Exhibition hall content self-adaptive central control display method and exhibition hall content self-adaptive central control display system

Abstract

The invention provides an exhibition hall content self-adaptive central control display method and system. The method comprises the following steps: acquiring and preprocessing exhibition hall display data; based on the exhibition hall display data obtained through preprocessing, generating feature vectors of the exhibition hall display data by adopting a multi-layer pulse neural network; based on the feature vector of the exhibition hall display data, outputting a content theme association matrix through a content association analysis model; acquiring real-time user behavior data, wherein the real-time user behavior data comprises user position information, sight focus information and stay time information; based on the content theme incidence matrix and the user behavior data, constructing a display strategy optimization model by adopting a space-time orthogonal propagation network so as to generate a self-adaptive display scheme through the display strategy optimization model; and encoding the self-adaptive display scheme by using a multicast addressing encoding scheme, and transmitting the self-adaptive display scheme to an exhibition hall display device for display.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence display technology, and specifically relates to a method and system for adaptive central control display of exhibition hall content. Background Art

[0002] In the exhibition hall display field, the intelligent and dynamic display of content is gaining increasing attention. Intelligent dynamic display aims to provide a personalized and immersive exhibition experience by using intelligent systems to perceive visitors' interests and needs in real time, dynamically adjusting the displayed content and presentation methods. Existing display technologies primarily implement intelligent content display through the following methods: first, using preset rules to switch display content according to fixed logic based on the different zones and time periods of the exhibition hall; second, based on simple human-computer interaction, allowing visitors to select content of interest through touch, gestures, and other methods; and third, using traditional machine learning methods to conduct statistical analysis of visitor behavioral data and adjust display strategies.

[0003] The existing display technology mainly has the following problems:

[0004] It is difficult to accurately match the display content with the interests and needs of visitors, resulting in unsatisfactory display effects; and there is a lack of semantic understanding of multimodal display content such as images, videos, and text, making it difficult to achieve intelligent organization of content and thematic correlation analysis. Summary of the Invention

[0005] This application provides a method and system for adaptively displaying exhibition hall content in a central control, aiming to address existing technical issues such as the lack of intelligent organization of display content and difficulty in real-time interaction with user behavior. The technical solutions adopted are as follows:

[0006] A method for adaptively displaying exhibition hall content in a central control system, comprising:

[0007] Acquiring and preprocessing exhibition hall display data, wherein the exhibition hall display data includes image data stream, video data stream and text data stream;

[0008] Based on the exhibition hall display data obtained through preprocessing, a multi-layer spiking neural network is used to generate a feature vector of the exhibition hall display data, wherein the multi-layer spiking neural network includes an input layer, a feature extraction layer, a time series integration layer, and an output layer;

[0009] Based on the feature vectors of the exhibition hall display data, outputting a content theme association matrix through a content association analysis model;

[0010] Acquire real-time user behavior data, including user location information, visual focus information, and dwell time information;

[0011] Based on the content topic association matrix and the user behavior data, a display strategy optimization model is constructed using a spatiotemporal orthogonal propagation network, so as to generate an adaptive display solution through the display strategy optimization model, wherein the spatiotemporal orthogonal propagation network includes a time dimension propagation layer, a space dimension propagation layer, and a spatiotemporal feature fusion layer;

[0012] The adaptive display scheme is encoded using a multicast addressing coding scheme and transmitted to the exhibition hall display device for display.

[0013] Before outputting the content topic association matrix through the content association analysis model, the method further includes:

[0014] A content association analysis model is constructed by combining the generalized Hebbian learning algorithm, including: constructing a content similarity matrix, designing temporal association strength update rules, and constructing a content topic association matrix.

[0015] The steps of constructing a content similarity matrix, designing a temporal correlation strength update rule, and constructing a content topic correlation matrix include:

[0016] Based on the feature vector, a content similarity matrix is calculated using a cosine similarity function;

[0017] Based on the preset time series window, a time series association strength update rule is designed to calculate the co-occurrence frequency of content pairs within the window and update the association strength;

[0018] An association graph is constructed based on the content similarity matrix and the updated association strength, and a community discovery algorithm is used to identify topic clusters to generate the content topic association matrix.

[0019] Before using a multi-layer pulse neural network to generate a feature vector of the exhibition hall display data, the method further includes:

[0020] Construct a multi-layer spiking neuron network architecture, wherein the input layer receives a standardized data stream, the feature extraction layer uses a multi-layer LeakyReLU spiking neuron, the time series integration layer uses a bidirectional LSTM unit, and the output layer is used to generate a feature vector;

[0021] The training of the spiking neuron network is achieved through a weight threshold leakage collaborative training mechanism, which includes: weight updating based on back propagation, dynamic adjustment of neuron firing thresholds, and achieving collaboration between forward activation and back propagation.

[0022] The method of using a multi-layer pulse neural network to generate a feature vector of the exhibition hall display data includes:

[0023] The multi-layer pulse neural network aligns and fuses different modal data in the feature space to generate a feature vector of the exhibition hall display data.

[0024] The step of constructing an association graph based on the content similarity matrix and the updated association strength, identifying topic clusters using a community discovery algorithm, and generating the content topic association matrix includes:

[0025] Applying a sparsity constraint and a temporal dependency constraint to the content similarity matrix;

[0026] Based on the L1 regularization term, the sparsity of the correlation matrix is controlled, and a sparse threshold is set to prune edges with correlation strength less than the threshold;

[0027] Construct a temporal dependency graph to calculate the temporal centrality of nodes, and adjust the association weight based on the temporal centrality;

[0028] Updating the association weights using a multi-objective optimization function, wherein the multi-objective optimization function includes a reconstruction error term, a sparsity constraint term, and a temporal consistency constraint term;

[0029] Based on the updated association weights, a community discovery algorithm is used to identify topic clusters and generate the content topic association matrix.

[0030] The method of constructing a display strategy optimization model based on the content theme association matrix and the user behavior data using a spatiotemporal orthogonal communication network so as to generate an adaptive display solution through the display strategy optimization model includes:

[0031] The time dimension propagation layer inputs the temporal features of user behavior and generates a time dimension feature vector through the temporal attention mechanism;

[0032] The spatial dimension propagation layer constructs a spatial graph structure based on the content topic association matrix and generates a spatial dimension feature vector through a graph attention mechanism;

[0033] The spatiotemporal feature fusion layer integrates the time dimension feature vector and the space dimension feature vector through a multi-head attention mechanism to generate a fused feature vector;

[0034] The adaptive display scheme is generated based on the fused feature vector, including calculating content display priority, optimizing display timing arrangement, and generating a display layout scheme.

[0035] The step of generating the adaptive display scheme based on the fused feature vector includes:

[0036] Calculate attention score based on user’s current location and gaze focus;

[0037] Estimate interest level based on content dwell time;

[0038] Calculating a display priority by combining the attention score and the interest level;

[0039] The display area is dynamically allocated according to the display priority and the content switching timing is optimized.

[0040] The method of encoding the adaptive display scheme using a multicast addressing coding scheme and transmitting the encoding scheme to an exhibition hall display device for display includes:

[0041] Construct a coding tree consisting of device layer, content type layer and content item layer, and assign a unique identifier to each display device;

[0042] Design the content addressing format, including the prefix code for identifying the device ID and content type, the infix code for identifying the presentation parameters, and the suffix code for identifying the priority and timing information, and generate the addressing table;

[0043] Executing a presentation plan based on the addressing table includes parsing presentation instructions, scheduling display resources, and monitoring presentation effects.

[0044] The present application also provides an exhibition hall content adaptive central control display system, including:

[0045] An acquisition module, configured to acquire and pre-process exhibition hall display data, wherein the exhibition hall display data includes image data stream, video data stream and text data stream;

[0046] A generation module, configured to generate a feature vector of the exhibition hall display data using a multi-layer spiking neural network based on the exhibition hall display data obtained through preprocessing, wherein the multi-layer spiking neural network includes an input layer, a feature extraction layer, a time series integration layer, and an output layer;

[0047] A content association module, configured to output a content theme association matrix through a content association analysis model based on the feature vectors of the exhibition hall display data;

[0048] The acquisition module is further used to acquire real-time user behavior data, wherein the real-time user behavior data includes user location information, sight focus information, and dwell time information;

[0049] The generation module is further configured to construct a display strategy optimization model using a spatiotemporal orthogonal propagation network based on the content topic association matrix and the user behavior data, so as to generate an adaptive display solution through the display strategy optimization model, wherein the spatiotemporal orthogonal propagation network includes a time dimension propagation layer, a space dimension propagation layer, and a spatiotemporal feature fusion layer;

[0050] The display module is used to encode the adaptive display scheme using a multicast addressing coding scheme and transmit it to the exhibition hall display device for display.

[0051] The beneficial effects of this application include:

[0052] 1. Deep feature extraction of multimodal content is achieved through a multi-layer spiking neural network, improving the accuracy of content understanding;

[0053] 2. Adopting a content association analysis model to achieve intelligent organization of displayed content, enhancing the thematic and systematic nature of the content display;

[0054] 3. The introduction of a spatiotemporal orthogonal communication network enables real-time adaptive optimization of display strategies and improves system response speed;

[0055] 4. Design a multicast addressing coding scheme to optimize the distribution efficiency of display content. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the disclosed embodiments in the embodiments of the present application, the following will briefly introduce the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative work.

[0057] Figure 1 A flowchart of the content adaptive display method provided in an embodiment of the present application;

[0058] Figure 2 This is a schematic diagram of the structure of a multi-layer pulse neural network in an embodiment of the present application;

[0059] Figure 3 This is a flowchart for constructing a content association analysis model in an embodiment of the present application;

[0060] Figure 4 This is a schematic diagram of the architecture of the spatiotemporal orthogonal communication network in an embodiment of the present application;

[0061] Figure 5 A schematic diagram illustrating content encoding and distribution in an embodiment of the present application;

[0062] Figure 6 This is a module structure diagram of the content adaptive display system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0064] It should be clear that the following embodiments of the present disclosure are described through specific concrete examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0065] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0066] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0067] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0068] like Figure 1 As shown, the content adaptive display method provided by this application includes the following steps:

[0069] Step 1: Acquire and preprocess exhibition hall display data

[0070] In this embodiment, we first need to acquire the multimodal display data within the exhibition hall. Exhibition halls typically contain a rich variety of display content, including static images and text descriptions, as well as dynamic video presentations. These different types of content together constitute a complete display system. Therefore, this step first collects this multimodal content and converts it into a standardized data stream for processing.

[0071] Specifically, for an image data stream I = {i1, i2, ..., in}, each image element ii represents a display image. These images may be physical photographs of exhibits, design schematics, or infographics. The system captures these images using a high-definition camera or image acquisition device, ensuring that the captured images meet display requirements.

[0072] For a video data stream V = {v1, v2, ..., vm}, each video element vi could be an operational demonstration of an exhibit, a historical image, or a 3D animation. The system captures these video contents in real time using a video acquisition device, while ensuring that the frame rate and resolution of the video meet the requirements for smooth display.

[0073] For a text data stream T = {t1, t2, ..., tk}, each text element ti may contain text information such as exhibit descriptions, historical background, or technical specifications. This text content can be extracted from display boards using text recognition technology or directly obtained from the exhibition hall content management system.

[0074] After acquiring this raw data, standardization preprocessing is required to ensure the accuracy and efficiency of subsequent processing. For image data, normalization is first performed to resize images of varying sizes to a standard size. The pixel distribution of the image is also standardized to meet the requirements of neural network processing. Image enhancement is also required to improve image clarity and contrast.

[0075] For video data, we first perform frame extraction, breaking down the continuous video sequence into keyframe sequences. These keyframes are then time-aligned to ensure effective comparison of temporal features across different videos. Video resolution and frame rate also need to be standardized.

[0076] For text data, we first need to clean the text to remove special characters and redundant information. Then, we perform word segmentation to break the continuous text into meaningful word units. Finally, we use word embedding technology to convert the text into a standardized vector representation to facilitate subsequent semantic analysis.

[0077] Through this preprocessing, standardized multimodal data streams are generated. These data streams retain the core information of the original content while exhibiting excellent computational properties, laying the foundation for subsequent feature extraction and content analysis. The standardized data format also facilitates unified management and rapid retrieval within the system, improving overall processing efficiency.

[0078] Step 2: Generate feature vectors using a multi-layer spiking neural network

[0079] like Figure 2 As shown, the constructed multi-layer pulse neural network includes:

[0080] 1) Input layer: receives standardized multimodal data stream;

[0081] 2) Feature extraction layer: uses multiple layers of LeakyReLU pulse neurons, with the number of neurons in each layer being {N1, N2, ..., Nk};

[0082] 3) Temporal integration layer: bidirectional LSTM units are used to capture content temporal dependencies;

[0083] 4) Output layer: Generate d-dimensional feature vector F = {f1, f2, ..., fd}.

[0084] The network is trained via weight threshold leakage co-training mechanism:

[0085] 1) Weight update rules:

[0086] Calculate gradients based on backpropagation

[0087] The momentum term μ is introduced to accelerate convergence;

[0088] Update weights

[0089] 2) Threshold adaptive adjustment:

[0090] Set the initial neuron firing threshold θ0;

[0091] Dynamically adjust the threshold θ(t) according to the input intensity;

[0092] When the membrane potential exceeds the threshold, the pulse is triggered.

[0093] In this step, it is necessary to build a multi-layer pulse neural network architecture in advance. The multi-layer pulse neural network includes an input layer, a feature extraction layer, a timing integration layer and an output layer.

[0094] The input layer receives a standardized data stream, the feature extraction layer uses a multi-layer LeakyReLU pulse neuron, the time series integration layer uses a bidirectional LSTM unit, and the output layer is used to generate a feature vector;

[0095] In addition, the training of the spiking neuron network is achieved through a weight threshold leakage collaborative training mechanism, which includes: weight updating based on back propagation, dynamic adjustment of neuron firing thresholds, and achieving collaboration between forward activation and back propagation.

[0096] Therefore, the method of using a multi-layer pulse neural network to generate the feature vector of the exhibition hall display data includes: the multi-layer pulse neural network aligns and fuses different modal data in the feature space to generate the feature vector of the exhibition hall display data.

[0097] The following is an explanation using a specific example.

[0098] For example, an intelligent robot exhibit in the exhibition hall includes a picture of the robot's appearance, a demonstration video, and technical documentation. After preprocessing, these three different modal data are input into a multi-layer spiking neural network for feature extraction.

[0099] For image data, assume that after preprocessing, a 224×224 pixel RGB image is obtained. First, the image data is converted into the input voltage of the neuron through the input layer. For example, each pixel value in the image can be mapped to the input voltage as follows:

[0100] V_input=(pixel_value / 255.0)*V_max

[0101] Where pixel_value is the pixel value (0-255) and V_max is the set maximum input voltage (e.g., 1.0mV). In this way, an RGB image is converted into the voltage values of 224×224×3 input neurons.

[0102] In the feature extraction layer, multiple layers of LeakyReLU spiking neurons are used for processing. Taking the first layer as an example, it contains 64 convolution kernels, each with a size of 3×3 and a stride of 1. When the input voltage exceeds the neuron's firing threshold θ (initially set to 0.5mV), the neuron will generate a pulse. The activation function of LeakyReLU can be expressed as:

[0103] f(x)={

[0104] x,if x>0

[0105] 0.01x,if x≤0

[0106] }

[0107] Assume that at time t, the membrane potential of a neuron in the first layer is:

[0108] V_mem(t)=0.8mV>θ

[0109] At this point the neuron fires a spike and the membrane potential resets:

[0110] V_mem(t+1)=0

[0111] Through the weight-threshold-leakage co-training mechanism, the firing threshold of a neuron is dynamically adjusted according to the input strength. For example, if a neuron continuously receives strong input, the threshold will increase accordingly:

[0112] θ(t+1)=θ(t)+β·V_mem(t)

[0113] Where β is a regulation coefficient (such as 0.1). This mechanism can prevent neurons from being overactivated.

[0114] For video data, the system first extracts a sequence of key frames. For example, suppose 30 key frames are extracted from a 10-second demo video. These key frames are processed through the same convolutional layer described above to obtain frame-level features. Then, in the temporal integration layer, bidirectional LSTM units are used to capture the temporal dependencies between frames.

[0115] Taking a certain LSTM unit as an example, the state update process at time t is:

[0116] f_t=σ(W_f·[h_(t-1),x_t]+b_f)

[0117] i_t=σ(W_i·[h_(t-1),x_t]+b_i)

[0118] c_t=f_t*c_(t-1)+i_t*tanh(W_c·[h_(t-1),x_t]+b_c)

[0119] o_t=σ(W_o·[h_(t-1),x_t]+b_o)

[0120] h_t=o_t*tanh(c_t)

[0121] Among them, f_t, i_t, o_t are the activation values of the forget gate, input gate and output gate respectively, c_t is the unit state, and h_t is the output state.

[0122] For text data, assuming a technical description document contains 500 words, the system first converts each word into a 300-dimensional word vector. These word vector sequences are also processed through an LSTM layer to capture the semantic dependencies between words.

[0123] In the output layer, features from different modalities are uniformly mapped to the same feature space. Assuming that the final feature vector dimension is 512, for the above exhibit, its feature vector can be expressed as:

[0124] F=[f_1,f_2,...,f_512]

[0125] The value range of each dimension is between [-1, 1], reflecting the feature strength of the exhibit in different semantic dimensions. For example:

[0126] f_1=0.85 / / may represent "intelligent" characteristics

[0127] f_2=0.62 / / may represent the "interactive" feature

[0128] f_3=-0.34 / / may represent the "complexity" feature ...

[0130] This feature vector integrates the visual, dynamic, and semantic features of the exhibits, providing a basis for subsequent content association analysis. Through training, the network has learned to map semantically similar content to adjacent regions in the feature space.

[0131] In practice, the system processes the content of multiple exhibits simultaneously. For example, if there are 100 exhibits in an exhibition hall, 100 512-dimensional feature vectors will be generated, forming a 100×512 feature matrix. This feature matrix provides a holistic semantic representation of the exhibition hall content, laying the foundation for the next step of content association analysis.

[0132] Step 3: Build a content association analysis model

[0133] like Figure 3 As shown, a content association analysis model is constructed based on the feature vector:

[0134] 1) Calculate the content similarity matrix:

[0135] Define the cosine similarity function sim(fi,fj);

[0136] Construct a similarity matrix S, where Sij = sim(fi,fj);

[0137] Set the similarity threshold τ for screening;

[0138] 2) Design the update rules of timing correlation strength:

[0139] Define the timing window size w;

[0140] Calculate the co-occurrence frequency C(fi,fj) of the content pairs in the window;

[0141] Update the association strength R = γ·R + η·C;

[0142] 3) Build a content theme correlation matrix:

[0143] Applying sparsity constraints and timing dependency constraints;

[0144] Set the sparse threshold λ to trim weak correlations;

[0145] Construct a temporal dependency graph to calculate centrality;

[0146] Multi-objective optimization is used to update the association weights;

[0147] Identifying topic clusters using community discovery.

[0148] In this step, a core innovation of the present invention is the construction of a content association analysis model using the generalized Hebbian learning algorithm. The generalized Hebbian learning algorithm is derived from the Hebbian learning rule in biological neural networks, which states that "connections between simultaneously activated neurons are strengthened." The present invention extends this principle to content association analysis, constructing a dynamic content association network by analyzing the similarity and temporal co-occurrence of content features.

[0149] Specifically, we first need to construct a content similarity matrix. Based on the feature vectors generated by the multi-layer spiking neural network in the previous step, the system calculates the similarity between the content using the cosine similarity function. The cosine similarity function measures the similarity of content by calculating the cosine value of the angle between the feature vectors. This value ranges from [-1, 1], with larger values indicating more similar content. For any two content feature vectors fi and fj, their similarity can be expressed as:

[0150] sim(fi,fj)=(fi·fj) / (||fi||×||fj||)

[0151] Here, "·" represents the vector dot product, and ||fi|| represents the modulus of vector fi. Based on this, we can construct a content similarity matrix S, where the matrix element Sij = sim(fi,fj). The similarity matrix reflects the static associations between content.

[0152] Next, we design rules for updating the temporal association strength. This step aims to capture the dynamic associations between content. The system defines a temporal sliding window and counts the co-occurrence frequency of different content within the window. For example, if a visitor frequently views a related piece of content after viewing a certain piece of content, there is a strong temporal association between the two pieces of content. The specific update rules are as follows:

[0153] First, define the time window size w, which determines the time range of the analysis. For any two pieces of content i and j, calculate their co-occurrence count C(fi,fj) within the window w. Then, use the cumulative update method to calculate the association strength:

[0154] Rij(t)=γ·Rij(t-1)+η·C(fi,fj)

[0155] Here, γ is a decay factor (0 < γ < 1), which gradually reduces the influence of history; η is the learning rate, which controls the degree to which new co-occurrences affect the strength of associations. This dynamic update mechanism enables the system to adapt to changes in content associations.

[0156] Finally, the content-topic correlation matrix is constructed. This step combines the static similarity and dynamic correlation strength obtained in the previous two steps to form the final content-topic correlation matrix. The specific process includes:

[0157] First, the similarity matrix S and the correlation strength matrix R are weighted fused:

[0158] M=α·S+β·R

[0159] Here, α and β are weight coefficients, satisfying α + β = 1. This fusion approach not only considers the inherent similarity of content but also reflects the associations revealed by visit behavior.

[0160] Then, the fusion matrix M is regularized. L1 regularization is applied to control the sparsity of the matrix, and a threshold λ is set to prune weak associations while retaining significant relationships. Simultaneously, a temporal dependency graph is constructed to calculate the temporal centrality of nodes. Based on this centrality, the association weights are adjusted to ensure that key content is prominent in the association network.

[0161] Finally, the association weights are updated using a multi-objective optimization function:

[0162] L=L1+αL2+βL3

[0163] in:

[0164] L1 is the reconstruction error term, which ensures the consistency of the updated correlation matrix with the original observations;

[0165] L2 is a sparse constraint term that controls the complexity of the correlation matrix;

[0166] L3 is the temporal consistency constraint, which maintains the temporal consistency of the association relationship;

[0167] By optimizing the aforementioned objective function, we can determine the optimal association weights that balance multiple objectives. Subsequently, we use a community discovery algorithm to identify content topic clusters within the association network, ultimately generating an association matrix that reflects the thematic organization of the content. This matrix not only reflects the similarity and association strength between content, but also reveals thematic divisions within the content, providing an important basis for optimizing subsequent display strategies.

[0168] The process of constructing an association graph based on the content similarity matrix and the updated association strength, identifying topic clusters using a community discovery algorithm, and generating the content topic association matrix specifically includes:

[0169] Sparse constraints and temporal dependency constraints are applied to the content similarity matrix; the sparsity of the association matrix is controlled based on the L1 regularization term, and a sparse threshold is set to prune edges with association strength less than the threshold; a temporal dependency graph is constructed to calculate the temporal centrality of the nodes, and the association weights are adjusted based on the temporal centrality; the association weights are updated using a multi-objective optimization function, the multi-objective optimization function including a reconstruction error term, a sparsity constraint term, and a temporal consistency constraint term; based on the updated association weights, a community discovery algorithm is used to identify topic clusters to generate the content topic association matrix.

[0170] Taking the aerospace technology development series displayed in the exhibition hall as an example, the system first obtains a content similarity matrix. Assuming the exhibition hall contains 50 related exhibits, each of which contains multimodal content such as images, videos, and text, the above steps can produce a 50×50 content similarity matrix S.

[0171] First, apply a sparsity constraint to the content similarity matrix. In reality, not all exhibits have strong correlations; most exhibits have weak correlations. Therefore, a sparsity constraint is needed to filter out significant correlations. Specifically, L1 regularization is used to control the sparsity of the correlation matrix:

[0172] L1(S)=λ||S||1=λΣΣ|Sij|

[0173] Where λ is the regularization coefficient, which is used to control the degree of sparsification. Taking λ = 0.1 as an example, assuming that the similarity between two items i and j in the original similarity matrix is:

[0174] Sij=0.15

[0175] Since this similarity value is small, it may be weakened to a value close to 0 after L1 regularization. Then, set a sparse threshold (such as 0.1) to trim the association strength: if Sij < 0.1, set Sij = 0. This will produce a sparse similarity matrix S', in which most elements are 0, retaining only significant associations.

[0176] Next, we construct a temporal dependency graph and calculate temporal centrality. For the theme of aerospace technology development, there's a clear temporal relationship between exhibits. For example, "The Beginning of Spaceflight" should be displayed before "Manned Spaceflight." Based on the visitor's browsing sequence, we can construct a temporal dependency graph G = (V, E):

[0177] The node set V represents each exhibition item;

[0178] The edge set E represents the temporal dependency between exhibits;

[0179] The weight of the edge represents the strength of the temporal dependency;

[0180] For each node v in the graph G, calculate its temporal centrality TC(v):

[0181] TC(v)=Σu∈V d(u,v)

[0182] Where d(u,v) represents the shortest path length from node u to node v. Taking the "Manned Spaceflight" exhibit as an example, assume that its temporal centrality is:

[0183] TC ("manned space flight") = 0.75

[0184] This indicates that the exhibit occupies a relatively important position in the temporal dependency network. Adjust the association weight based on temporal centrality:

[0185] Wij=Sij'*(1+α*(TC(i)+TC(j)))

[0186] Where α is the adjustment coefficient (e.g., 0.2). In this way, the association weights between exhibits with higher temporal centrality will be appropriately enhanced.

[0187] Next, the association weights are updated using a multi-objective optimization function. This optimization function consists of three main terms:

[0188] L=L1+β1L2+β2L3

[0189] in:

[0190] 1. L1 is the reconstruction error term, which measures the consistency of the optimized correlation matrix with the original observations:

[0191] L1=||WS||F2

[0192] where ||·||F represents the Frobenius norm

[0193] 2. L2 is a sparse constraint term, such as the aforementioned L1 regularization term:

[0194] L2=λ||W||1

[0195] 3. L3 is a timing consistency constraint that ensures that the association relationship satisfies the timing dependency:

[0196] L3=Σ(i,j)∈E|Wij-Wji|

[0197] β1 and β2 are balancing coefficients (e.g., β1 = 0.3, β2 = 0.4). By minimizing the objective function L through an optimization algorithm such as gradient descent, the optimal association weight matrix W* that balances the constraints can be obtained.

[0198] Finally, we use a community discovery algorithm to identify topic clusters. We consider the association weight matrix W* as the adjacency matrix of a weighted undirected graph and use the Louvain algorithm for community detection. This algorithm discovers community structures in a network by optimizing modularity:

[0199] Q=(1 / 2m)Σij[Wij*-(ki*kj) / (2m)]δ(ci,cj)

[0200] in:

[0201] m is the sum of the weights of all edges in the network;

[0202] ki, kj are the degrees of nodes i and j;

[0203] ci, cj are the communities to which nodes i and j belong;

[0204] δ(ci,cj) is the indicator function, which is 1 when ci=cj, otherwise it is 0;

[0205] Taking the aerospace technology exhibit as an example, the algorithm may identify the following theme clusters:

[0206] Cluster 1: exhibits related to the start of the aerospace industry;

[0207] Cluster 2: exhibits related to manned space flight;

[0208] Cluster 3: Deep space exploration related exhibits;

[0209] Cluster 4: exhibits related to aerospace applications;

[0210] Based on these theme clusters, a content theme correlation matrix T is constructed. Matrix T not only reflects the correlation strength between exhibits, but also reflects the hierarchical structure of the themes. For example:

[0211] T_ij={

[0212] 1.0, if i and j belong to the same topic cluster

[0213] 0.5, if i and j belong to strongly related topic clusters

[0214] 0.2, if i and j belong to weakly related topic clusters

[0215] 0.0, if i and j are unrelated

[0216] }

[0217] The content theme association matrix obtained in this way not only maintains the original association relationship, but also adds semantic information at the theme level, providing an important basis for subsequent display strategy optimization.

[0218] Step 4: Get real-time user behavior data

[0219] Deploy the data collection system to obtain:

[0220] 1) User location information:

[0221] Positioning is achieved through a Bluetooth beacon array; RFID readers track movement trajectories; and position sequence P(t) is generated.

[0222] 2) Eye focus information:

[0223] The infrared camera captures the sight direction; coordinate mapping conversion; generates the focus sequence G(t);

[0224] 3) Stay time information:

[0225] Count the content residence time; record the switching frequency; generate the time series D(t);

[0226] For example:

[0227] First, install the eye tracking device:

[0228] Deploy infrared camera arrays C = {c1, c2, ..., cn} at key locations in the exhibition hall;

[0229] Configure the gaze tracking algorithm parameters, including the sampling frequency f and the accuracy threshold ε;

[0230] Establish the mapping relationship M(x,y) from the line of sight coordinates to the display screen;

[0231] Second, deploy location-aware sensors:

[0232] Arrange a Bluetooth beacon array B = {b1, b2, ..., bm} for indoor positioning;

[0233] Set up an RFID reader array R = {r1, r2, ..., rk} to track the user's movement trajectory;

[0234] Build a position coordinate system to achieve multi-sensor data fusion;

[0235] Finally, configure the behavior data collection interface:

[0236] Define data collection protocols, including data format and transmission methods;

[0237] Set up data caching and preprocessing modules;

[0238] Build real-time data stream processing pipelines;

[0239] Real-time collection of user behavior data based on the deployed system:

[0240] First, obtain the user location trajectory P(t):

[0241] Calculate the user's real-time location coordinates (x, y) using Bluetooth beacon array B;

[0242] Track the user's moving direction θ based on the RFID reader array R;

[0243] Generate position trajectory sequence P(t) = {p1(t), p2(t), ..., pn(t)};

[0244] Secondly, record the sight focus sequence G(t):

[0245] Capture the user's sight direction through the infrared camera array C;

[0246] Use the mapping relationship M(x,y) to convert to screen coordinates;

[0247] Generate a sight focus sequence G(t) = {g1(t), g2(t), ..., gm(t)};

[0248] Finally, count the content dwell time D(t):

[0249] Calculate the user's dwell time in front of each displayed content;

[0250] Record content switching frequency and sequence;

[0251] Generate a dwell time sequence D(t) = {d1(t), d2(t), ..., dk(t)};

[0252] Step 5: Build a display strategy optimization model

[0253] like Figure 4 As shown, the space-time orthogonal propagation network includes:

[0254] 1) Time dimension propagation layer:

[0255] Input user behavior temporal characteristics;

[0256] Temporal attention mechanism At(t);

[0257] Generate time feature vector Ft;

[0258] 2) Spatial dimension propagation layer:

[0259] Construct graph structure based on topic correlation matrix;

[0260] Graph attention mechanism As(s);

[0261] Generate spatial feature vector Fs;

[0262] 3) Spatiotemporal feature fusion layer:

[0263] Multi-head attention mechanism Mh;

[0264] Residual connections prevent gradient disappearance;

[0265] Output fused feature vector F*;

[0266] Generate display strategy based on fusion features:

[0267] 1) Calculate display priority:

[0268] Calculate attention score based on position and gaze;

[0269] Estimate interest level based on dwell time;

[0270] Calculate the priority score comprehensively;

[0271] 2) Optimize display arrangements:

[0272] Dynamically allocate display area;

[0273] Optimize content switching timing;

[0274] Generate display layout plan;

[0275] Step 5 mainly includes:

[0276] The time dimension propagation layer inputs the temporal features of user behavior and generates a time dimension feature vector through the temporal attention mechanism;

[0277] The spatial dimension propagation layer constructs a spatial graph structure based on the content topic association matrix and generates a spatial dimension feature vector through a graph attention mechanism;

[0278] The spatiotemporal feature fusion layer integrates the time dimension feature vector and the space dimension feature vector through a multi-head attention mechanism to generate a fused feature vector;

[0279] The adaptive display scheme is generated based on the fused feature vector, including calculating content display priority, optimizing display timing arrangement, and generating a display layout scheme.

[0280] Specifically, the system's spatiotemporal orthogonal propagation network contains three key levels, which respectively deal with the time dimension, space dimension and the fusion of spatiotemporal features.

[0281] In the temporal propagation layer, the system first inputs the temporal features of the user's behavior, including the position sequence P(t), the gaze sequence G(t), and the dwell sequence D(t). This sequential data is processed using the temporal attention mechanism At(t), which captures temporal dependencies by calculating the attention score between the current moment t and the previous k moments. Specifically, for any moment t, the system calculates the attention score between it and the previous k moments: score(qi,kj) = (Wqqi)T(Wkkj) / √dk, where qi is the query vector at the current moment, kj is the key vector at the previous k moments, Wq and Wk are learnable parameter matrices, and dk is the dimension of the key vector. After normalizing the attention scores using the softmax function, a temporal feature vector Ft = Σαi(t)·Vi is generated, where Vi is the value vector at each moment.

[0282] In the spatial communication layer, the system constructs a spatial graph structure Gs = (Vs, Es) based on the content-theme correlation matrix. The node set Vs represents the display area, and the edge set Es represents the spatial correlation between areas. Edge weights are determined by the content-theme correlation matrix and physical distance, calculated as w(i, j) = exp(-d(i, j) / σ)*T(i, j). The spatial dependencies between nodes are learned through the graph attention mechanism As(s). For each node v, the attention coefficient evu = LeakyReLU(aT[Whv||Whu]) is calculated for its neighbor u, where W is the weight matrix and a is the attention vector. Attention weights αvu = softmax(evu) are obtained through normalization, ultimately generating a spatial feature vector Fs = Σαvu·hu.

[0283] In the spatiotemporal feature fusion layer, the system uses a multi-head attention mechanism Mh to integrate temporal and spatial features. The implementation uses eight attention heads, each with a dimension of 32. First, the temporal and spatial features are projected to obtain Ft' = WtFt and Fs' = WsFs. Each attention head then performs the Attention operation: headi = Attention(WiQFt', WiKFs', WiVFs'), where WiQ, WiK, and WiV are learnable parameter matrices. Finally, the outputs of all attention heads are concatenated and linearly transformed to obtain the fused feature vector F* = WO[head1; head2; ...; head8]. To prevent vanishing gradients, a residual connection is added: F* = F* + Ft + Fs.

[0284] Based on the obtained fusion feature vector F*, the system generates an adaptive display plan. First, the content display priority is calculated, and the priority score score(c) = w1·α(c) + w2·β(c) is calculated for each displayable content c, where α(c) is the attention score calculated based on the user's current position and visual focus, and β(c) is the interest score calculated based on the historical stay time. Then, the display timing is optimized by constructing a Markov decision process, with the current display content combination as the state, content switching as the action, and the visitor's attention and interest changes as the reward. Through the policy gradient algorithm Optimize the display sequence. Finally, generate a specific display layout plan for each display screen based on priority, including detailed design such as area allocation, content switching timing, and transition animation effects.

[0285] Taking the aerospace science and technology exhibition hall as an example, the hall is equipped with multiple large display screens to showcase content on different topics. The system needs to dynamically optimize the display strategy based on the real-time behavior of visitors. The following details the processing steps at each layer of the spatiotemporal orthogonal communication network.

[0286] 1. Time dimension propagation layer

[0287] This layer mainly processes the temporal characteristics of user behavior. Take the behavior data of a certain visitor group within 10 minutes as an example:

[0288] First, organize the user behavior temporal features into sequence form:

[0289] Position sequence P(t) = {p1, p2, ..., pt}, where pt represents the position coordinate at time t

[0290] Gaze sequence G(t) = {g1, g2, ..., gt}, where gt represents the gaze focus at time t

[0291] Dwell sequence D(t) = {d1, d2, ..., dt}, where dt represents the dwell time at time t

[0292] These sequences are then processed by the temporal attention mechanism At(t). For any time t, the attention score between it and the previous k times is calculated:

[0293] score(qi,kj)=(Wqqi)T(Wkkj) / √dk

[0294] in:

[0295] qi is the query vector at the current time t

[0296] kj is the key vector of the previous k moments

[0297] Wq and Wk are learnable parameter matrices

[0298] dk is the dimension of the key vector

[0299] For example, if the position at time t is the center of the exhibition hall (5.0, 3.0), the system will focus on:

[0300] Whether it was in the adjacent area at the previous moment;

[0301] Whether there are significant changes in movement speed and direction;

[0302] Whether there is clustering in the group's location;

[0303] Normalize the attention scores through the softmax function:

[0304] α(t)=softmax(score(qt,K))

[0305] Finally, the time dimension feature vector is generated:

[0306] Ft=Σαi(t)·Vi

[0307] Where Vi is the value vector at each moment. This feature vector encodes the temporal pattern of visit behavior and may have a dimension of 256.

[0308] 2. Spatial Dimension Propagation Layer

[0309] This layer builds a spatial graph structure based on the content theme association matrix. Assume that the exhibition hall has 20 display areas, and each area may display multiple themes.

[0310] First, construct the spatial graph Gs = (Vs, Es):

[0311] The node set Vs represents the display area;

[0312] The edge set Es represents the spatial association between regions;

[0313] The edge weight is determined by the content topic association matrix and the physical distance:

[0314] w(i,j)=exp(-d(i,j) / σ)*T(i,j)

[0315] in:

[0316] d(i,j) is the physical distance between regions i and j;

[0317] σ is the distance attenuation parameter (e.g., 5.0);

[0318] T(i,j) is the corresponding topic association strength;

[0319] Then, the graph attention mechanism As(s) is applied to learn the spatial dependencies between nodes. For each node v, the attention coefficient between it and its neighbor node u is calculated:

[0320] evu=LeakyReLU(aT[Whv||Whu])

[0321] in:

[0322] W is the weight matrix;

[0323] a is the attention vector;

[0324] || indicates a splicing operation;

[0325] The attention weight is obtained by normalization:

[0326] αvu=softmax(evu)

[0327] Finally, the spatial dimension feature vector is generated:

[0328] Fs=Σαvu·hu

[0329] This feature vector encodes the spatial distribution characteristics of the displayed content, and its dimension may also be 256.

[0330] 3. Spatiotemporal feature fusion layer

[0331] This layer integrates temporal and spatial features through a multi-head attention mechanism Mh. Assume that 8 attention heads are used and the dimension of each head is 32:

[0332] First, project the temporal and spatial features:

[0333] Time characteristics: Ft'=WtFt

[0334] Spatial characteristics: Fs'=WsFs

[0335] Then, each attention head i performs the following operations:

[0336] headi=Attention(WiQFt',WiKFs',WiVFs')

[0337] Among them, WiQ, WiK, and WiV are learnable parameter matrices.

[0338] Finally, the outputs of all attention heads are concatenated and passed through a linear transformation:

[0339] F*=WO[head1;head2;...;head8]

[0340] To prevent the gradient from disappearing, add a residual connection:

[0341] F*=F*+Ft+Fs

[0342] The fused feature vector F* obtained in this way has a dimension of 256 and comprehensively encodes the temporal and spatial features.

[0343] 4. Generate an adaptive display strategy

[0344] Based on the fused feature vector F*, the system generates a specific display strategy:

[0345] First, calculate the content display priority. For each displayable content c, calculate its priority score:

[0346] score(c)=w1·α(c)+w2·β(c)

[0347] in:

[0348] α(c) is the attention score, which is calculated based on the user's current position and gaze focus:

[0349] α(c)=exp(-||p-pc||2 / σp)*exp(-||g-gc||2 / σg)

[0350] β(c) is the interest score, calculated based on the historical stay time:

[0351] β(c)=(1 / Z)Σexp(di / τ)

[0352] w1 and w2 are weight coefficients (such as 0.6 and 0.4)

[0353] Then, optimize the display timing. Construct a Markov decision process:

[0354] Status: the current content combination displayed;

[0355] Action: Switch to a new content combination;

[0356] Reward: Changes in visitor attention and interest;

[0357] Optimize the display sequence through the policy gradient algorithm:

[0358]

[0359] Where J(θ) is the expected reward.

[0360] Finally, generate the display layout plan. For each display:

[0361] 1) Allocate display areas according to priority;

[0362] 2) Design content switching timing;

[0363] 3) Plan transition animation effects;

[0364] For example, for a 4K display, possible layouts are:

[0365] Main area (2160×3840): displays the highest priority content;

[0366] Sidebar (2160×240): displays previews of related content;

[0367] Bottom (120×3840): Displays timeline or progress information;

[0368] Through the above processing, the system can generate an adaptive display strategy based on the real-time behavior of visitors, which not only ensures the consistency of the display theme but also responds to changes in user interests in a timely manner.

[0369] Step 6: Execute the display plan;

[0370] like Figure 5 As shown, multicast addressing encoding is used:

[0371] 1) Construct the coding tree:

[0372] Equipment layer identification display equipment;

[0373] The content type layer distinguishes data types;

[0374] The content item layer identifies the specific content;

[0375] 2) Design addressing format:

[0376] The prefix code identifies the device and type;

[0377] Infix code identification display parameters;

[0378] The suffix code identifies the priority information;

[0379] 3) Execute display control:

[0380] Parse display instructions;

[0381] Scheduling display resources;

[0382] Monitor display performance;

[0383] Record performance data.

[0384] To achieve adaptive display of exhibition hall content, the system uses a multi-layered coding tree structure to organize and manage display resources. This coding tree consists of three levels: the top device layer identifies different display devices; the middle content type layer distinguishes different types of display data (such as images, videos, and text); and the bottom content item layer identifies specific display content. The system assigns a unique identifier to each display device. This allocation mechanism ensures that display instructions are accurately delivered to the target device.

[0385] The system uses a three-segment encoding structure in its content addressing format design. The prefix code primarily identifies the device ID and content type. For example, "DEV001_IMG" represents image-type content on a device with identifier 001. The infix code identifies specific display parameters, including configuration information such as display position, size, and transparency. For example, "POS(x,y)_SIZE(w,h)_ALPHA(0.8)" indicates the content's priority and timing. For example, "PRI5_T20230901" indicates content with a priority of 5 and scheduled for display at a specific time. Based on this encoding format, the system generates a complete addressing table, which enables unified management and efficient indexing of display resources.

[0386] During the execution phase of the presentation plan, the system first needs to parse the presentation instructions. This parsing process deconstructs the encoded instructions into specific operational parameters, including target device, content type, presentation parameters, and timing requirements. Based on the parsed results, the system schedules display resources, including content loading, cache management, and rendering preparation. Simultaneously, the system monitors the presentation in real time, collecting performance data including device status, content playback progress, and user feedback. This data is used to evaluate presentation effectiveness and inform subsequent presentation strategy optimization.

[0387] For example, suppose a demonstration video about aerospace technology is to be displayed on a 4K display device named "DEV001." The system-generated address code might be "DEV001_VID_4K / MainArea_Size(3840,2160)_Priority3_T20230901143000," indicating that the 4K video should be displayed in the main area of device 001, with a display size of 3840×2160, a priority of 3, and a scheduled playback start time of 2:30 PM on September 1, 2023. The system then schedules the corresponding video resource based on this code, controlling its playback start time and location, and monitoring the playback status in real time to ensure the desired presentation quality.

[0388] This multicast addressing and encoding scheme enables flexible scheduling and precise control of display content, while ensuring the stability and reliability of the entire display process. The advantage of this solution lies in its ability to effectively handle the concurrent display requirements of multiple devices and multiple content types, and supports dynamic adjustment based on priority, thus providing visitors with a smooth and coherent display experience.

[0389] like Figure 6 As shown, the present application also provides an exhibition hall content adaptive central control display system, including:

[0390] An acquisition module, configured to acquire and pre-process exhibition hall display data, wherein the exhibition hall display data includes image data stream, video data stream and text data stream;

[0391] A generation module, configured to generate a feature vector of the exhibition hall display data using a multi-layer spiking neural network based on the exhibition hall display data obtained through preprocessing, wherein the multi-layer spiking neural network includes an input layer, a feature extraction layer, a time series integration layer, and an output layer;

[0392] A content association module, configured to output a content theme association matrix through a content association analysis model based on the feature vectors of the exhibition hall display data;

[0393] The acquisition module is further used to acquire real-time user behavior data, wherein the real-time user behavior data includes user location information, sight focus information, and dwell time information;

[0394] The generation module is further configured to construct a display strategy optimization model using a spatiotemporal orthogonal propagation network based on the content topic association matrix and the user behavior data, so as to generate an adaptive display solution through the display strategy optimization model, wherein the spatiotemporal orthogonal propagation network includes a time dimension propagation layer, a space dimension propagation layer, and a spatiotemporal feature fusion layer;

[0395] The display module is used to encode the adaptive display scheme using a multicast addressing coding scheme and transmit it to the exhibition hall display device for display.

[0396] The above modules are connected through a data bus and work together to achieve adaptive display of display content.

[0397] After understanding the principles of the above embodiments, those skilled in the art may make various modifications to these embodiments, but these modifications all fall within the scope of protection of this application.

[0398] The system of the embodiments of the present disclosure can execute the method provided by the embodiments of the present disclosure, and the implementation principles are similar. The actions performed by each module in the system of each embodiment of the present disclosure correspond to the steps in the method of each embodiment of the present disclosure. For the detailed functional description of each module of the system, please refer to the description in the corresponding method shown in the previous text, which will not be repeated here.

[0399] The above description is only an optional implementation method for some implementation scenarios of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present disclosure, other similar implementation methods based on the technical ideas of the present disclosure also fall within the protection scope of the embodiments of the present disclosure.

Claims

1. A method for adaptively displaying exhibition hall contents in a central control, characterized in that: The following steps are involved: Acquiring and preprocessing exhibition hall display data, wherein the exhibition hall display data includes image data stream, video data stream and text data stream; Based on the exhibition hall display data obtained through preprocessing, a multi-layer spiking neural network is used to generate a feature vector of the exhibition hall display data, wherein the multi-layer spiking neural network includes an input layer, a feature extraction layer, a time series integration layer, and an output layer; Based on the feature vectors of the exhibition hall display data, outputting a content theme association matrix through a content association analysis model; Acquire real-time user behavior data, including user location information, visual focus information, and dwell time information; Based on the content topic association matrix and the user behavior data, a display strategy optimization model is constructed using a spatiotemporal orthogonal propagation network, so as to generate an adaptive display solution through the display strategy optimization model, wherein the spatiotemporal orthogonal propagation network includes a time dimension propagation layer, a space dimension propagation layer, and a spatiotemporal feature fusion layer; The adaptive display scheme is encoded using a multicast addressing coding scheme and transmitted to the exhibition hall display device for display.

2. The method according to claim 1, characterized in that Before outputting the content topic association matrix through the content association analysis model, the method further includes: A content association analysis model is constructed by combining the generalized Hebbian learning algorithm, including: constructing a content similarity matrix, designing temporal association strength update rules, and constructing a content topic association matrix.

3. The method according to claim 2, characterized in that The steps of constructing a content similarity matrix, designing a temporal association strength update rule, and constructing a content topic association matrix include: Based on the feature vector, a content similarity matrix is calculated using a cosine similarity function; Based on the preset time series window, a time series association strength update rule is designed to calculate the co-occurrence frequency of content pairs within the window and update the association strength; An association graph is constructed based on the content similarity matrix and the updated association strength, and a community discovery algorithm is used to identify topic clusters to generate the content topic association matrix.

4. The method according to claim 1, wherein Before using a multi-layer pulse neural network to generate a feature vector of the exhibition hall display data, the method further includes: Construct a multi-layer spiking neuron network architecture, wherein the input layer receives a standardized data stream, the feature extraction layer uses a multi-layer LeakyReLU spiking neuron, the time series integration layer uses a bidirectional LSTM unit, and the output layer is used to generate a feature vector; The training of the spiking neuron network is achieved through a weight threshold leakage collaborative training mechanism, which includes: weight updating based on back propagation, dynamic adjustment of neuron firing thresholds, and achieving collaboration between forward activation and back propagation.

5. The method according to claim 4, characterized in that The method of using a multi-layer pulse neural network to generate a feature vector of the exhibition hall display data includes: The multi-layer pulse neural network aligns and fuses different modal data in the feature space to generate a feature vector of the exhibition hall display data.

6. The method according to claim 3, characterized in that The step of constructing an association graph based on the content similarity matrix and the updated association strength, identifying topic clusters using a community discovery algorithm, and generating the content topic association matrix includes: Applying a sparsity constraint and a temporal dependency constraint to the content similarity matrix; Based on the L1 regularization term, the sparsity of the correlation matrix is controlled, and a sparse threshold is set to prune edges with correlation strength less than the threshold; Construct a temporal dependency graph to calculate the temporal centrality of nodes, and adjust the association weight based on the temporal centrality; Updating the association weights using a multi-objective optimization function, wherein the multi-objective optimization function includes a reconstruction error term, a sparsity constraint term, and a temporal consistency constraint term; Based on the updated association weights, a community discovery algorithm is used to identify topic clusters and generate the content topic association matrix.

7. The method according to claim 1, characterized in that The method of constructing a display strategy optimization model based on the content theme association matrix and the user behavior data using a spatiotemporal orthogonal communication network so as to generate an adaptive display solution through the display strategy optimization model includes: The time dimension propagation layer inputs the temporal features of user behavior and generates a time dimension feature vector through the temporal attention mechanism; The spatial dimension propagation layer constructs a spatial graph structure based on the content topic association matrix and generates a spatial dimension feature vector through a graph attention mechanism; The spatiotemporal feature fusion layer integrates the time dimension feature vector and the space dimension feature vector through a multi-head attention mechanism to generate a fused feature vector; The adaptive display scheme is generated based on the fused feature vector, including calculating content display priority, optimizing display timing arrangement, and generating a display layout scheme.

8. The method according to claim 7, characterized in that The generating the adaptive display scheme based on the fused feature vector includes: Calculate attention score based on user’s current location and gaze focus; Estimate interest level based on content dwell time; Calculating a display priority by combining the attention score and the interest level; The display area is dynamically allocated according to the display priority and the content switching timing is optimized.

9. The method according to claim 1, characterized in that The method of encoding the adaptive display scheme by using a multicast addressing coding scheme and transmitting the encoding scheme to an exhibition hall display device for display includes: Construct a coding tree consisting of device layer, content type layer and content item layer, and assign a unique identifier to each display device; Design the content addressing format, including the prefix code for identifying the device ID and content type, the infix code for identifying the presentation parameters, and the suffix code for identifying the priority and timing information, and generate the addressing table; Executing a presentation plan based on the addressing table includes parsing presentation instructions, scheduling display resources, and monitoring presentation effects.

10. An exhibition hall content adaptive central control display system, characterized in that: include: An acquisition module, configured to acquire and pre-process exhibition hall display data, wherein the exhibition hall display data includes image data stream, video data stream and text data stream; A generation module, configured to generate a feature vector of the exhibition hall display data using a multi-layer spiking neural network based on the exhibition hall display data obtained through preprocessing, wherein the multi-layer spiking neural network includes an input layer, a feature extraction layer, a time series integration layer, and an output layer; A content association module, configured to output a content theme association matrix through a content association analysis model based on the feature vectors of the exhibition hall display data; The acquisition module is further used to acquire real-time user behavior data, wherein the real-time user behavior data includes user location information, sight focus information, and dwell time information; The generation module is further configured to construct a display strategy optimization model using a spatiotemporal orthogonal propagation network based on the content theme association matrix and the user behavior data, so as to generate an adaptive display solution through the display strategy optimization model, wherein: The spatiotemporal orthogonal propagation network includes a time dimension propagation layer, a space dimension propagation layer and a spatiotemporal feature fusion layer; The display module is used to encode the adaptive display scheme using a multicast addressing coding scheme and transmit it to the exhibition hall display device for display.

Images