Binance Chain), covering the mainstream blockchain ecosystem.

[0091] 3. Markov Chain Propagation Path Prediction Engine Core technology: State-space modeling: User behavior state: Define user state as S = {activity level, content type preference, social relationship strength}, and classify state categories using clustering algorithms (such as DBSCAN).

[0092] Content dissemination status: Define the content status as C = {dissemination platform, topic popularity, interaction decay rate}, and construct a state transition matrix.

[0093] Dynamic prediction algorithm: Markov chain optimization: Hidden Markov Model (HMM) is used in conjunction with the Viterbi algorithm to predict the optimal propagation path.

[0094] A reinforcement learning (RL) mechanism is introduced, and the state transition probabilities (such as platform selection weights) are dynamically adjusted through Q-Learning.

[0095] Real-time feedback closed loop: Deploy edge computing nodes to collect real-time data from social media platform APIs (such as Weibo topic growth rate) and update model parameters.

[0096] Design an adaptive decay function to dynamically reduce the recommendation weight of outdated content (e.g., decay by 50% if there is no interaction within 24 hours).

[0097] Innovative breakthrough: Improved prediction accuracy: In the pilot project at Dazu Rock Carvings in Chongqing, the accuracy rate of propagation path prediction reached 89.2% (compared to 67.4% for traditional methods).

[0098] Computational efficiency optimization: Through model pruning and quantization, inference speed is improved by 3 times (latency reduced from 120ms to 40ms). In summary, this invention, 1. The combination of light effects and culture in the precise mapping of light effects and culture solves the problem of misidentification under complex light and shadow conditions. It achieves a symbol matching accuracy of 92.3% on the test set (compared to only 78.5% for traditional methods) and supports millisecond-level real-time inference.

[0099] 2. A three-dimensional dynamic incentive mechanism, pioneering a quantitative model for cultural depth, landscape analysis, and dissemination influence, calculates contribution value in real time through an on-chain oracle, increasing rewards for high-quality content by 3 times, and supports cross-chain circulation of NFTs on both Ethereum and Binance Chain.

[0100] 3. The adaptive propagation engine, which integrates Markov chains and reinforcement learning prediction models, achieved a propagation path accuracy of 89.2% in the pilot test, improved computational efficiency by 3 times (reduced latency from 120ms to 40ms), and extended the topic lifecycle to 5 days.

[0101] Although embodiments of the present invention have been shown and described, those skilled in the art will be able to make various changes, modifications, substitutions and variations to these embodiments without departing from the principles and basis of the present invention. The scope of the present invention is defined by the appended claims and their equivalents. Therefore, the embodiments of the present invention are merely illustrative examples and should not be construed as limiting the present invention in any way.

Claims

1. A cultural heritage dissemination system based on multimodal analysis and blockchain incentives, characterized in that, include, The data acquisition module is used to collect scene data, lighting effect data, and user behavior data of cultural heritage scenes, forming a multi-dimensional data matrix of "lighting effect-behavior-space", and establishing a light and shadow culture knowledge graph database that maps and associates the multi-dimensional data matrix of the cultural heritage with its cultural symbols. The multimodal AI parsing module is used to extract scene data, lighting effect data, and user behavior data from user-uploaded UGC content. Based on the multidimensional data matrix of the cultural heritage and the association mapping relationship between cultural symbols in the lighting knowledge graph database, dynamic narrative content is generated, and sentiment analysis is performed based on user comments on UGC content. The blockchain incentive module is used to evaluate the value of UGC content and award rewards according to a tiered system based on a pre-defined reward contract. The adaptive propagation module is used to dynamically adjust the propagation path based on user behavior and environmental parameters, and dynamically generate contextualized tags.

2. The cultural heritage dissemination system based on multimodal analysis and blockchain incentives according to claim 1, characterized in that, The multimodal AI parsing module includes, The light effect feature extraction module adopts an improved dual-stream convolutional neural network. The physical parameter stream processes temporal light effect data through a 1D convolutional layer, while the visual feature stream extracts spatial light distribution features in the image based on ResNet-50. The two are semantically aligned through a cross-attention mechanism. The dynamic narrative generation module combines 3D-ResNet and LSTM networks to analyze the light effect gradation process in UGC content and automatically generate explanatory text that conforms to the context of cultural heritage. The cultural cognition assessment module uses the BERT model to perform sentiment analysis and knowledge density calculation on user comments, constructs a "scene perception-cultural understanding" correlation map, and identifies implicit correlations not recorded in the literature.

3. The cultural heritage dissemination system based on multimodal analysis and blockchain incentives according to claim 1, characterized in that, The blockchain incentive module includes, The multi-dimensional contribution quantification module constructs an evaluation system from three dimensions: cultural depth, scene resolution, and dissemination influence. It obtains data in real time through an on-chain oracle and calculates the contribution value. The NFT tiered reward module allows users to unlock rare digital collectibles by reaching a certain contribution threshold. The metadata of these collectibles is stored in IPFS and ownership is confirmed anonymously through zero-knowledge proofs.

4. A cultural heritage dissemination system based on multimodal analysis and blockchain incentives according to claim 1, characterized in that, The adaptive propagation module includes, The state-space modeling module encodes user behavior and environmental parameters into a joint state vector and dynamically adjusts the content distribution weights through the Q-Learning algorithm. The semantic tag dynamic generation module generates contextualized push tags based on real-time data, guiding social media platforms to conduct precise topic operations.

5. The method for cultural heritage dissemination based on multimodal analysis and blockchain incentives according to claim 1, characterized in that, Includes the following steps, 1) Environmental perception; Collect scene data, lighting effect data, and user behavior data of cultural heritage to form a multi-dimensional data matrix of "lighting effect-behavior-space", and establish a knowledge graph database of light and scene culture that maps and associates the multi-dimensional data matrix of cultural heritage with its cultural symbols; 2) Cultural Decoding; After users upload UGC content, the system analyzes the cultural heritage scene data, lighting effect data, and user behavior data in the UGC content. Based on the mapping relationship between the multidimensional data matrix of the cultural heritage and cultural symbols in the Light and Scene Culture Knowledge Graph Database, the system automatically generates a dynamic narrative and performs sentiment analysis based on user comments on the UGC content. 3) Value-based incentives; The value of user-uploaded UGC content is assessed, and tiered rewards are given according to a pre-set reward contract. 4) Dynamic propagation; The propagation path is dynamically adjusted based on user behavior and environmental parameters, and contextualized tags are dynamically generated.

6. The method for cultural heritage dissemination based on multimodal analysis and blockchain incentives according to claim 5, characterized in that, The scene data includes the name, geography, and local weather of the cultural heritage site; the lighting effect data includes light intensity, color temperature, and angle of incidence; and the user behavior data includes the user's gaze area and dwell time.

7. A method for cultural heritage dissemination based on multimodal analysis and blockchain incentives according to claim 5, characterized in that, In step 2), the parsing of UGC content adopts a dual-stream convolutional neural network. The physical parameter stream processes the temporal light effect data through a 1D convolutional layer, while the visual feature stream extracts the spatial light distribution features in the image based on ResNet-50. The two are semantically aligned through a cross-attention mechanism to realize the mapping association between specific light effects and cultural symbols. The dynamic narrative includes cultural interpretation content generated by analyzing the gradual changes in light effects using a 3D-ResNet+LSTM model; The sentiment analysis includes using the BERT model to analyze the sentiment tendency and cultural knowledge density in user comments, constructing a dynamic cultural cognition map, and discovering hidden connections.

8. A method for cultural heritage dissemination based on multimodal analysis and blockchain incentives according to claim 5, characterized in that, In step 3), the value assessment of UGC content includes calculating contribution value from three dimensions: cultural depth, accuracy of scene interpretation, and dissemination influence, and rewarding content according to the pre-set reward contract based on the contribution value.

9. A method for cultural heritage dissemination based on multimodal analysis and blockchain incentives according to claim 5, characterized in that, In step 4), the user behavior state includes activity level, content preference, and social relationship strength, and the state categories are divided by DBSCAN clustering; the environmental parameters include light intensity, weather forecast, and geographical location, which together with the user behavior state form a joint state vector, and the state transition probability is calculated using a Markov chain prediction model. The recommendation weight is adjusted according to the real-time click rate using a Q-Learning model to maximize the dissemination benefits.

Images