Encryption method and system of LED conference display screen

Abstract

The invention relates to the technical field of display screen encryption, and discloses an encryption method and system for an LED conference display screen, and the method comprises the steps: collecting and analyzing conference scene data, and obtaining a conference feature vector, a content feature vector and user role information; performing multi-layer neural network processing to obtain a dynamic encryption demand score and a plurality of candidate encryption strategies; constructing a double-layer model based on the dynamic encryption demand score and a plurality of candidate encryption strategies; solving according to the double-layer model to obtain a global optimal encryption strategy; lED display content is scanned and graded in real time, and a multi-element content set is obtained; according to the method, the multi-level encrypted display data stream is generated based on the dynamic sensitivity level of the multi-element content set, decryption control is carried out, the decrypted display content which can be accessed by the authorized user is obtained, the conference scene, the content characteristics and the user behaviors can be analyzed in real time, the encryption strategy is adjusted in a self-adaptive mode, and multi-level and differentiated content protection is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of display screen encryption, and in particular to an encryption method and system for an LED conference display screen. Background Art

[0002] LED conference displays are widely used in modern conference rooms, facilitating information sharing and collaboration. However, meetings often involve sensitive information and confidential data, and traditional display encryption methods are difficult to meet security requirements in dynamic conference environments. Existing encryption technologies mostly use static encryption strategies, which cannot be flexibly adjusted according to real-time changes in conference content and user roles, making it difficult to balance security and user experience. In addition, traditional methods are inefficient when processing multimedia content, making it difficult to achieve differentiated protection for different types of data.

[0003] In complex conference scenarios, user rights management and content access control also face huge challenges. The existing role-based access control model lacks flexibility and is difficult to adapt to dynamically changing conference needs and temporary authorization situations. At the same time, traditional watermarking technology has limited effectiveness in protecting intellectual property rights and tracking information leaks, and is difficult to deal with various tampering and attack behaviors in high-definition display screens. Summary of the invention

[0004] The present invention provides an encryption method and system for an LED conference display screen, which are used for real-time analysis of conference scenes, content features and user behaviors, adaptively adjusting encryption strategies, and realizing multi-level and differentiated content protection.

[0005] In a first aspect, the present invention provides an encryption method for an LED conference display screen, the encryption method for an LED conference display screen comprising:

[0006] Collect and analyze conference scene data to obtain conference feature vectors, content feature vectors, and user role information;

[0007] Performing multi-layer neural network processing on the conference feature vector, the content feature vector and the user role information to obtain a dynamic encryption requirement score and multiple candidate encryption strategies;

[0008] constructing a two-level model based on the dynamic encryption requirement score and multiple candidate encryption strategies;

[0009] By solving the two-level model through equivalent complementary constraints, a global optimal encryption strategy is obtained;

[0010] Based on the global optimal encryption strategy, the LED display content is scanned and graded in real time to obtain a multi-element content set with dynamic sensitivity levels;

[0011] A multi-level encrypted display data stream is generated based on the dynamic sensitivity level of the multi-element content set and decryption control is performed to obtain decrypted display content accessible to authorized users.

[0012] In a second aspect, the present invention provides an encryption system for an LED conference display screen, the encryption system for an LED conference display screen comprising:

[0013] The acquisition module is used to collect and analyze the conference scene data to obtain the conference feature vector, content feature vector and user role information;

[0014] A processing module, configured to perform multi-layer neural network processing on the conference feature vector, the content feature vector and the user role information to obtain a dynamic encryption requirement score and a plurality of candidate encryption strategies;

[0015] A construction module for constructing a two-level model based on the dynamic encryption requirement score and a plurality of candidate encryption strategies;

[0016] A solution module, used for solving the global optimal encryption strategy according to the two-level model through equivalent complementary constraints;

[0017] A grading module, used for scanning and grading the LED display content in real time based on the global optimal encryption strategy to obtain a multi-element content set with dynamic sensitivity levels;

[0018] The control module is used to generate a multi-level encrypted display data stream based on the dynamic sensitivity level of the multi-element content set and perform decryption control to obtain decrypted display content accessible to authorized users.

[0019] The third aspect of the present invention provides an encryption device for an LED conference display screen, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the encryption device for the LED conference display screen executes the above-mentioned encryption method for the LED conference display screen.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned encryption method for the LED conference display screen.

[0021] In the technical solution provided by the present invention, through the real-time collection and analysis of conference scene data, combined with multi-layer neural network processing, the method can accurately identify conference features, content features and user role information, adopt a two-level model and a mixed integer quadratic programming method, and achieve global optimization of security and delay. At the same time, through flexible authority allocation based on RBAC, the LED display content is scanned in real time and dynamically graded for sensitivity. Combined with layered differential encryption and dynamic watermark embedding technology, refined protection of different types of content is achieved, and multi-factor authentication and trust evaluation mechanisms are introduced. Combined with homomorphic decryption and secure multi-party computing technology, flexible decryption control is achieved while ensuring data security, effectively preventing unauthorized access. Through isolated computing and secure rendering technology, including memory isolation, GPU virtualization, trusted execution environment and other measures, the security of decrypted display content is ensured, and potential security threats in the rendering process are effectively prevented. The anti-tampering ability and traceability of the content are enhanced, and the transmission efficiency of encrypted data is optimized through compression and redundant data addition technology, while improving the error recovery ability of the system and enhancing the reliability of the display system.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of an embodiment of an encryption method for an LED conference display screen in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of an embodiment of an encryption system for an LED conference display screen in an embodiment of the present invention;

[0026] Figure 3 The figure is a schematic diagram of an embodiment of an encryption device for an LED conference display screen in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0029] To facilitate understanding of this embodiment, firstly, an encryption method for an LED conference display screen disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps:

[0030] 101. Collect and analyze conference scene data to obtain conference feature vectors, content feature vectors, and user role information;

[0031] It is understandable that the execution subject of the present invention may be an encryption system of an LED conference display screen, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as an execution subject as an example.

[0032] Specifically, cluster analysis is performed on the meeting type data, and the different types of meeting data are classified using clustering algorithms to generate a meeting type feature matrix. This feature matrix can effectively capture the common features of different types of meetings. Based on the meeting type feature matrix, feature extraction is performed on the ongoing meeting data to obtain a meeting feature vector, including the core information of the meeting, such as the meeting theme, the number of participants, and the main content of the discussion. At the same time, optical character recognition and image segmentation are performed on the content displayed on the LED display. Optical character recognition technology is used to convert the text information on the screen into a processable text data set, while image segmentation divides the displayed image content into different areas to generate an image data set. The generated text data set is vectorized, and the natural language processing technology is used to convert the text information into a text feature matrix, which can represent the semantic content and structural features of the text. At the same time, feature extraction is performed on the image data set, and the image is converted into an image feature matrix using a deep learning algorithm. The text feature matrix and the image feature matrix are tensor fused to combine the features of the text and the image to form a comprehensive content feature vector. At the same time, the user's behavior log is analyzed, and the user's behavior pattern in the meeting is tracked through time series analysis technology. The key features of user behavior are identified using frequent pattern mining technology to generate a user behavior feature set. Based on this feature set, a user similarity graph is constructed, and the graph is used to show the similarity relationship between different users. Spectral clustering is performed on the user similarity graph, and users with high similarity are grouped into the same group to generate initial user role groups, and the user's role in the meeting is preliminarily determined, such as speaker, listener, etc. The initial user role grouping is dynamically adjusted, and accurate user role information is obtained by comprehensively considering the user's behavior changes and interaction.

[0033] 102. Perform multi-layer neural network processing on the conference feature vector, content feature vector and user role information to obtain a dynamic encryption requirement score and multiple candidate encryption strategies;

[0034] Specifically, the conference feature vector, content feature vector and user role information are combined into tensors to obtain an input tensor. The input tensor is extracted using a three-dimensional convolutional neural network. The three-dimensional convolutional neural network has the ability to capture spatial and temporal features, and can deeply mine the complex structure of the input tensor to generate a high-dimensional feature map, which contains multi-dimensional representations of conference features, content features and user role information, reflecting the spatial and temporal correlation of these information. The high-dimensional feature map is weighted by the self-attention mechanism. The self-attention mechanism calculates the importance of each part in the feature map to generate an attention weight matrix, and weights the feature map based on the matrix to generate a weighted feature map. The weighted feature map can better highlight key features and suppress secondary information, thereby increasing the model's attention to important information. The weighted feature map is input into the long short-term memory network for temporal feature extraction. The long short-term memory network can effectively handle the long-range dependency problem of sequence data, extract temporal correlation features, and reflect the important patterns that change over time in the meeting. The temporal correlation features are processed by residual connection and layer normalization. Residual connection can prevent the gradient vanishing problem in deep networks, while layer normalization can stabilize the network training process and obtain optimized feature representation. Based on the optimized feature representation, a multi-head attention mechanism is applied to perform multi-scale feature processing. The multi-head attention mechanism can analyze features at different scales, capture the diversity and complexity of features, and obtain multi-scale features. The multi-scale features are input into the fully connected layer for processing to obtain the initial encryption requirement score and candidate encryption parameters. For the initial encryption requirement score, the softmax function is used to calculate and generate the final dynamic encryption requirement score. The softmax function can standardize the score into a probability distribution, thereby more intuitively reflecting the relative importance of each encryption requirement. Based on the dynamic encryption requirement score and candidate encryption parameters, a strategy generation network is constructed to generate multiple candidate encryption strategies. These strategies are different encryption schemes for conference content, which can be selected and applied according to actual needs, thereby ensuring the security and flexibility of conference content.

[0035] 103. Construct a two-level model based on dynamic encryption requirement scoring and multiple candidate encryption strategies;

[0036] Specifically, the dynamic encryption requirement score is converted into a fuzzy set. By converting the explicit encryption requirement score into a fuzzy encryption requirement matrix, the uncertainty and ambiguity in the encryption requirement can be better handled, making the selection of encryption strategy more flexible and adaptable. Based on the fuzzy encryption requirement matrix, multiple candidate encryption strategies are weighted summed to generate the initial display content flow. The genetic algorithm is applied to perform genetic crossover and mutation operations on the initial display content flow to generate multiple candidate display content flows. The crossover and mutation operations of the genetic algorithm imitate the process of natural selection, allowing these candidate display content flows to explore a wider range of possibilities and increase the chance of finding the optimal solution. Multiple candidate display content flows are evaluated, and the optimal display content flow is selected through a series of evaluation criteria (such as security, latency, user experience, etc.). The upper optimization objective function is constructed based on the optimal display content flow. The optimization objective function includes two main items: security maximization and latency minimization. Security maximization ensures the confidentiality of the conference content, while latency minimization ensures the real-time and fluency of the content. At the same time, the user role information is hierarchically analyzed to obtain the role importance weight vector. Hierarchical analysis is an effective decision-making method that quantifies the importance of different user roles through hierarchical analysis. Based on the obtained role importance weight vector, a role-based access control (RBAC) permission matrix is ​​constructed, and an initial permission allocation scheme is generated. The RBAC model ensures that key roles can obtain appropriate access rights by considering the importance of different user roles. After the initial permission allocation scheme is generated, iterative optimization is performed. According to the actual situation, the permission allocation is continuously adjusted to make the permission allocation scheme more reasonable and efficient, and an improved permission allocation scheme is obtained. Based on the improved permission allocation scheme, the lower-level constraints are constructed, mainly including role inheritance and mutual exclusion. The role inheritance relationship ensures that permissions can be transferred according to the hierarchical relationship of roles, while the mutual exclusion relationship prevents permission conflicts and ensures the security and consistency of the system. The upper-level optimization objective function and the lower-level constraints are combined to form a complete two-level model. By comprehensively considering security, real-time and user permissions, the two-level model ensures that the encryption and display process of conference content meets security standards and can flexibly respond to complex conference scenario requirements.

[0037] 104. The global optimal encryption strategy is obtained by solving the two-level model through equivalent complementary constraints;

[0038] Specifically, the upper objective function of the two-level model is Lagrangian relaxed, and the complex constraints of the upper level are relaxed into a more tractable form, which simplifies the difficulty of solving the problem and obtains the equivalent lower level problem. The equivalent lower level problem is merged with the lower level constraint conditions to generate a single-level equivalent model. The single-level equivalent model simplifies the two-level model into a more unified problem framework by integrating the upper and lower level problems and constraints. In this model, there are nonlinear terms, which usually increase the complexity of problem solving. In order to solve this problem, the nonlinear terms are envelope linearized to obtain a linearized constraint set. Envelope linearization is a commonly used method that simplifies the problem solving process by approximating nonlinear relationships as linear relationships. The logical constraints involved in the linearized constraint set are transformed into mixed integer constraints. Mixed integer constraints enable the model to handle continuous variables and discrete variables at the same time, providing flexibility and accuracy for solving. Based on the mixed integer constraints, an initial mixed integer programming model is constructed. The variables of the initial mixed integer programming model are fixed and unnecessary constraints are eliminated to obtain a simplified model. The simplified model is cut, and the feasible solution space is reduced by introducing effective inequality constraints, thus converting the problem into a mixed integer quadratic programming problem. Mixed integer quadratic programming problems can better reflect the actual characteristics of the problem by introducing quadratic objective functions or constraints. A tree search algorithm is applied to systematically explore the solution space to obtain a feasible solution set. The tree search algorithm can ensure that the optimal solution or a near-optimal solution is found by systematically searching all possible solutions. Local search is performed on the feasible solution set to fine-tune and optimize the solution to obtain an improved solution. Local search can explore better solutions within the neighborhood of the solution to avoid falling into the local optimum. Sensitivity analysis is performed on the improved solution. By analyzing the sensitivity of the solution to different parameters, the robustness and adaptability of the solution are evaluated, and finally the global optimal encryption strategy is obtained.

[0039] 105. Based on the global optimal encryption strategy, the LED display content is scanned and graded in real time to obtain a multi-element content set with dynamic sensitivity levels;

[0040] Specifically, the parameters of the global optimal encryption strategy are parsed to obtain the sensitivity evaluation function and the set of graded thresholds. Real-time image acquisition is performed on the LED display screen. By capturing each frame on the display screen, the original display content frame sequence is generated, which contains all the visual information of the current display content. Image enhancement is performed on the original display content frame sequence to improve the clarity and contrast of the image and generate an optimized image sequence. Based on the optimized image sequence, content segmentation is performed to divide the image into multiple content element areas. Each area corresponds to a part of the display content, such as text, graphics or images. Object detection and semantic segmentation are performed on the content element area. Object detection is used to identify specific objects in the content, while semantic segmentation assigns semantic labels to different parts of the image to generate content element type labels. Text extraction is performed on text content elements to generate text content data. Semantic analysis is performed on text content data and non-text content elements. Semantic feature vectors are generated through natural language processing and deep learning technology to quantify the semantic information of content elements. The semantic feature vectors are input into the sensitivity evaluation function to calculate the sensitivity score of each content element. The sensitivity score reflects the potential sensitivity of content elements and determines the encryption level they require during display. The initial sensitivity level is analyzed for contextual information and time series relationships to more accurately reflect the actual sensitivity of the content and obtain a dynamic sensitivity level. Contextual information takes into account the correlation between content, while time series analysis ensures that the sensitivity is dynamically adjusted over time. The dynamic sensitivity level is quantified based on the set of graded thresholds to obtain discrete sensitivity levels and generate a multi-element content set with different sensitivity levels. Each element in the content set is assigned an appropriate sensitivity level to ensure that it can be properly encrypted and protected according to its sensitivity during display, thereby effectively balancing information security and user experience.

[0041] 106. Generate a multi-level encrypted display data stream based on the dynamic sensitivity level of the multi-element content set and perform decryption control to obtain decrypted display content accessible to authorized users.

[0042] Specifically, according to the dynamic sensitivity level and content type, the multi-element content set is clustered and analyzed, and the content with similar sensitivity and type is clustered together to form content encryption groups. Each encryption group generates a corresponding key through key distribution technology according to its characteristics and security requirements to build a key pool. Each key in the key pool corresponds to an encryption group to ensure the independence and security of encryption operations. Homomorphic encryption is performed on each encryption group. Homomorphic encryption allows computing operations to be performed directly on encrypted data and content processing to be performed while keeping the data encrypted. Primary encrypted content is obtained through encryption. In order to ensure that the data can be effectively verified during transmission and storage, metadata is embedded in the primary encrypted content to generate encrypted content with hidden information. Metadata includes information about the encryption algorithm, key identifiers, and other content-related verification information. An encrypted content hash is generated based on the encrypted content with hidden information and stored in a distributed ledger. Distributed ledger technology provides an untamperable storage environment so that any unauthorized modification of the encrypted content will be detected, and tamper-proof encrypted content is obtained. At the same time, a digital watermark corresponding to the conference information and timestamp is generated based on a dynamic watermark generation algorithm. Digital watermarks can be embedded in image or video data without affecting their visual effects, but can provide tracking and verification functions when the content is copied or tampered with without authorization. The digital watermark is fused with the tamper-proof encrypted content at the pixel level to obtain the watermark encrypted content. In order to reduce the bandwidth consumption during transmission, the watermark encrypted content is compressed to generate a low-bandwidth encrypted data stream. Compression technology can effectively reduce the amount of data while retaining the necessary information quality. On this basis, redundant data is added to enhance the error recovery capability of the data during transmission and generate an encrypted data packet with error recovery capability. Even if this data packet encounters partial data loss or damage in the network environment, it can be restored through redundant information to ensure the integrity and reliability of the data. Dynamic routing and transmission optimization of encrypted data packets ensure that the data stream can choose the best path during transmission, reduce delay and packet loss rate, and form a multi-level encrypted display data stream. Decryption control of the multi-level encrypted display data stream. By using the keys in the key pool, the data stream is decrypted layer by layer to restore the original content. The decryption process is strictly controlled according to the permissions of authorized users, and only users with corresponding permissions can access specific decrypted content. Decrypted display content accessible to authorized users ensures the security and privacy of conference content during transmission and display.

[0043] The biometric data provided by the user is collected and preprocessed, and the fingerprint, face, iris and other biometric features are enhanced, denoised and standardized to generate biometric authentication results. A time-synchronized one-time password algorithm is used to generate a dynamic password, which is verified in combination with the user's device information to obtain a dynamic password verification result. The dynamic verification method ensures that even if the biometrics are stolen, the user's security can still be protected through dual verification of time and device. The user's behavior pattern is analyzed in real time, and anomaly detection algorithms are used to identify anomalies in user behavior and generate a behavioral authentication score. The score is based on the user's operating habits, usage frequency and other behavioral characteristics to detect whether there is a potential security threat. The biometric authentication results, dynamic password verification results and behavioral authentication scores are input into the fuzzy comprehensive evaluation model for comprehensive evaluation to obtain the multi-factor authentication credibility, which reflects the user's current identity authentication strength and security. The multi-factor authentication credibility is weighted and fused with the user's RBAC role information. The RBAC role information reflects the user's authority and responsibilities in the system. The user's access trust level is obtained through comprehensive calculation by the trust evaluation algorithm. The access trust level determines the data range and decryption authority that the user can access. Based on this trust level, homomorphic decryption technology is applied to partially decrypt the multi-level encrypted display data stream to obtain an intermediate data stream. Homomorphic decryption allows calculations to be performed while the data is encrypted, ensuring that data security is maintained during the decryption process. After the intermediate data stream is generated, attribute access control is performed to ensure that only the data subset that meets the user's permissions can be accessed. Attribute access control performs refined permission management based on the user's role, the sensitivity of the data content, and the access context. The system distributes decryption keys and generates distributed decryption key shares through a secure multi-party computing protocol. Combined with the data subset that meets the user's permissions and the distributed decryption key shares, the final decryption operation is performed to generate decrypted data. The decrypted data is securely rendered through isolated computing. The isolated computing environment ensures that the data will not be accessed by unauthorized users or programs during the rendering process, thereby improving the security of the system. The securely rendered decrypted data generates decrypted display content that is accessible to authorized users, ensuring that users can safely and accurately view sensitive information in the meeting without unauthorized disclosure or access.

[0044] Perform memory isolation on the decrypted data, prevent data leakage and unauthorized access by dividing independent memory areas, and generate isolated memory areas. Set access control lists for the isolated memory areas to create a protected data buffer. Virtualize the graphics processing unit (GPU) based on the protected data buffer to generate a virtual GPU instance. The virtual GPU instance allows multiple independent rendering tasks to be executed securely on the same physical GPU and avoid mutual interference. Configure security policies for the virtual GPU instance to create a GPU environment with isolation characteristics, ensure that each rendering task runs in an isolated environment, and prevent data leakage and attacks. Encapsulate the GPU environment with isolation characteristics into a trusted execution environment to obtain a trusted rendering container. This container ensures the integrity and confidentiality of the rendering process through hardware-supported isolation technologies such as Intel SGX or AMD SEV. Perform integrity measurement on the trusted rendering container and calculate the trusted measurement value, which is used to verify the trustworthiness of the rendering environment and ensure that the rendering instructions have not been tampered with or destroyed. Perform credibility verification on the rendering instructions based on the trusted measurement value to generate a trusted rendering instruction set. The trusted rendering instruction set is obfuscated at the instruction level. The obfuscation technology makes the instruction sequence difficult to be reverse engineered and analyzed, thereby improving security. The obfuscated rendering instructions are dynamically reordered to generate a random execution sequence, making the order of rendering instructions more random when executed, preventing attackers from predicting the order of instruction execution. The GPU shader is dynamically compiled according to the random execution sequence to generate a secure shader program. Dynamic compilation makes the shader code different each time it is executed, increasing the difficulty of cracking. The code integrity of the secure shader program is checked to ensure that the compiled shader code has not been tampered with, and an integrity check result is generated. Based on this check result, the rendering pipeline is configured to obtain a secure rendering pipeline. According to the secure rendering pipeline, the geometric data is transformed and rasterized to generate pixel-level rendering data, and the initial rendering result is obtained through pixel shading technology. The initial rendering result is encrypted by frame buffer object to generate an encrypted frame buffer. The encrypted frame buffer is securely transmitted to generate encrypted data on the display end. To ensure the security of the display end, the display driver is isolated according to the encrypted data on the display end, and an isolated display channel is generated to prevent malware or unauthorized programs from stealing display content. In the isolated display channel, the data is decrypted using dynamic key decryption technology to obtain decrypted display data. The decrypted display data is displayed in a timing control to ensure that the display content is presented in the correct order and timing, and finally the decrypted display content accessible to authorized users is obtained.

[0045] In the embodiment of the present invention, by real-time collection and analysis of conference scene data, combined with multi-layer neural network processing, the method can accurately identify conference features, content features and user role information, adopt a two-level model and a mixed integer quadratic programming method, and achieve global optimization of security and delay. At the same time, through flexible permission allocation based on RBAC, the LED display content is scanned in real time and dynamically graded for sensitivity. Combined with layered differential encryption and dynamic watermark embedding technology, refined protection of different types of content is achieved, and multi-factor authentication and trust evaluation mechanisms are introduced. Combined with homomorphic decryption and secure multi-party computing technology, flexible decryption control is achieved while ensuring data security, effectively preventing unauthorized access. Through isolated computing and secure rendering technology, including memory isolation, GPU virtualization, trusted execution environment and other measures, the security of decrypted display content is ensured, and potential security threats in the rendering process are effectively prevented. The anti-tampering ability and traceability of the content are enhanced, and the transmission efficiency of encrypted data is optimized through compression and redundant data addition technology. At the same time, the error recovery ability of the system is improved, and the reliability of the display system is enhanced.

[0046] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0047] Perform cluster analysis on the conference type data to obtain a conference type feature matrix, and perform feature extraction on the real-time conference data based on the conference type feature matrix to obtain a conference feature vector;

[0048] Perform optical character recognition and image segmentation on the LED display screen content to obtain text data sets and image data sets;

[0049] Perform vectorization on the text data set to obtain a text feature matrix, and perform feature extraction on the image data set to obtain an image feature matrix;

[0050] Perform tensor fusion on the text feature matrix and the image feature matrix to obtain the content feature vector;

[0051] Perform time series analysis and frequent pattern mining on user behavior logs to obtain user behavior feature sets, and build user similarity graphs based on the user behavior feature sets;

[0052] Spectral clustering is performed on the user similarity graph to obtain initial user role groupings, and the initial user role groupings are dynamically adjusted to obtain user role information.

[0053] Specifically, through cluster analysis of a large amount of conference type data, a representative conference type feature matrix is ​​summarized. Using clustering algorithms such as K-means or hierarchical clustering, different types of conference data are clustered together. Let X = {x1, x2, …, x n} represents n conference samples, where each x i All contain features of multiple dimensions (such as meeting time, number of participants, meeting topic, etc.). Through the clustering algorithm, these samples are divided into k clusters, each cluster represents a type of meeting. The goal is to minimize the sum of the squares of the distances from the samples in each cluster to the center of the cluster, expressed as:

[0054]

[0055] Among them, μ j Represents cluster C j In this way, the generated cluster center (i.e., the eigenvector μ j ) constitutes the conference type feature matrix M, which represents the typical features of different types of meetings. Feature extraction is performed on real-time conference data based on the conference type feature matrix. For example, in an actual conference scenario, various information about the meeting is collected in real time, such as the number of speeches made by participants, the frequency of topic changes, and the interaction status. These real-time data are compared with the conference type feature matrix M, and the similarity is calculated or through matrix transformation to extract the conference feature vector v, which can describe the characteristics of the current meeting in a high-dimensional space. Perform optical character recognition (OCR) and image segmentation on the content on the LED display. Through OCR technology, the text information in the displayed content is extracted into a processable text dataset T, while the image segmentation technology extracts the picture part of the displayed content into an image dataset I. These datasets represent the important text information and visual content in the meeting, respectively. Vectorize the text dataset T. Through word embedding technology, such as Word2Vec or TF-IDF, each word or phrase is mapped to a vector space to generate a text feature matrix T f . Matrix T f It can represent the semantic information of text data. At the same time, the image dataset I is feature extracted, and the image is converted into an image feature matrix I using deep learning models such as convolutional neural networks. f , which can capture the visual features and structural information of the image. f and the image feature matrix I f Tensor fusion is performed to obtain a comprehensive content feature vector. Tensor fusion captures the association between text and images by jointly modeling data of different modalities in high-dimensional space. Time series analysis and frequent pattern mining are performed on the data of user behavior logs. User behavior logs usually include the sequence of user operations in the system, such as clicks, browsing, and speaking. Through time series analysis, the dynamic patterns of user behavior are captured, and through frequent pattern mining, high-frequency patterns in user behavior are identified, such as a user always checking the meeting agenda before opening related materials. Through these analyses, the user behavior feature set U is generated. f, this set of features captures the typical behavior patterns of users. Based on the user behavior feature set U f , construct a user similarity graph G. The nodes of the graph represent different users, and the edge weights represent the similarity between users. The similarity can be calculated based on the cosine similarity of user behavior characteristics or other similarity measurement methods. For example, if the operations performed by two users in a meeting are very similar, the similarity between them is high and the edge weights are also large. Perform spectral clustering on the user similarity graph G to obtain the initial user role grouping. Spectral clustering analyzes the eigenvectors of the Laplacian matrix of the graph and assigns users with high similarity to the same group to form the initial user role grouping R0. This role grouping can represent different roles in the meeting, such as speakers, listeners, hosts, etc. In order to adapt to the dynamic meeting environment, the initial user role grouping R0 is dynamically adjusted. Combined with real-time data, by analyzing the new behavior or role changes of users in the meeting, their grouping is adjusted, and finally the user role information R is obtained.

[0056] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0057] Perform tensor combination on the conference feature vector, content feature vector and user role information to obtain an input tensor;

[0058] Based on the three-dimensional convolutional neural network, the input tensor is feature extracted to obtain a high-dimensional feature map, and the high-dimensional feature map is weighted by the self-attention mechanism to obtain the attention weight matrix and the weighted feature map;

[0059] The weighted feature map is input into the long short-term memory network to extract the time series features, obtain the time series correlation features, and the time series correlation features are residually connected and layer normalized to obtain the optimized feature representation;

[0060] Based on the multi-head attention mechanism, multi-scale feature processing is performed on the optimized feature representation to obtain multi-scale features, and the multi-scale features are input into the fully connected layer to obtain the initial encryption requirement score and candidate encryption parameters;

[0061] A softmax function operation is performed on the initial encryption requirement score to obtain a dynamic encryption requirement score, and a strategy generation network is constructed based on the candidate encryption parameters to generate multiple candidate encryption strategies.

[0062] Specifically, the conference feature vector, content feature vector and user role information are combined into a tensor. Conference feature vector v m , content feature vector v c and user role information v uThey represent the overall characteristics of the meeting, the characteristics of the display content, and the role information of the participants. By combining these vectors into tensors, a three-dimensional input tensor T is generated. The input tensor T can be expressed as in Represents a tensor product operation. The input tensor combines various key information of the meeting to form a multi-dimensional data representation that can be further processed in spatial and temporal dimensions. A three-dimensional convolutional neural network (3D-CNN) is used to extract features from the input tensor T. The convolution kernel of 3D-CNN slides in three dimensions and performs convolution operations on local areas of the input tensor to capture the complex interactions between different information. After the convolution operation, it is processed by an activation function (such as ReLU) to obtain a high-dimensional feature map F, which represents the high-level features of the input data. Each position in the high-dimensional feature map F reflects the feature combination of the original input in the local area. The high-dimensional feature map F is weighted by the self-attention mechanism. The self-attention mechanism generates an attention weight matrix A by calculating the importance of each position in the feature map. The attention weight matrix A can be expressed as

[0063]

[0064] Among them, Q i is the query vector, K j is the key vector, Q i ·K jrepresents the dot product operation of the query vector and the key vector. The calculated value is normalized by the softmax function to obtain the attention distribution. The attention weight matrix A reflects which parts of the feature map are more important. The attention weight matrix A is applied to the high-dimensional feature map F to obtain the weighted feature map F′. The weighted feature map F′ is input into the long short-term memory network (LSTM) for temporal feature extraction. LSTM is a neural network suitable for processing sequence data, which can capture the temporal relationship and dependency in the input data. Through the memory unit and forget gate mechanism of LSTM, the temporal correlation features S can be extracted from the weighted feature map F′. These features S reflect the dynamic pattern of the change of meeting, content and user role information over time. The temporal correlation features S are subjected to residual connection and layer normalization. Residual connection introduces skip connection in the deep network, making it easier for the network to propagate gradients during training and preventing the gradient vanishing problem. Layer normalization reduces internal covariate shift and stabilizes the training process by normalizing the output of each layer. After residual connection and layer normalization, the optimized feature representation S′ is obtained. The optimized feature representation is processed with multi-scale features based on the multi-head attention mechanism. The multi-head attention mechanism performs attention calculations on multiple different attention heads in parallel. Each attention head focuses on different parts of the input data or information at different scales, so that the model can analyze input features from multiple perspectives at the same time and generate multi-scale features. The multi-scale features are input into the fully connected layer for processing to generate an initial encryption requirement score and candidate encryption parameters. The initial encryption requirement score indicates the degree of need to encrypt the data in the current context, while the candidate encryption parameters provide different encryption strategies or settings. The initial encryption requirement score P is operated with the softmax function to obtain the dynamic encryption requirement score P′. The form of the softmax function is:

[0065]

[0066] Among them, P iRepresents the i-th score in the initial encryption requirement score. Through the softmax function operation, all scores are normalized into a probability distribution, reflecting the relative importance of each encryption requirement. The final dynamic encryption requirement score P′ can intuitively indicate which encryption strategies are more appropriate under the current conditions. Based on the dynamic encryption requirement score P′ and the candidate encryption parameters C, a policy generation network is constructed to generate multiple candidate encryption strategies. These strategies are generated according to different requirements and conditions to ensure that the conference content can receive the most suitable encryption protection during transmission and storage. For example, if the dynamic encryption requirement score P′ indicates that the current content is highly sensitive, the policy generation network may select a more stringent encryption algorithm, such as AES-256, to ensure data security. On the contrary, if the content is less sensitive, a lighter encryption scheme may be selected to optimize performance.

[0067] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0068] Performing fuzzy set transformation on the dynamic encryption requirement score to obtain a fuzzy encryption requirement matrix, and performing weighted summation on multiple candidate encryption strategies based on the fuzzy encryption requirement matrix to obtain an initial display content stream;

[0069] Performing genetic crossover and mutation operations on the initial display content flow to obtain multiple candidate display content flows, and evaluating the multiple candidate display content flows to obtain an optimal display content flow;

[0070] Based on the optimal display content flow, the upper layer optimization objective function is constructed, including the security maximization and delay minimization items, and the user role information is hierarchically analyzed to obtain the role importance weight vector;

[0071] Based on the role importance weight vector, the RBAC permission matrix is ​​constructed to obtain the initial permission allocation scheme, and the initial permission allocation scheme is iteratively optimized to obtain an improved permission allocation scheme;

[0072] Based on the improved permission allocation scheme, the lower-level constraints are constructed, including role inheritance and mutual exclusion, and the upper-level optimization objective function and the lower-level constraints are combined to obtain a two-level model.

[0073] Specifically, the dynamic encryption requirement score P′ is converted into a fuzzy set, and the encryption requirement score is converted into a fuzzy encryption requirement matrix F, so that the system can better handle uncertainty and ambiguity. In fuzzy set theory, the score can be converted into a fuzzy set, and each encryption requirement score is mapped to a membership value in the fuzzy set through the membership function μ(x). The membership value represents the applicability of a certain policy under a given score. For example, an encryption requirement score can be converted into a fuzzy set through the following membership function:

[0074]

[0075] Where x represents the encryption requirement score, x0 and x1 are the thresholds in the conversion process, and μ(x) is the membership value. Through the conversion, the fuzzy encryption requirement matrix F obtained contains the fuzzy representation of each encryption requirement score. Based on the fuzzy encryption requirement matrix F, multiple candidate encryption strategies are weighted and summed to generate the initial display content stream S0. Assume that the candidate encryption strategy is C = {c1, c2, …, c n}, the weight of each strategy is determined by the value in the fuzzy encryption requirement matrix, and the initial display content flow can be expressed as:

[0076]

[0077] Among them, F i For the fuzzy encryption requirement matrix and strategy c i The relevant membership degree, S0 is the initial display content stream after weighted summation, representing the display content generated after comprehensive consideration of various encryption strategies and requirements. The initial display content stream S0 is operated by genetic algorithm, including genetic crossover and mutation, to generate a more diverse set of candidate display content streams S1, S2, ..., S m The genetic crossover operation combines the characteristics of two or more candidate content flows to generate a new candidate display content flow by simulating the gene exchange process in natural inheritance. The mutation operation randomly changes some characteristics in the candidate content flow to explore a wider solution space. After evaluating multiple candidate display content flows, the optimal display content flow S is selected based on preset evaluation indicators (such as security, latency, user experience, etc.). * For example, in an actual conference environment, it is important to select a display content stream that provides both high security and low latency. * As the final output. Based on the optimal display content stream S * , construct an upper-level optimization objective function, which includes two parts: maximizing security and minimizing latency. The goal of maximizing security can be achieved by increasing encryption strength or improving data protection level, while minimizing latency requires minimizing processing time during display content streaming and rendering. Assume that security and latency are represented by S(S * ) and L(S * ), the upper optimization objective function can be expressed as:

[0078] max(α·S(S * )-β·L(S * ));

[0079] Among them, α and β are weight coefficients used to balance the importance of security and latency. By optimizing this objective function, the display content stream that best suits the current meeting scenario is found. At the same time, a hierarchical analysis is performed on the user role information. The hierarchical analysis method is a tool commonly used in decision analysis to help determine the importance weight of each user role. By constructing a pairwise comparison matrix, the importance weight vector w = {w1, w2, …, w k}, where w i Represents the importance of the i-th role. Based on the role importance weight vector w, the role-based access control (RBAC) permission matrix R is constructed to generate the initial permission allocation scheme. The RBAC model ensures that each role can exercise its authority in the system by defining the relationship between roles and permissions. After the initial permission allocation scheme is generated, the permission allocation scheme is adjusted through iterative optimization, considering the dynamically changing role importance and system status, and the improved permission allocation scheme R is obtained. * . Construct lower-level constraints, which include role inheritance and mutual exclusion. Role inheritance ensures that the permissions of the upper-level role can be passed to the lower-level role, while mutual exclusion prevents certain conflicting roles from having certain permissions at the same time. These constraints define the boundary conditions for permission allocation to ensure the security and consistency of the system. Combining the upper-level optimization objective function with the lower-level constraints, a complete two-level model is obtained. The two-level model can reasonably allocate user permissions while considering security and latency, ensuring the secure transmission and display of conference content, while meeting user access needs and system performance requirements.

[0080] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0081] Perform Lagrangian relaxation on the upper objective function of the two-level model to obtain the relaxed upper problem, and transform the relaxed upper problem to obtain the equivalent lower problem;

[0082] The equivalent lower-level problem and the lower-level constraint conditions are combined to obtain a single-level equivalent model, and the nonlinear terms in the single-level equivalent model are envelope linearized to obtain a linearized constraint set;

[0083] The logical constraints in the linearized constraint set are converted into mixed integer constraints to obtain an initial mixed integer programming model, and the variables of the initial mixed integer programming model are fixed and the constraints are eliminated to obtain a simplified model;

[0084] Cut the simplified model to generate valid inequality constraints, obtain a mixed integer quadratic programming problem, and perform tree search on the mixed integer quadratic programming problem to obtain a feasible solution set;

[0085] A local search is performed on the feasible solution set to obtain an improved solution, and a sensitivity analysis is performed on the improved solution to obtain the global optimal encryption strategy.

[0086] Specifically, the upper-level objective function is usually used to optimize some macro indicators, such as maximizing security and minimizing latency, while the lower-level constraints involve specific operational constraints, such as role inheritance and mutual exclusion. Lagrangian relaxation is performed on the upper-level objective function of the two-level model. Lagrangian relaxation is a process of transforming a constrained optimization problem into an unconstrained optimization problem by integrating constraints into the objective function. Assuming that the upper-level objective function is f(x) and its constraint is g(x)≤0, the Lagrangian multiplier λ is introduced to construct the Lagrangian function:

[0087]

[0088] In this way, the original constrained optimization problem can be relaxed into a new unconstrained optimization problem, which enables the system to handle constraints more flexibly, especially in complex multi-level models. By relaxing, a relaxed upper-level problem is obtained. The relaxed upper-level problem is transformed into an equivalent lower-level problem. The equivalent lower-level problem is usually easier to handle than the original upper-level problem, especially when multiple complex constraints need to be considered. For example, when optimizing security and delay, the problem can be expressed as a combination of a simpler objective function and corresponding constraints through appropriate transformations. The equivalent lower-level problem is combined with the lower-level constraints to construct a single-level equivalent model. The nonlinear terms are linearized through envelope linearization technology to obtain a linearized constraint set. Envelope linearization usually approximates the original nonlinear function by constructing a linear envelope, so as to bring it into the scope of linear programming. When the logical constraints in the linearized constraint set need to be processed, they are converted into mixed integer constraints. This conversion usually obtains an initial mixed integer programming model (MIP) by expressing the logical conditions (such as "if-then" relations) as a combination of integer variables and linear inequalities. The MIP model contains both linear relationships of continuous variables and logical constraints of discrete variables, so that the model can handle more complex decision problems. Fix the variables in the MIP model and eliminate unnecessary constraints. For example, for some variables, if their values ​​can be determined under certain circumstances, they can be fixed to reduce the complexity of the model. Cut the simplified model to generate valid inequality constraints and transform the problem into a mixed integer quadratic programming problem. Cutting technology is usually used in the branch and bound process of optimization problems. By adding inequality constraints, infeasible solutions are excluded while retaining the feasible solution space. Perform tree search on mixed integer quadratic programming problems. The tree search algorithm systematically explores every possible solution in the solution space by constructing a decision tree. Through branch and bound and backtracking, the tree search algorithm can gradually narrow the solution space and finally obtain a feasible solution set. The feasible solution set contains all solutions that meet the constraints, and these solutions are close to the optimal solution to a certain extent. Perform local search on the feasible solution set, fine-tune and optimize the solutions in the solution set, and obtain a better solution. Local search obtains improved solutions by carefully exploring local areas of the solution space. Sensitivity analysis is performed on the improved solutions to evaluate the robustness and reliability of the solutions under different conditions. By adjusting key parameters in the model and observing the changes in the solutions, sensitivity analysis can help the system understand the stability of the solutions and their sensitivity to parameter changes. Based on the results of the sensitivity analysis, the global optimal encryption strategy is finally determined, which is not only optimal under the current conditions, but also able to cope with possible changes in the future.

[0089] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0090] The parameters of the global optimal encryption strategy are parsed to obtain the sensitivity evaluation function and the classification threshold set, and the real-time image acquisition of the LED display screen is performed to obtain the original display content frame sequence;

[0091] Performing image enhancement on the original display content frame sequence to obtain an optimized image sequence, and performing content segmentation on the optimized image sequence to obtain multiple content element regions;

[0092] Perform object detection and semantic segmentation on the content element area to obtain the content element type label, and perform text extraction on the text content elements to obtain text content data;

[0093] Perform semantic analysis on text content data and non-text content elements to obtain semantic feature vectors, and input the semantic feature vectors into the sensitivity evaluation function to obtain an initial sensitivity level by calculating the sensitivity score;

[0094] The context information and time series relationship of the initial sensitivity level are analyzed to obtain a dynamic sensitivity level, and the dynamic sensitivity level is quantified according to a grading threshold set to obtain a multi-element content set with discrete sensitivity levels.

[0095] Specifically, key parameters are extracted from the global optimal encryption strategy, which determine how to evaluate the sensitivity of the content and how to grade these evaluation values. Assume that the sensitivity evaluation function S(v) is used to calculate the sensitivity score of the content, where v represents the input feature vector. The function can be defined as:

[0096] S(v)=w T v+b;

[0097] Where w is the weight vector, which indicates the contribution of each feature to the sensitivity, and b is the bias term, which is used to adjust the baseline of the overall sensitivity score. Through the analytical strategy, the hierarchical threshold set {T1, T2, …, T n}, these thresholds are used to quantify the continuous sensitivity scores into different discrete levels. Real-time image acquisition is performed on the LED display screen, capturing each frame of the displayed content and generating a sequence I = {I1, I2, ..., I t}, where I t represents the image frame collected at time point t. Image enhancement is performed on the image frame to improve the quality and clarity of the image. Assume that the image enhancement function is Enhance(I t ), then the enhanced image sequence {I′1,I′2,…,I′ t} can be obtained by the following method: I′ t =α·I t+β, where α and β represent the contrast adjustment coefficient and the brightness adjustment coefficient respectively. Through image enhancement, a clearer image sequence is obtained. The optimized image sequence is subjected to content segmentation to divide the image into multiple content element areas. Content segmentation technology divides the image into several areas with independent semantics or functions by analyzing features such as edges, colors and textures in the image. For example, in a conference presentation page, content segmentation may separate the title area, text area, chart area and background area in the slide. Target detection and semantic segmentation are performed on the segmented content element areas to identify specific objects in the area and assign a type label to each content element. For text content, optical character recognition (OCR) technology is used to extract text data. Semantic analysis is performed on text data and non-text content to extract deep semantic features of each content element and generate a semantic feature vector. The text data is processed by natural language processing technology to extract semantic information of words, phrases and sentences; for image content, deep learning methods such as convolutional neural networks are used to extract visual features. The semantic feature vector v t,i Input into the sensitivity evaluation function S(v) to calculate the initial sensitivity score of each content element:

[0098] S t,i =w T v t,i +b;

[0099] Where S t,i Represents the initial sensitivity score of the ith region in the tth frame. The sensitivity level is adjusted based on the context information and the time series relationship. The dynamic sensitivity level is compared with the grading threshold set {T1, T2, T3} and quantized to obtain the discrete sensitivity level L t,i :

[0100]

[0101] In this way, all content elements in region R t,i They are assigned clear sensitivity levels, which are used in the subsequent development of encryption and access control policies to ensure that the content is appropriately protected during transmission and display.

[0102] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0103] According to the dynamic sensitivity level and content type, cluster analysis is performed on the multi-element content set to obtain content encryption groups, and a key is generated for each encryption group based on the key distribution technology to obtain a key pool;

[0104] Perform homomorphic encryption on each encrypted group to obtain primary encrypted content, and embed metadata into the primary encrypted content to obtain encrypted content with hidden information;

[0105] Generate an encrypted content hash based on the encrypted content with hidden information and store it in a distributed ledger to obtain tamper-proof encrypted content;

[0106] Generate digital watermarks corresponding to conference information and timestamps based on a dynamic watermark generation algorithm, and fuse the tamper-proof encrypted content and digital watermark at the pixel level to obtain watermark encrypted content;

[0107] Compressing the watermark encrypted content to obtain a low-bandwidth encrypted data stream, and adding redundant data to the low-bandwidth encrypted data stream to obtain an encrypted data packet with error recovery capability;

[0108] Dynamically route and optimize transmission of encrypted data packets to obtain multi-level encrypted display data streams;

[0109] Decryption control is performed on the multi-level encrypted display data stream to obtain the decrypted display content accessible to authorized users.

[0110] Specifically, based on the dynamic sensitivity level and content type of the content, a cluster analysis is performed on the multi-element content set. Through the clustering algorithm, the content is classified according to these two factors to form encrypted groups. Assume that the content set X contains multiple elements {x1, x2, …, x n}, each element x i With specific sensitivity levels i and content type i The goal of clustering is to minimize the differences within the group and maximize the differences between groups. The formula can be expressed as:

[0111]

[0112] Among them, C k is the kth cluster, and Represents cluster C k The sensitivity level and content type of all elements in the cluster are averaged. α and β are weight parameters used to balance the influence of sensitivity level and content type in clustering. After clustering, a key is generated for each encrypted group based on key distribution technology, and these keys form a key pool. Assume that there are K encrypted groups in total, and each group C k Corresponding to a key K k , the key pool K can be expressed as {K1,K2,…,K K}. Homomorphic encryption technology is applied to each encrypted group to generate primary encrypted content. Homomorphic encryption allows calculations to be performed in an encrypted state, ensuring the privacy of the data while not affecting the operability of the data. For example, if the original content m i Encryption is performed, and the encrypted content c is obtained i Meet c i =E(m i ,K k ), where E represents the encryption function. Homomorphic encryption allows the system to operate on these encrypted contents without decryption, ensuring the security of the data. Metadata is embedded in the primary encrypted content to form encrypted content with hidden information. These metadata may include information such as the creation time of the content, the identity of the creator, and the importance level of the content. The embedding of these metadata can be achieved through information hiding techniques, such as based on LSB (least significant bit) or more complex steganography techniques. In order to ensure the integrity and tamper-proofness of the encrypted content, a hash value of the encrypted content is generated and stored in a distributed ledger. The hash value is obtained by encrypting the content c = {c1, c2, …, c n}, usually using SHA-256 or other cryptographic hash algorithms. The hash value h=H(c) generated by the hash function H(·) is stored in the distributed ledger, making the encrypted content tamper-proof. Based on the dynamic watermark generation algorithm, a digital watermark corresponding to the meeting information and timestamp is generated. A digital watermark is a unique identifier that combines meeting information and timestamps to ensure the time sensitivity and authenticity of the source of the data. The generated digital watermark is pixel-wise fused with the tamper-proof encrypted content, and the concealment and robustness of the watermark are guaranteed by modulating the pixel values. The watermarked encrypted content is compressed to generate a low-bandwidth encrypted data stream, reducing the amount of data during transmission while retaining the integrity and security of the data. The compression algorithm can use standard compression methods such as H.264 or JPEG2000. In order to enhance the reliability of data during transmission, redundant data is added to the low-bandwidth encrypted data stream to generate an encrypted data packet with error recovery capabilities. Redundant data addition is usually implemented through forward error correction codes (such as Reed-Solomon codes or LDPC codes), so that even if part of the data is lost or damaged during transmission, it can still be restored through redundant data. During the transmission stage, the encrypted data packets are dynamically routed and optimized for transmission, and the optimal transmission path is selected to improve transmission efficiency and security. The dynamic routing algorithm usually selects the path with the lowest latency or the most secure path based on real-time feedback of the network status to transmit the encrypted data stream to the destination. The receiver decrypts and controls the multi-level encrypted display data stream to restore the decrypted display content accessible to authorized users. The decryption process is the inverse operation of the previous encryption step. Decryption is performed by retrieving the corresponding key in the key pool, and the integrity of the hash value and watermark is verified to ensure that the received content has not been tampered with or forged.

[0113] In a specific embodiment, the execution step performs decryption control on the multi-level encrypted display data stream to obtain the decrypted display content accessible to the authorized user, which may specifically include the following steps:

[0114] Collect and pre-process the biometric data provided by the user to obtain the biometric authentication result, and obtain the dynamic password verification result based on the time-synchronized one-time password algorithm combined with the user's device information;

[0115] Real-time analysis of user behavior patterns is performed to obtain behavioral authentication scores through anomaly detection algorithms. The biometric authentication results, dynamic password verification results, and behavioral authentication scores are input into the fuzzy comprehensive evaluation model to obtain the credibility of multi-factor authentication.

[0116] The multi-factor authentication credibility and user RBAC role information are weighted and integrated, and the user access trust level is obtained through the trust evaluation algorithm. Based on the homomorphic decryption technology, the multi-level encrypted display data stream is partially decrypted to obtain the intermediate state data stream;

[0117] Perform attribute access control on the intermediate data stream to obtain a data subset that meets the user's permissions, distribute decryption keys, and obtain distributed decryption key shares through a secure multi-party computing protocol;

[0118] Decrypted data is obtained based on the data subset that meets the user's permissions and the distributed decryption key share;

[0119] Through isolated computing, the decrypted data is securely rendered to obtain decrypted display content accessible to authorized users.

[0120] Specifically, the biometric data provided by the user, such as fingerprints, facial recognition or iris scans, is collected and preprocessed. The collection of biometric data includes converting the raw biometric data captured by the sensor into digital form and standardizing it to eliminate noise and external interference factors. The preprocessing step may include filtering, image enhancement, feature extraction and other technologies to ensure that the features extracted from the raw data have high reliability. After preprocessing, the extracted features are matched with the templates in the database to obtain the biometric authentication result. Assume that the raw biometric data is B raw , the feature vector obtained after preprocessing is B pre , the matching score is S b At the same time, the time-synchronized one-time password algorithm (TOTP) is combined with user device information, such as device ID, IP address, etc. to generate a dynamic password and verify it. The TOTP algorithm generates a short-term password formula based on the current time and the user's secret key:

[0121] TOTP = HMAC_SHA1(K,T)

[0122] Where K is the secret key of the user device, T is the timestamp, and HMAC_SHA1 represents the HMAC algorithm using the SHA-1 hash function. Combined with the user device information D i (such as device ID and IP address), verify whether the generated dynamic password matches the password entered by the user, and obtain the dynamic password verification result S d . Perform real-time analysis on user behavior patterns, monitor user operation behaviors in the system, such as login time, access frequency, operation sequence, etc., and analyze the normality of these behaviors through anomaly detection algorithms. Anomaly detection algorithms can build user behavior models based on historical data and compare the deviation between current behavior and the model in real time. Assume that the behavior feature set is A = {a1, a2, …, a n}, where a i Represents a specific behavior feature. The behavior authentication score S is calculated by the anomaly detection algorithm f(A) a , used to evaluate the credibility of user behavior. The biometric authentication result S b , Dynamic password verification result S d and behavioral certification score S a Input into the fuzzy comprehensive evaluation model to generate the multi-factor authentication credibility. The fuzzy comprehensive evaluation model comprehensively evaluates multiple authentication results through given rules and weights, and finally obtains the authentication credibility C m represents the overall credibility of the user identity. Assume that the weight set is W = {w b ,w d ,w a}, then the multi-factor authentication credibility formula is:

[0123] C m =w b ·S b +w d ·S d +w a ·S a ;

[0124] Combined with the user's RBAC (role-based access control) role information, the multi-factor authentication credibility C m The user role information R is weighted and fused to generate the user's access trust level L t RBAC role information R is the level of authority granted to the user in the system. Different roles correspond to different access rights. m Combined with R, the final access trust level L is obtained t, indicating the user's trustworthiness in the current environment. In the data access phase, based on the user's access trust level L t , the multi-level encrypted display data stream is partially decrypted through homomorphic decryption technology to obtain the intermediate data stream. Homomorphic decryption allows the system to perform specific calculations without decrypting the original data, thereby ensuring the privacy of the data. In this process, only the data that meets the user's access trust level will be partially decrypted, and the rest will remain encrypted. Assume that the encrypted data is E d , the decrypted intermediate data stream is D m Attribute access control is performed on the intermediate data stream to ensure that users can only access the data subset that meets their permissions. Attribute access control determines the range of data that users can access based on user attributes, such as department, position, and permission level. Through screening and filtering, the content in the intermediate data stream is limited to the scope allowed by the user's permissions, generating a data subset D that meets the user's permissions. s . By distributing decryption keys and combining them with a secure multi-party computing protocol, a distributed decryption key share is generated. The secure multi-party computing protocol allows multiple participants to jointly calculate a function without sharing their private data, ensuring that the privacy of each participant is not leaked. Each participant will receive a decryption key share, and only when all the shares are combined can the data be decrypted. The key shares are distributed to authorized users, enabling them to use the data subset D that meets their permissions. s and the distributed decryption key share K share Perform the final decryption to obtain the complete decrypted data D f To ensure the security of the decrypted data, the decrypted data is securely rendered through isolated computing. The isolated computing environment can be a trusted execution environment (TEE), such as InteISGX, to ensure that the data will not be leaked or tampered with by the external environment after decryption. Through isolated computing, the decrypted data is securely rendered to generate the final decrypted display content D render , the content is visible only to authorized users, thus ensuring the security and integrity of the data.

[0125] In a specific embodiment, the execution step securely renders the decrypted data through isolated computing to obtain the decrypted display content accessible to the authorized user, which may specifically include the following steps:

[0126] Perform memory isolation on the decrypted data to obtain an isolated memory area, and set an access control list on the isolated memory area to obtain a protected data buffer;

[0127] Virtualizing a graphics processing unit according to a protected data buffer to obtain a virtual GPU instance, and configuring a security policy for the virtual GPU instance to obtain a GPU environment with an isolation feature;

[0128] Encapsulate the GPU environment with isolation characteristics into a trusted execution environment to obtain a trusted rendering container, and measure the integrity of the trusted rendering container to obtain a trusted measurement value;

[0129] Performing credibility verification on the rendering instruction according to the credibility metric value to obtain a trusted rendering instruction set, and performing instruction-level obfuscation on the trusted rendering instruction set to obtain obfuscated rendering instructions;

[0130] Dynamically reorder the obfuscated rendering instructions to obtain a random execution sequence, and dynamically compile the GPU shader according to the random execution sequence to obtain a secure shader program;

[0131] Performing code integrity check on the security shader program to obtain an integrity check result, and configuring the rendering pipeline according to the integrity check result to obtain a secure rendering pipeline;

[0132] Transform and rasterize the geometric data according to the safe rendering pipeline to obtain pixel-level rendering data, and perform pixel shading on the pixel-level rendering data to obtain an initial rendering result;

[0133] Encrypting the frame buffer object of the initial rendering result to obtain an encrypted frame buffer, and securely transmitting the encrypted frame buffer to obtain encrypted data on the display end;

[0134] Display driver isolation is performed according to the encrypted data of the display end to obtain an isolated display channel, the isolated display channel is dynamically decrypted with a key to obtain decrypted display data, and the display timing of the decrypted display data is controlled to obtain decrypted display content accessible to authorized users.

[0135] Specifically, memory isolation is performed on the decrypted data to ensure that the data is not accessed by unauthorized processes during processing. Memory isolation technology loads the decrypted data into an independent memory area, which is physically or logically isolated from other memory parts of the system to ensure that only processes with specific permissions can access it. An access control list is set for the isolated memory area, which defines which processes or users have the right to access the data in the memory, forming a protected data buffer. The graphics processing unit (GPU) is virtualized, and a virtual GPU instance is created using the protected data buffer. GPU virtualization technology allows multiple independent virtual machines or containers to share the same physical GPU resources, while each virtual GPU instance can run independently without affecting other instances. Security policies are configured for the virtual GPU instance to ensure that graphics processing operations are performed in an isolated and controlled environment to prevent malicious code from being executed on the GPU and stealing or tampering with data. Through these security policies, a GPU environment with isolation characteristics is created. The GPU environment with isolation characteristics is encapsulated in a trusted execution environment (TEE) to form a trusted rendering container. TEE is a hardware-supported security environment that can protect data and code in the container from external attacks. In order to verify the security of the trusted rendering container, integrity measurement is performed. Integrity measurement is a technique that detects tampering or other security threats by calculating and verifying the hash value of the data inside the container. By calculating the trust metric value M trust , quantify the integrity and trustworthiness of the container. Assume H(x) is a hash function, D c Ontainer is the data in the rendering container, and the trust metric value can be expressed as

[0136] M trust =H(D container );

[0137] Based on the obtained trust metric value, the credibility of the rendering instructions to be executed is verified to ensure that these instructions have not been tampered with or forged. The verified rendering instruction set forms a trusted rendering instruction set. The trusted rendering instruction set is obfuscated at the instruction level to make the instruction sequence difficult to be reverse engineered or tampered with. The obfuscation process involves disrupting the order of instructions or inserting pseudo instructions to make the instruction sequence more complicated. The obfuscated rendering instruction set is input into the dynamic reordering module for out-of-order execution. By changing the execution order of instructions, it is difficult for an attacker to understand their logic or tamper with them even if they can intercept some instructions. The rendering instructions executed out of order are dynamically compiled to generate a secure shader program. The shader program is the code used to perform graphics rendering tasks on the GPU, and its security directly affects the final rendering result. After the secure shader program is generated, the code integrity check is performed on it to ensure that the shader program has not been tampered with or damaged during the compilation and transmission process. According to the integrity check result, the rendering pipeline is configured to create a secure rendering pipeline. The rendering pipeline is the core part of graphics processing, involving the conversion process from geometric data to final pixels. By configuring a secure rendering pipeline, it is ensured that data will not be leaked or tampered with during the transformation, rasterization, and shading of geometric data. The final geometric data is rasterized to obtain pixel-level rendering data, which is further processed to generate the initial rendering result. The initial rendering result is encrypted by frame buffer object. The frame buffer object is an intermediate data structure that stores rendering output. It is protected by encryption technology to prevent the rendering result from being stolen or tampered with during transmission or storage. The encrypted frame buffer object is sent to the display end through a secure transmission channel to form display end encrypted data. At the display end, display driver isolation is performed based on the received encrypted data. Display driver isolation technology prevents unauthorized processes from accessing or tampering with display data by isolating display data from other parts of the system. The isolated display channel uses a dynamic key to decrypt and restore the decrypted display data. Dynamic key decryption is a technology that decrypts data with a regularly updated key, which helps prevent the key from being cracked by brute force. Display timing control is performed on the decrypted display data to ensure that the data is displayed in a predetermined order and time interval, generating decrypted display content accessible to authorized users.

[0138] The above describes the encryption method of the LED conference display screen in the embodiment of the present invention. The following describes the encryption system of the LED conference display screen in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an encryption system for an LED conference display screen includes:

[0139] The collection module 201 is used to collect and analyze the conference scene data to obtain the conference feature vector, content feature vector and user role information;

[0140] Processing module 202, used to perform multi-layer neural network processing on the conference feature vector, content feature vector and user role information to obtain a dynamic encryption requirement score and multiple candidate encryption strategies;

[0141] A construction module 203 is used to construct a two-level model based on the dynamic encryption requirement score and multiple candidate encryption strategies;

[0142] A solution module 204, used for solving a global optimal encryption strategy according to a two-level model through equivalent complementary constraints;

[0143] The grading module 205 is used to perform real-time scanning and grading of LED display content based on a global optimal encryption strategy to obtain a multi-element content set with dynamic sensitivity levels;

[0144] The control module 206 is used to generate a multi-level encrypted display data stream based on the dynamic sensitivity level of the multi-element content set and perform decryption control to obtain decrypted display content accessible to authorized users.

[0145] Through the collaboration of the above components, through the real-time collection and analysis of conference scene data, combined with multi-layer neural network processing, this method can accurately identify conference features, content features and user role information. The dual-level model and mixed integer quadratic programming method are used to achieve global optimization of security and latency. At the same time, through flexible permission allocation based on RBAC, the LED display content is scanned in real time and dynamically graded for sensitivity. Combined with layered differential encryption and dynamic watermark embedding technology, refined protection of different types of content is achieved. Multi-factor authentication and trust evaluation mechanism are introduced, combined with homomorphic decryption and secure multi-party computing technology, while ensuring data security, flexible decryption control is achieved to effectively prevent unauthorized access. Through isolated computing and secure rendering technology, including memory isolation, GPU virtualization, trusted execution environment and other measures, the security of decrypted display content is ensured, and potential security threats in the rendering process are effectively prevented. The anti-tampering ability and traceability of the content are enhanced. Through compression and redundant data addition technology, the transmission efficiency of encrypted data is optimized, while the error recovery ability of the system is improved, and the reliability of the display system is enhanced.

[0146] above Figure 2 The encryption system of the LED conference display screen in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The encryption device of the LED conference display screen in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0147] Figure 3It is a structural schematic diagram of an encryption device for an LED conference display screen provided by an embodiment of the present invention. The encryption device 300 for the LED conference display screen may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more massive storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the encryption device 300 for the LED conference display screen. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the encryption device 300 for the LED conference display screen to implement the steps of the encryption method of the above-mentioned LED conference display screen.

[0148] The encryption device 300 of the LED conference display screen may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It can be understood by those skilled in the art that Figure 3 The encryption device structure of the LED conference display screen shown does not constitute a limitation on the encryption device of the LED conference display screen provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0149] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the encryption method of the LED conference display screen.

[0150] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0151] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0152] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An encryption method for a LED conference display screen, characterized in that: The method comprises: Collect and analyze conference scene data to obtain conference feature vectors, content feature vectors, and user role information; Performing multi-layer neural network processing on the conference feature vector, the content feature vector and the user role information to obtain a dynamic encryption requirement score and multiple candidate encryption strategies; constructing a two-level model based on the dynamic encryption requirement score and multiple candidate encryption strategies; By solving the two-level model through equivalent complementary constraints, a global optimal encryption strategy is obtained; Based on the global optimal encryption strategy, the LED display content is scanned and graded in real time to obtain a multi-element content set with dynamic sensitivity levels; A multi-level encrypted display data stream is generated based on the dynamic sensitivity level of the multi-element content set and decryption control is performed to obtain decrypted display content accessible to authorized users.

2. The encryption method of the LED conference display screen according to claim 1 is characterized in that: The conference scene data is collected and analyzed to obtain conference feature vectors, content feature vectors and user role information, including: Performing cluster analysis on the conference type data to obtain a conference type feature matrix, and performing feature extraction on the real-time conference data based on the conference type feature matrix to obtain a conference feature vector; Perform optical character recognition and image segmentation on the LED display screen content to obtain text data sets and image data sets; Performing vectorization processing on the text data set to obtain a text feature matrix, and performing feature extraction on the image data set to obtain an image feature matrix; Performing tensor fusion on the text feature matrix and the image feature matrix to obtain a content feature vector; Performing time series analysis and frequent pattern mining on user behavior logs to obtain a user behavior feature set, and constructing a user similarity graph based on the user behavior feature set; Spectral clustering is performed on the user similarity graph to obtain initial user role groups, and the initial user role groups are dynamically adjusted to obtain user role information.

3. The encryption method of the LED conference display screen according to claim 1 is characterized in that: The multi-layer neural network processing is performed on the conference feature vector, the content feature vector and the user role information to obtain a dynamic encryption requirement score and multiple candidate encryption strategies, including: Performing tensor combination on the conference feature vector, the content feature vector and the user role information to obtain an input tensor; Performing feature extraction on the input tensor based on a three-dimensional convolutional neural network to obtain a high-dimensional feature map, and performing self-attention mechanism weighting on the high-dimensional feature map to obtain an attention weight matrix and a weighted feature map; Inputting the weighted feature map into a long short-term memory network to extract time series features to obtain time series correlation features, and performing residual connection and layer normalization on the time series correlation features to obtain optimized feature representation; Performing multi-scale feature processing on the optimized feature representation based on a multi-head attention mechanism to obtain multi-scale features, and inputting the multi-scale features into a fully connected layer to obtain an initial encryption requirement score and candidate encryption parameters; A softmax function operation is performed on the initial encryption requirement score to obtain a dynamic encryption requirement score, and a strategy generation network is constructed based on the candidate encryption parameters to generate multiple candidate encryption strategies.

4. The encryption method of the LED conference display screen according to claim 1 is characterized in that: The two-level model is constructed based on the dynamic encryption requirement score and multiple candidate encryption strategies, including: Performing fuzzy set conversion on the dynamic encryption requirement score to obtain a fuzzy encryption requirement matrix, and performing weighted summation on multiple candidate encryption strategies based on the fuzzy encryption requirement matrix to obtain an initial display content stream; Performing genetic crossover and mutation operations on the initial display content flow to obtain multiple candidate display content flows, and evaluating the multiple candidate display content flows to obtain an optimal display content flow; Based on the optimal display content flow, an upper-layer optimization objective function is constructed, including security maximization and delay minimization items, and user role information is hierarchically analyzed to obtain a role importance weight vector; Constructing an RBAC authority matrix based on the role importance weight vector to obtain an initial authority allocation scheme, and iteratively optimizing the initial authority allocation scheme to obtain an improved authority allocation scheme; Based on the improved authority allocation scheme, lower-level constraints are constructed, including role inheritance and mutual exclusion relationships, and the upper-level optimization objective function and the lower-level constraints are combined to obtain a two-level model.

5. The encryption method of the LED conference display screen according to claim 4 is characterized in that: The method of obtaining a global optimal encryption strategy by solving the two-level model through equivalent complementary constraints includes: Performing Lagrangian relaxation on the upper objective function of the two-level model to obtain a relaxed upper problem, and transforming the relaxed upper problem to obtain an equivalent lower problem; The equivalent lower-layer problem is combined with the lower-layer constraint condition to obtain a single-layer equivalent model, and nonlinear terms in the single-layer equivalent model are envelope linearized to obtain a linearized constraint set; Converting the logical constraints in the linearized constraint set into mixed integer constraints to obtain an initial mixed integer programming model, and performing variable fixing and constraint elimination on the initial mixed integer programming model to obtain a simplified model; Cutting the simplified model to generate valid inequality constraints, obtaining a mixed integer quadratic programming problem, and performing a tree search on the mixed integer quadratic programming problem to obtain a feasible solution set; A local search is performed on the feasible solution set to obtain an improved solution, and a sensitivity analysis is performed on the improved solution to obtain a global optimal encryption strategy.

6. The encryption method of the LED conference display screen according to claim 1 is characterized in that: Based on the global optimal encryption strategy, the LED display content is scanned and graded in real time to obtain a multi-element content set with dynamic sensitivity levels, including: Parameters of the global optimal encryption strategy are parsed to obtain a sensitivity evaluation function and a grading threshold set, and real-time image acquisition is performed on the LED display screen to obtain a frame sequence of original display content; Performing image enhancement on the original display content frame sequence to obtain an optimized image sequence, and performing content segmentation on the optimized image sequence to obtain a plurality of content element regions; Performing target detection and semantic segmentation on the content element area to obtain a content element type label, and performing text extraction on the text content element to obtain text content data; Performing semantic analysis on the text content data and non-text content elements to obtain a semantic feature vector, and inputting the semantic feature vector into the sensitivity evaluation function to obtain an initial sensitivity level by calculating a sensitivity score; The initial sensitivity level is subjected to context information and time series relationship analysis to obtain a dynamic sensitivity level, and the dynamic sensitivity level is quantified according to the grading threshold set to obtain a multi-element content set with discrete sensitivity levels.

7. The encryption method of the LED conference display screen according to claim 1 is characterized in that: The generating of a multi-level encrypted display data stream based on the dynamic sensitivity level of the multi-element content set and performing decryption control to obtain decrypted display content accessible to authorized users includes: Performing cluster analysis on the multi-element content set according to the dynamic sensitivity level and the content type to obtain content encryption groups, and generating a key for each encryption group based on a key distribution technology to obtain a key pool; Performing homomorphic encryption on each encrypted group to obtain primary encrypted content, and embedding metadata into the primary encrypted content to obtain encrypted content with hidden information; Generate an encrypted content hash based on the encrypted content with hidden information and store it in a distributed ledger to obtain tamper-proof encrypted content; Generate a digital watermark corresponding to the conference information and the timestamp based on a dynamic watermark generation algorithm, and perform pixel-level fusion on the tamper-proof encrypted content and the digital watermark to obtain watermark encrypted content; Compressing the watermark encrypted content to obtain a low-bandwidth encrypted data stream, and adding redundant data to the low-bandwidth encrypted data stream to obtain an encrypted data packet with error recovery capability; Dynamically routing and optimizing the transmission of the encrypted data packets to obtain a multi-level encrypted display data stream; The multi-level encrypted display data stream is decrypted and controlled to obtain decrypted display content accessible to authorized users.

8. The encryption method of the LED conference display screen according to claim 7 is characterized in that: The decryption control of the multi-level encrypted display data stream to obtain the decrypted display content accessible to the authorized user includes: Collect and pre-process the biometric data provided by the user to obtain the biometric authentication result, and obtain the dynamic password verification result based on the time-synchronized one-time password algorithm combined with the user's device information; Perform real-time analysis on user behavior patterns, obtain a behavior authentication score through an anomaly detection algorithm, and input the biometric authentication result, the dynamic password verification result and the behavior authentication score into a fuzzy comprehensive evaluation model to obtain the multi-factor authentication credibility; The multi-factor authentication credibility and the user RBAC role information are weightedly integrated, and the user access trust level is obtained through a trust evaluation algorithm, and the multi-level encrypted display data stream is partially decrypted based on the homomorphic decryption technology to obtain an intermediate state data stream; Perform attribute access control on the intermediate data stream to obtain a data subset that meets the user's authority, distribute decryption keys, and obtain distributed decryption key shares through a secure multi-party computing protocol; Obtaining decrypted data according to the data subset that meets the user's authority and the distributed decryption key share; The decrypted data is securely rendered through isolated computing to obtain decrypted display content accessible to authorized users.

9. The encryption method of the LED conference display screen according to claim 8 is characterized in that: The method of securely rendering the decrypted data through isolated computing to obtain decrypted display content accessible to authorized users includes: Performing memory isolation on the decrypted data to obtain an isolated memory area, and setting an access control list on the isolated memory area to obtain a protected data buffer; Virtualizing a graphics processing unit according to the protected data buffer to obtain a virtual GPU instance, and configuring a security policy for the virtual GPU instance to obtain a GPU environment with isolation characteristics; Performing trusted execution environment encapsulation on the GPU environment with isolation characteristics to obtain a trusted rendering container, and performing integrity measurement on the trusted rendering container to obtain a trusted measurement value; Performing credibility verification on the rendering instruction according to the credibility metric value to obtain a trusted rendering instruction set, and performing instruction-level obfuscation on the trusted rendering instruction set to obtain obfuscated rendering instructions; Dynamically reordering the obfuscated rendering instructions to obtain a random execution sequence, and dynamically compiling the GPU shader according to the random execution sequence to obtain a secure shader program; Performing a code integrity check on the secure shader program to obtain an integrity check result, and configuring the rendering pipeline according to the integrity check result to obtain a secure rendering pipeline; Transforming and rasterizing the geometric data according to the secure rendering pipeline to obtain pixel-level rendering data, and performing pixel shading on the pixel-level rendering data to obtain an initial rendering result; Encrypting the initial rendering result by means of a frame buffer object to obtain an encrypted frame buffer, and securely transmitting the encrypted frame buffer to obtain encrypted data on the display end; Display drive isolation is performed according to the display end encrypted data to obtain an isolated display channel, dynamic key decryption is performed on the isolated display channel to obtain decrypted display data, and display timing control is performed on the decrypted display data to obtain decrypted display content accessible to authorized users.

10. An encryption system for a LED conference display screen, characterized in that: Used to execute the encryption method of the LED conference display screen according to any one of claims 1 to 9, the system comprises: The acquisition module is used to collect and analyze the conference scene data to obtain the conference feature vector, content feature vector and user role information; A processing module, configured to perform multi-layer neural network processing on the conference feature vector, the content feature vector and the user role information to obtain a dynamic encryption requirement score and a plurality of candidate encryption strategies; A construction module for constructing a two-level model based on the dynamic encryption requirement score and a plurality of candidate encryption strategies; A solution module, used for solving the global optimal encryption strategy according to the two-level model through equivalent complementary constraints; A grading module, used for scanning and grading the LED display content in real time based on the global optimal encryption strategy to obtain a multi-element content set with dynamic sensitivity levels; The control module is used to generate a multi-level encrypted display data stream based on the dynamic sensitivity level of the multi-element content set and perform decryption control to obtain decrypted display content accessible to authorized users.

Images