AR interaction method and system for health and epidemic prevention science popularization

Abstract

The invention discloses an AR interaction method and system for health and epidemic prevention science popularization, and relates to the technical field of AR interaction, and the method comprises the steps: constructing an epidemic disease transmission mechanism equation set based on literature retrieval and a parameter card, inputting an interaction instruction in a computer interface, and enabling the interaction instruction to correspond to a protection action and to be mapped into a control quantity of the mechanism equation; the system achieves relation fitting between interactive actions and multi-dimensional parameters of a parameter card through a biaxial evolution knowledge graph network, calculates intuitive parameters in real time, and generates an anti-fact comparison result and an action contribution degree. A result is displayed in a double-track mode on a computer interface, an interaction action state and a mechanism evolution curve are visually displayed, and instant feedback is provided based on an anti-fact difference value. When a complex environmental condition (such as existence of a plurality of local flow fields) is detected, a hybrid calculation module is triggered, a global flow field is generated through local template scaling and attention superposition, and parameters are calibrated. The method improves scientificity and interactivity of epidemic prevention science popularization.

Description

Technical Field

[0001] The present invention relates to the field of AR interaction technology, and in particular to an AR interaction method and system for popularizing health and epidemic prevention science. Background Art

[0002] With the frequent occurrence of public health incidents, society's demand for epidemic prevention knowledge is growing. However, existing epidemic prevention education methods still rely primarily on text-based publicity, video explanations, and passive dissemination, lacking interactivity and personalized feedback. This is particularly true when addressing diverse populations, making it difficult to intuitively demonstrate the causal relationship between protective actions and disease transmission mechanisms. This results in the public's understanding of epidemic prevention measures remaining superficial and lacking scientific validity.

[0003] On the other hand, modern computer technology and augmented reality (AR) have been gradually applied to education and medical popularization. However, most systems remain limited to static visualization or simple action demonstrations, failing to implement mechanism extraction, parameterized modeling, and interactive evolutionary analysis based on scientific literature. For example, while traditional epidemiological models such as SIR and SEIR can simulate macroscopic transmission processes, they cannot accurately couple individual actions with microscopic transmission factors, nor can they provide immediate feedback based on counterfactual comparisons. This makes it difficult for users to truly understand the specific contribution of a particular action to reducing infection risk. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an AR interactive method for health and epidemic prevention popularization to solve the problem of how to couple the mechanism knowledge in epidemic prevention literature with interactive actions to model, realize interactive communication evolution visualization and provide scientific feedback based on counterfactual comparison.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an AR interactive method for popularizing health and epidemic prevention science, which includes parsing relevant literature on epidemic prevention and extracting parameters to form a parameter card; Based on the search of the relevant literature and the parameter card, generating the disease transmission mechanism equation group; Input interactive instructions through the computer interface, the interactive instructions corresponding to the selection or switching of protective actions, and mapped to the control quantity of the epidemic mechanism equation; Injecting the control variables into the disease transmission mechanism equations, calculating the infection probability, exposure dose, air concentration, and surface pathogen load in real time, and generating counterfactual comparison results and action contributions based on the interactive instructions; The interactive action status and mechanism evolution curve are displayed in a dual-track manner in the computer interface, and instant feedback is provided based on the counterfactual difference; During the calculation process of the disease transmission mechanism equation group, when complex environmental conditions are detected, the hybrid calculation module is triggered, and the calculation results are used to calibrate the parameters of the disease transmission mechanism equation group.

[0007] As a preferred solution of the AR interactive method for health and epidemic prevention science popularization described in the present invention, wherein: The relevant literature includes articles on the target disease and a collection of credible sources; Analyze the contents of the relevant literature to obtain the micro-mechanisms and macro-mechanisms of the disease; According to the microscopic mechanism and the macroscopic mechanism, a directly corresponding parameter card is generated; The coupling relationship between microscopic mechanisms, between macroscopic mechanisms, and between microscopic mechanisms and macroscopic mechanisms is used to supplement the coupling relationship of the parameter card.

[0008] As a preferred solution of the AR interactive method for health and epidemic prevention science popularization described in the present invention, wherein: The microscopic mechanism is the individual properties of pathogenic microorganisms, including the following elements: the morphological characteristics of the pathogenic microorganisms, their sensitivity to different daily detergents, their individual size and ability to penetrate different isolation barriers, their survival and decay processes in different environments, and their individual pathogenic processes; The macro-mechanisms include: the spatial spread of the disease, the intensity of the disease response in different populations, and the susceptible population; The generation process of the parameter card is as follows: Step 1: Conduct the first stage of learning by searching the relevant literature. During the first stage of learning, only any mechanism elements and the multidimensional parameters corresponding to the mechanism elements in the microscopic mechanism or the macroscopic mechanism are retrieved. Each retrieval result is used as an evidence, and an anchor point is embedded in each evidence position. The similarity of the retrieval is used as the evidence weight. Generate a parameter file based on the j-dimensional parameters of the mechanism element in each piece of evidence i; integrate all evidence to obtain parameters containing the J dimensions of all evidence; map each parameter file in the J-dimensional parameter space generated by all parameter files, and mark the evidence weight to obtain the parameter card; Where i represents the evidence index, j represents the parameter dimension reflected in the evidence, and J represents the dimension after taking the union of all dimensions of all evidence; Step 2: After locating the anchor points of each piece of evidence during the first phase of learning by searching the relevant literature, we proceed to the second phase of learning: analyzing the coupling relationships between the anchor points and generating candidate edges for rules. The rules include: co-occurrence rules, causal triggering rules, conditional rules, and mutually exclusive or collaborative rules. Meanwhile, the confidence of each edge is taken as the corresponding weight.

[0009] As a preferred scheme of the AR interaction method for health and epidemic prevention popularization, wherein: The epidemic disease transmission mechanism equation set comprises an interactive action obtained in a computer interface and an input, and is coded as an interactive action operator. In the related literature, after the anchor point is positioned, a third stage of learning is performed, in which similar actions to the interactive action d are searched, and a control process or an interactive action similar to the action principle of the interactive action is searched as an action node through transfer learning , and the action operator of the action node is coded according to the expression of each action node in the corresponding literature, thereby generating a vector difference between the action operator of the action node and the interactive action operator coded by the interactive action d as the offset feature of each action node ; wherein e represents the index of the action node, and the e-th action node of the interactive action d represents the e-th action node of the interactive action d. The epidemic disease transmission mechanism equation set is built through a double-axis evolution knowledge graph network to generate the interaction relationship between the interactive action and each dimension parameter in each mechanism element; the double-axis evolution knowledge graph network is composed of two channels of longitudinal flow and transverse flow and a cross-axis fusion head, and is connected through a modulation elimination layer between the two channels.

[0010] As a preferred scheme of the AR interaction method for health and epidemic prevention popularization, wherein: The transverse flow is composed of two layers of graph convolution networks with attention mechanism, takes the bipartite graph of the action node and the parameter card multidimensional parameter node as input, and performs attention modulation through the direction and size of the offset vector, the edge type and the evidence confidence in the information transmission process; is used to learn the structural differences between the interactive action and the parameter card multidimensional parameter and output the dimension-by-dimension transverse adjustment clues; The longitudinal flow is composed of a fully connected neural network with monotonic constraint, takes the action depth as input, can generate a monotonic influence curve between the action depth and the parameter card multidimensional parameter, and introduces the evidence confidence and the condition label when aggregating, so that the evidence that does not meet the applicable conditions is automatically de-weighted or shielded: if the evidence and the condition only partially match, the automatic de-weighting operation is performed; if the evidence and the condition do not match at all, the automatic shielding operation is performed. The modulation elimination layer is placed between the transverse flow and the longitudinal flow, and realizes mutual elimination of the transverse features and the longitudinal features through residual orthogonalization and decorrelation constraint, to avoid coupling interference of the transverse and longitudinal channels in information learning. The cross-axis fusion head consists of a two-layer fully connected network with physical feasible domain projection and vertical and horizontal bidirectional smoothing mechanism at the output stage.

[0011] As a preferred solution of the AR interactive method for health and epidemic prevention science popularization described in the present invention, wherein: The control variable is injected into the disease transmission mechanism equation group to generate the corresponding value of each dimension parameter on the parameter card; and according to the corresponding value, it is mapped into: infection probability, exposure dose, air concentration and surface pathogen load through a preset mapping function; The actual state of the interaction action and the corresponding counterfactual baseline state are simultaneously calculated; the counterfactual baseline state is generated by setting the interaction action corresponding to the interaction instruction to a virtual state of not being executed or having the minimum execution intensity, while keeping other actions unchanged; By comparing the calculation results of the disease transmission mechanism equation group under the actual state and the counterfactual baseline state, the difference in infection probability, exposure dose, air concentration and surface pathogen load is obtained as the counterfactual comparison result; Based on the counterfactual comparison results and the weight of evidence of each dimension parameter in the parameter card, the action contribution of each interactive action is calculated.

[0012] As a preferred solution of the AR interactive method for health and epidemic prevention science popularization described in the present invention, wherein: The complex environmental conditions include the presence of multiple locally different air flow fields during the simulation of disease transmission; The hybrid calculation module includes scaling and superimposing a pre-built local flow field template to make the global flow field conform to the current environment, allocating global attention parameters according to the scaling and superimposition process of the local flow field template, and obtaining a global attention weight through global normalization; The output results of injecting the control amount into the disease propagation mechanism equation group are globally distributed according to attention; The local flow field template includes: assuming that the global space is composed of multiple fixed unit minimum spaces, and obtaining a local typical flow field space by simulating the typical space; each space unit is equipped with an attention parameter, which is proportional to the overall strength of the local flow field template.

[0013] In a second aspect, the present invention provides an AR interactive system for popularizing health and epidemic prevention science, including a collection unit that parses and extracts parameters from relevant documents on epidemic prevention to form a parameter card; An analysis unit generates the disease transmission mechanism equations based on the search of the relevant literature and the parameter card; A control unit inputs interactive instructions through a computer interface, wherein the interactive instructions correspond to the selection or switching of protective actions and are mapped into control quantities of the disease mechanism equation; The feedback unit injects the control quantity into the disease transmission mechanism equation group, calculates the infection probability, exposure dose, air concentration and surface pathogen load in real time, and generates counterfactual comparison results and action contribution based on the interactive instructions; displays the interactive action state and mechanism evolution curve in a dual-track manner in the computer interface, and provides instant feedback based on the counterfactual difference.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the AR interaction method for health and epidemic prevention science popularization as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the AR interaction method for health and epidemic prevention science popularization as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: through literature parsing and parameter card construction, epidemic prevention knowledge is transformed from abstract text into quantifiable mechanism parameters; through the system of equations for the mechanism of disease transmission and the dual-axis evolution knowledge graph network, accurate fitting of interactive actions and multi-dimensional parameters is achieved, so that the scientific effects of protective behaviors can be presented in real time. Compared with the traditional one-way popular science method, the present invention can generate counterfactual comparison results and action contributions, and intuitively reveal "how the risk of infection will change if a certain action is not performed", which greatly enhances the interactivity and persuasiveness of popular science. At the same time, a hybrid computing module is introduced in a complex environment, and a global flow field matrix is ​​generated by superimposing the local template flow field and attention scaling to calibrate the model, effectively improving the calculation accuracy and environmental adaptability. In summary, the present invention not only improves the public's scientific understanding of epidemic prevention measures, but also helps users understand the effects of multi-action combinations, thereby realizing personalized and interactive health and epidemic prevention popular science, and has a high application and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 The figure is a flowchart of the AR interaction method used for health and epidemic prevention science popularization. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0022] Reference Figure 1 , is an embodiment of the present invention, which provides an AR interaction method for health and epidemic prevention science popularization, including the following steps: S1: Analyze and extract parameters from relevant literature on epidemic prevention to form a parameter card.

[0023] Furthermore, the relevant literature includes articles on the target disease, constructing a collection of credible sources. The content of these relevant literature articles is dissected to derive the micro- and macro-mechanisms of the disease's action. First, a credible source collection is established, consisting of articles related to the target disease, selected from medical databases, public health agencies, academic journals, and authoritative guides. By searching and organizing these relevant literature, the system dissects the article content and extracts the micro- and macro-mechanisms involved.

[0024] A directly corresponding parameter card is generated based on the microscopic mechanism and the macroscopic mechanism. At the same time, the coupling relationship between microscopic mechanisms, between macroscopic mechanisms, and between microscopic mechanisms and macroscopic mechanisms is used to supplement the parameter card.

[0025] The microscopic mechanism is the individual attribute of pathogenic microorganisms, including elements such as their morphological characteristics, sensitivity to different daily detergents, individual size and ability to penetrate different isolation barriers, individual survival and decay processes in different environments, and individual pathogenic processes. By extracting the microscopic mechanism, we can provide parameter support for the "pathogen's own characteristics" level for the parameter card. This is used to demonstrate the role of onlookers and achieve epidemic prevention explanations.

[0026] The macro mechanism includes elements: spatial propagation process, the intensity of the epidemic in different groups of people, and susceptible population; the extraction of the macro mechanism provides the parameter card with parameter support at the level of "group transmission and environmental conditions".

[0027] It is to be noted that the purpose of the micro mechanism is to directly convert the "pathogen itself characteristics" into the underlying parameters that can drive the interaction and AR demonstration, so that each operation of the user can produce a rational and visual response at the micro level. After the morphological characteristics, the sensitivity to the detergent, the particle size and the barrier penetration, the survival / decay under different temperature and humidity conditions, and the pathogenic process are unified into controllable quantities and boundary conditions in the parameter card, the system can map these quantities one by one to the source term, the inactivation term, the barrier penetration term and the receptor susceptible site of the mechanism equation. When the user selects or adjusts the action of "cleaning / disinfection, wearing a barrier, changing environmental conditions" and the like on the computer interface, the AR layer will synchronously demonstrate the "microscopic action" process of microorganisms being inactivated, blocked, diluted or still penetrating, and the numerical evolution is constrained by the evidence range and monotonic direction given by the parameter card, so as to realize the explainable explanation of "why this action can reduce the risk", instead of relying on static animation.

[0028] The purpose of the macro mechanism is to gather the influence of "group transmission and environmental conditions" into the control quantity at the scene level, so that the system can give the risk evolution result coupled with the interaction action and the counterfactual comparison at the spatial and population scale. After modeling the transmission path and intensity, the reaction intensity and susceptibility of different groups of people as the ventilation / mixing efficiency, near-field exchange, contact network edge weight and population susceptibility coefficient in the parameter card, the mechanism equation can solve the dynamic curves of "air concentration, surface load, exposure dose and infection probability" on the room, floor or contact graph of the population in real time; when the user superimposes or switches the action, the system can immediately generate the counterfactual baseline of "if the action is not taken" and calculate the contribution of the action. The macro mechanism also provides target quantities (such as local concentration peak, ventilation short circuit, partition difference) for the calibration of complex environments, which facilitates the triggering of the mixing calculation module to generate a global flow field by superimposing the template flow field and attention scaling, and online correction of the model. In this way, the macro layer not only supports situational and personalized popular science interaction, but also ensures that the numerical changes at the micro level can reflect the actual effect at the spatial and population level, forming a "action-mechanism-visualization-counterfactual" closed loop.

[0029] The generation process of the parameter card is as follows: Step 1: Conduct the first stage of learning by searching the relevant literature. During this first stage, only the mechanistic elements and the corresponding multidimensional parameters of the microscopic or macroscopic mechanisms are retrieved. For example, literature describing microbial survival may include temperature, humidity, and survival time. Each search result is considered as an evidence, and an anchor point is embedded at each evidence location. The similarity of the search results is used as the evidence weight.

[0030] A parameter file is generated based on the j-dimensional parameters of the mechanism element in each piece of evidence i; all evidence is integrated to obtain parameters containing J dimensions of all evidence; each parameter file is mapped in the J-dimensional parameter space generated by all parameter files, and the evidence weight is marked to obtain the parameter card.

[0031] Among them, i represents the evidence index, j represents the parameter dimension reflected in the evidence, and J represents the dimension after taking the union of all dimensions of all evidence.

[0032] Step 2: After locating the anchor points of each piece of evidence during the first phase of learning by searching the relevant literature, the second phase of learning begins: analyzing the coupling relationships between anchor points and generating candidate edges for rules. Rules include: co-occurrence rules, causal triggering rules, conditionalization rules, and mutually exclusive or synergistic rules. Co-occurrence rules: Mechanism parameters that frequently appear together in multiple papers, such as the co-occurrence of "increased ventilation rate" and "decreased aerosol concentration." Causal triggering rules: Changes in one mechanism parameter directly lead to significant changes in another, such as "increased temperature" leading to "shortened pathogen survival time." Conditionalization rules: A relationship only holds under specific conditions, such as "the effect of humidity on transmission intensity is only significant in low-temperature environments." Mutually exclusive or synergistic rules: Mutually exclusive or synergistic effects exist between different parameters, such as additive rather than mutually exclusive relationships.

[0033] Each rule generates a candidate edge, assigning different weights based on the confidence and frequency of the literature evidence. Ultimately, the system not only retains the parameter values ​​corresponding to individual mechanism elements in the parameter card, but also uses these rules to supplement the coupling relationships between different parameters, making the parameter card a comprehensive structure that encompasses both value ranges and relationship networks.

[0034] S2: Based on the retrieval of the relevant literature and the parameter card, the set of equations for the disease transmission mechanism is generated.

[0035] The disease transmission mechanism equation group includes obtaining interactive actions with inputs on a computer interface and encoding them into interactive action operators.

[0036] After locating the anchor point in the relevant literature, the third stage of learning is carried out. In the relevant literature, actions similar to the interactive action d are retrieved, and control processes or interactive actions similar to the working principle of the interactive action are retrieved through transfer learning as action nodes. ; and according to the expression of each action node in the corresponding document, it is encoded as an action node The action operator generates the action node The vector difference between the action operator and the interactive action operator encoded by the interactive action d is used as the vector difference between each action node Offset feature .

[0037] Among them, e represents the index of the action node, Represents the e-th action node of the interaction action d. The disease transmission mechanism equations are constructed using a dual-axis evolutionary knowledge graph network, generating the interaction relationship between the interaction actions and each dimensional parameter in each mechanism element. The dual-axis evolutionary knowledge graph network consists of two channels for vertical and horizontal flow, as well as a cross-axis fusion head, connected by a modulation elimination layer.

[0038] First, the user's interactive actions entered into the computer interface are encoded as interactive action operators. This aims to transform previously ambiguous interactive behaviors (such as "open the window for ventilation" or "wear a mask") into numerical inputs that can be manipulated by the mechanistic equations, allowing the actions to directly affect the parameters of the propagation equations. Second, anchor points are located in relevant literature and similar actions are retrieved to provide a control group with scientific evidence for the current interaction. Transfer learning is further extended to control processes with similar operating principles, allowing the system to rely not only on a single description but also integrate evidence from different sources to form a set of "action nodes." The action nodes are encoded as action operators, and the difference between them and the interactive action operators forms an offset feature. This step aims to quantify the difference between the "user's current action" and the "action in existing scientific evidence." This difference includes both direction (the direction of the action's mechanistic effect) and magnitude (the impact of the action's strength on the parameters), providing data support for subsequent relationship fitting. Finally, a dual-axis evolutionary knowledge graph network is introduced to systematically learn complex relationships. The lateral stream focuses on learning the propagation patterns of lateral correlations and offset magnitudes between different action nodes within parameter dimensions; the longitudinal stream focuses on learning the monotonic influence curve between action depth and parameter dimensions. The purpose of the modulation elimination layer is to eliminate the mutual interference between lateral and longitudinal features, ensuring more stable and interpretable learning results. The purpose of the cross-axis fusion head is to fuse the two results into a consistent modulation variable, which is directly used to numerically update the system of equations for the disease transmission mechanism.

[0039] Specifically, the lateral flow input is a bipartite graph consisting of action nodes and parameter dimension nodes (previously, the parameter card represents the parsed results of micro- and macro-mechanisms, with each mechanism element broken down into multiple dimensional parameters; for example, survival rate at temperature, survival rate at humidity, aerosol particle size distribution, crowd density, and ventilation rate). Parameter dimension nodes explicitly represent these dimensional parameters as nodes, each representing an independently adjustable and analyzable numerical dimension). First, embedding preprocessing is performed on both types of nodes. The action semantic vectors obtained from text parsing and the multidimensional vectors of the parameter card are subjected to linear layers and layer normalization, respectively, to obtain stable initial embeddings. Anisotropic attention weights are then calculated on each edge from the action to the parameter dimension, allowing information transfer to be modulated by the offset direction and size, edge type, and evidence confidence. Two layers of graph convolution sequentially perform message aggregation and nonlinear transformation, with residuals and layer normalization added between layers to ensure numerical stability and robustness.

[0040] Attention weight calculation: ; Represents the weight of the message transmitted from the e-th action node to the k-th parameter dimension node; represents the normalized softmax operation on all incoming edges relative to the same receiving node k; represents the learnable coefficient of the directional consistency term; Represents the inner product of the unit offset direction and the desired action direction; Represents the unit vector of the desired action direction obtained by encoding the edge type and condition label; represents the learnable coefficient of the offset amplitude term; Represents a monotonic mapping function to the offset modulus such as logarithmic or linear compression; The learnable coefficient representing the confidence term of the evidence; represents the confidence weight of the evidence of this edge; Indicates the action node index; Indicates the parameter dimension node index; Indicates the unit direction of the offset vector; Indicates the magnitude of the offset vector.

[0041] Aggregate output of the first layer of graph convolution: ; represents the horizontal representation of the parameter dimension node k; Represents a nonlinear activation function such as GELU or ReLU; represents the self-loop update matrix (the self-loop update matrix is ​​an extension of the adjacency matrix with self-loops. It ensures that during graph convolution propagation, nodes retain their own characteristics while absorbing neighbor information, and is used to achieve stable evolution of parameter card dimensions and action nodes); represents the initial embedding of node k; represents the set of action nodes connected to node k; represents the message transformation matrix; represents the concatenation of the action node embedding and its offset vector; represents the embedding of the e-th action node; Represents the offset vector of the e-th action node.

[0042] The second layer of graph convolution reuses the same weighting mechanism but operates on the output of the previous layer to enhance nonlinear representation. The final output of the crossflow is a direction and magnitude clue for each parameter dimension, indicating the direction in which the dimension should be adjusted and the approximate magnitude range. The entire crossflow process is subject to the numerical upper and lower bounds and monotonic direction constraints specified in the parameter card. During training and inference, out-of-bounds components are clipped and sign-preserved to ensure physical feasibility and interpretability.

[0043] The implementation of vertical flow takes as input the depth vector of the interaction action, namely, the user's intensity or compliance settings at each control point. To ensure that physically monotonic effects are captured, such as increased ventilation not causing an increase in air concentration, a multilayer perceptron with monotonic constraints is employed; this constraint is implemented by imposing non-negativity on the weights and using non-negative activations. The vertical flow first normalizes the input depth and clips the feasible domain, then establishes a monotonic influence curve for each parameter dimension. During the aggregation phase, evidence confidence and conditional labels are introduced for amplitude modulation and masking. During the aggregation phase of the vertical flow, a dynamic modulation mechanism based on evidence confidence and conditional labels is introduced to address the varying applicability of different evidence in specific interaction scenarios. This mechanism includes two processing methods: "automatic downgrading" and "automatic masking," which are sequential, rather than parallel, approaches. Automatic downgrading occurs when a piece of evidence only partially matches the current interaction context. For example, if a piece of literature evidence is based on experimental conditions with a humidity of 30%, and the humidity of the current interaction environment is 50%, the system will generate a correction factor less than 1 through the difference function, so that the amplitude contribution of the evidence in the vertical curve aggregation is reduced, but it still retains its reference function. When a piece of evidence is completely inconsistent with the current interaction environment, or there is a logical conflict with other high-confidence evidence, the system will perform an automatic shielding operation. The shielding operation is achieved by directly setting the weight of the evidence to zero, so that the evidence no longer affects the output of the vertical curve. For example, if a piece of evidence is only applicable to "bacterial transmission in an anaerobic environment" and the current interaction scenario is an aerobic environment, the evidence will be completely shielded. For example, if the conditional label shows that the evidence is only applicable to children, and the current interaction object is the elderly, the evidence will also be shielded.

[0044] Through this sequential processing mechanism, the longitudinal flow can first detect whether there is a complete conflict or inapplicable situation during the aggregation process, and if so, perform shielding; if there is no conflict, perform weight reduction according to the difference. This way not only guarantees the scientific rigor and result stability of the longitudinal curve, but also improves the adaptability of the parameter card in different interactive scenarios. The final output of the longitudinal adjustment curve not only meets the monotonicity and physical feasibility constraints, but also has higher explainability and robustness, thereby providing reliable input for dynamic control of the mechanism equation.

[0045] Single-dimensional monotonic mapping: ; represents the adjustment amount component of the longitudinal flow to the parameter dimension k; represents the monotonic mapping function family of dimension k; represents the depth scalar of the interactive action at the control site k; represents the number of hidden units; represents the non-negative coefficient of the rth hidden unit to the output to ensure overall monotonicity; represents a non-negative activation function such as ReLU; represents the non-negative weight to the input; represents the bias.

[0046] Evidence modulation and conditional aggregation: ; represents the normalized weight of the e th action evidence to dimension k; represents the evidence confidence weight; represents an indication of whether the condition label is met, taking zero or one; represents a small positive number for numerical stability; represents the aggregated longitudinal adjustment amount. is not a new independent variable, but a summation index symbol, meaning: in the denominator, all action nodes (represented by the index e') connected to the same parameter dimension k are traversed, and their weights and condition labels are all added up.

[0047] The longitudinal flow output is a dimension-by-dimension adjustment curve or adjustment amount at a given depth; the output is also constrained by the upper and lower bounds of the parameter card and the monotonic direction and will be projected back to the feasible region at the boundary.

[0048] The modulation cancellation layer, located between the horizontal and vertical layers, cancels the predictable components of the two paths before entering the output, preventing statistical confounding between structural differences and depth effects. This is achieved using bidirectional residual orthogonalization supplemented by a decorrelation loss.

[0049] Bidirectional residual orthogonalization: ; Represents the horizontal representation after mutual cancellation; Represents the horizontal representation before mutual cancellation; Represents the learnable linear mapping matrix from the longitudinal subspace to the transverse subspace; It represents the vertical representation after mutual cancellation; It represents the vertical representation before mutual cancellation; Represents the learnable linear mapping matrix from the horizontal subspace to the vertical subspace.

[0050] Decorrelation loss: ; represents the de-correlation penalty term; represents the empirical covariance matrix between the horizontal and vertical representations after mutual cancellation; represents the Frobenius norm.

[0051] After processing by this layer, two non-interfering residual representations are obtained for subsequent fusion.

[0052] Implementation of the cross-axis fusion head: The fusion head receives the representations after two-way mutual cancellation, first performs feature splicing, and then regresses the adjustment amount of each parameter dimension through a two-layer fully connected network. The output is then projected into the physical feasible domain and smoothed bidirectionally, ensuring that the result complies with the mechanism constraints and has the stability of interactive presentation.

[0053] Fusion and projection: ; Represents the final adjustment of dimension k; Proj represents the projection operation implemented according to the upper and lower bounds and monotonicity requirements given by the parameter card, including clipping and sign constraints; Represents the fusion layer weight matrix; Represents the vector concatenation of two representations; Represents the fusion layer bias.

[0054] Longitudinal smoothing: ; Indicates the adjustment amount after time or interactive step smoothing; Indicates that the smoothing coefficient takes a value between zero and one; Represents the smoothed value of the previous moment or the previous interaction step; Indicates the current fusion output.

[0055] Horizontal Smoothing: ; Represents the final adjustment vector after consistency constraint is applied on the parameter dimension subgraph; represents the identity matrix with the same number of parameter dimensions; in represents the graph smoothing intensity coefficient which is non-negative and not greater than one; The undirected Laplacian matrix representing the parameter-dimensional subgraph; Represents the smoothed stacked vector of all dimensions.

[0056] The final adjustment amount is written into the corresponding coefficients or boundaries of the mechanism equation in the form of a control site mapping table, driving the update of indicators such as infection probability, exposure dose, air concentration and surface load in real time, while providing input for the calculation of counterfactual baseline and action contribution.

[0057] The advantage of this structure is that it simultaneously captures the "correspondence between actions and parameters" and the "continuous impact of action intensity on parameters," without the disconnected nature of simple mappings between actions and parameters in traditional methods. The lateral flow uses bipartite graph convolution to finely model the differential information across different action and parameter dimensions. This allows for the combined effects of offset direction, offset magnitude, edge relationships, and evidence confidence, avoiding the rigidity of relying solely on empirical rules to set weights, resulting in results that are closer to actual mechanisms.

[0058] Vertical flow utilizes a monotonic constrained neural network to ensure that the depth changes of interactive actions maintain physical consistency in the numerical mapping. This prevents counterintuitive fluctuations in model outputs such as infection probability and air concentration when users adjust the intensity of their actions on the interface, significantly enhancing the system's reliability and trustworthiness.

[0059] The introduction of a modulation cancellation layer solves the problem of horizontal and vertical coupling interference. In conventional methods, structural differences and depth effects are often confused, easily leading to repeated calculations or bias amplification. By orthogonalizing the residuals and applying decorrelation constraints, this layer assigns the horizontal layer to difference learning and the vertical layer to intensity learning, ensuring a clear division of labor and non-redundant information representation, thereby improving model convergence speed and interpretability.

[0060] The advantage of the cross-axis fusion head is that it not only integrates both horizontal and vertical features, but also ensures reasonable and visually friendly results through physically feasible domain projection and bidirectional smoothing. This prevents parameter out-of-bounds and numerical anomalies while also ensuring smooth and coherent output curves in AR displays, enhancing the user's interactive experience.

[0061] In summary, this architecture offers advantages over existing technologies in that it achieves high-precision matching of action and mechanism parameters, maintains consistency in physical laws, eliminates interference from learning across different dimensions, and ultimately outputs interpretable results that are both scientifically sound and suitable for interactive presentation. This allows epidemic prevention education to move beyond demonstration and provide users with intuitive and reliable feedback in a data-driven manner.

[0062] S3: Input interactive instructions through the computer interface. The interactive instructions correspond to the selection or switching of protective actions and are mapped into control quantities of the epidemic mechanism equation.

[0063] User actions on the computer interface are first captured as selection signals, which carry the control identifier, raw value, and event type. The system debounces and reshapes these selection signals to ensure only valid actions are retained. Using a control-action mapping table, these signals are converted into corresponding action semantics, such as "mask selection," "window ventilation," or "surface cleaning." Subsequently, the raw values ​​are normalized or converted to different levels, unifying different inputs into intensities or levels. For example, slider displacements are normalized to a range of 0–1, and mask models are converted to different protection levels. The system further incorporates the scene context to bind actions to specific areas or objects. Parameter cards are used for range checking and conflict resolution to ensure that the generated signals are both numerically and logically consistent with reality. Finally, the system assigns a confidence level to the action and encapsulates elements such as action type, intensity, scope, and confidence level into interaction signals. Thus, the selection signals on the interface are converted into interaction signals that can be directly invoked by the disease transmission mechanism equations, achieving an automatic mapping from user actions to mechanism parameter controls.

[0064] S4: Injecting the control quantity into the disease transmission mechanism equation group, calculating the infection probability, exposure dose, air concentration and surface pathogen load in real time, and generating counterfactual comparison results and action contribution based on the interactive instructions.

[0065] The control quantity is injected into the disease transmission mechanism equation group to generate the corresponding numerical value of each dimensional parameter on the parameter card; and according to the corresponding numerical value, it is mapped into: infection probability, exposure dose, air concentration and surface pathogen load through a preset mapping function.

[0066] The microscopic layer, centered on pathogen characteristics, drives processes such as inactivation, size-dependent penetration, and attenuation on various surfaces and environments. This is used to present "close-up mechanism animations" of pathogenicity and protection in AR. This layer does not involve complex spatial environments and therefore does not trigger the hybrid computing module. The macroscopic layer, centered on spatial and population transmission, continuously calculates near- and far-field air concentrations, surface pathogen loads caused by deposition, contact-resuspension pathways, and removal processes resulting from changes in ventilation and filtration. Control variables act in real time on corresponding coefficients or boundary conditions, forming an evolutionary trajectory that updates with user interaction. At each moment (or interaction step), the mechanism state variables are mapped into four user-interpretable metrics: air concentration, surface pathogen load, exposure dose, and infection probability. Air concentration is directly derived from the near- and far-field states of the macroscopic layer; surface pathogen load is derived from the dynamic balance between deposition and inactivation; exposure dose is derived from the temporal accumulation of air concentration and contact pathways within the user's domain; and infection probability is derived from the risk value obtained by substituting the dose into a pre-set dose-response relationship. The system unifies the units, thresholds, and color codes these outputs, providing ready-to-use data for AR overlay and dual-track curve display.

[0067] During the calculation of the disease transmission mechanism equations, if complex environmental conditions are detected, the hybrid computation module is triggered and the parameters of the disease transmission mechanism equations are calibrated using the computational results. These complex environmental conditions include the presence of multiple, locally distinct air flow fields during the disease transmission simulation (the AR process implemented in this solution encompasses both the microscopic pathogenic and epidemic prevention processes, which do not involve complex environments, and the macroscopic transmission and epidemic prevention processes, which may involve transmission in complex environments).

[0068] The system continuously monitors for complex environments. When the scene contains features such as heterogeneous ventilation, localized airflow channels, obstructed partitions, and multiple sources, or when the deviation between the macro-level prediction and the observation / empirical template exceeds a threshold, it determines the presence of multiple, locally distinct air flow fields. At this point, the hybrid computing module is triggered. If the system is in the purely microscopic explanation phase (close-range disease and protection), it is not triggered to ensure smooth performance and interaction.

[0069] The hybrid calculation module includes scaling and superimposing the pre-built local flow field template to make the global flow field conform to the current environment. According to the scaling and superimposition process of the local flow field template, the global attention parameter is allocated, and the global attention weight is obtained through global normalization. The output result of injecting the control amount into the disease transmission mechanism equation group is globally distributed according to the attention. The local flow field template includes assuming that the global is composed of multiple fixed unit minimum spaces, and obtaining a local typical flow field space by simulating the typical space; each spatial unit is equipped with an attention parameter, which is proportional to the overall strength of the local flow field template.

[0070] The specific implementation process of the hybrid computing module is as follows: 1. Local template extraction and scaling. The system divides the global space into multiple fixed minimum space units (grid or room partition), and pre-establishes a local flow field template library for each type of typical space (including typical patterns such as air inlet / outlet, window door opening, fan jet, thermal plume, etc.). In runtime, the template is scaled in size and intensity according to the current scene parameters and action intensity, so that the unit-level flow field matches the current environment.

[0071] 2. Template superposition to generate global flow field. Align and superimpose the local flow fields of each unit in the global coordinates, and perform consistency correction and boundary continuity check during superposition to ensure that the overall field domain has no obvious discontinuity.

[0072] 3. Global attention allocation. Assign an attention parameter to each unit, which is related to the overall intensity of the unit template, the position relationship with the source, the ventilation path, the obstacle shielding, and the action scope matching degree; then perform global normalization to obtain the global attention weight of each unit at the current time.

[0073] 4. Global distribution of mechanism output according to attention. Distribute and reweight the output of the mechanism equation set or the component affected by the controlled quantity in space according to the global attention weight, to obtain a concentration and deposition distribution that better fits the real multi-flow field conditions.

[0074] First, it ensures accuracy under multi-flow field conditions. Conventional macroscopic propagation models often assume uniform mixing of space air, but in reality, there are often heterogeneous ventilation, obstacles, and multiple source superposition, resulting in significant local concentration differences. If a single global parameter is used for estimation, the output air concentration and surface deposition will deviate from reality. Through the local template scaling and superposition mechanism, the real distribution of multi-flow field superposition can be flexibly restored, making the simulation results more consistent with the actual scene. Second, it improves the adaptability and robustness of the model. Through the attention parameter mechanism, the system can automatically identify which local spaces have a greater impact on the overall propagation results, such as air flow channels and major personnel gathering areas, and dynamically allocate calculation weights. This global attention allocation not only avoids resource waste caused by uniform processing, but also ensures that the propagation and protection effects of key areas are highlighted. Third, it provides the ability to calibrate and feedback parameters. Finally, it optimizes the coherence of AR interaction experience. The actions taken by the user in the interface, such as opening windows, turning on fans, or moving positions, often change the local flow field. The hybrid computing module can immediately reflect the effects of these operations in the global flow field, and through attention weighting, the real propagation distribution is smoothly presented on the visualization, avoiding abrupt jumps in values or images, thereby enhancing the immersion and credibility of the interaction.

[0075] S5: Display the interactive action status and mechanism evolution curve in a dual-track manner in the computer interface, and provide instant feedback based on the counterfactual difference.

[0076] In this step, the system first calculates the user's interactive action in real time, obtaining the actual state including infection probability, exposure dose, air concentration, and surface pathogen load. Simultaneously, the system automatically constructs a counterfactual baseline state corresponding to the interactive action. This counterfactual baseline is generated by setting the action corresponding to the interactive instruction to "not executed" or "minimum intensity execution" while keeping the remaining actions and environmental parameters unchanged, thus forming a control scenario in which only the impact of the action is missing.

[0077] After completing the baseline generation, the system parallelizes the disease transmission equations for both the actual state and the counterfactual baseline, generating two sets of results. By comparing the differences between the two sets of results, the system can determine the differences in four indicators: the reduction in infection probability, the degree of exposure dose reduction, the dilution of airborne pathogen concentration, and the attenuation of surface pathogen load. These differences, known as the counterfactual comparison results, provide a clear picture of the protective effect of the interaction under current conditions.

[0078] To further quantify the importance of interactive actions, the system combines counterfactual comparison results with the multi-dimensional parameters in the parameter card and performs a weighted analysis using the weight of evidence. The weight of evidence is derived from the credibility and applicability assessment of each parameter during the literature analysis phase, thus ensuring the scientific nature of the contribution calculation. In this way, the system calculates the contribution of each interactive action to the overall protection process, expressed as a numerical value or percentage, to reflect the relative value of the action in reducing the risk of transmission.

[0079] In the computer interface, the system uses a dual-track visualization. The upper track displays the actual state and evolution of interactive actions, clearly indicating the user's current actions and their intensity or duration. The lower track dynamically displays the mechanism's evolution as a curve, plotting both the actual state and the counterfactual baseline results. Shading or difference annotations highlight the immediate protective effect of the action. This allows users to intuitively understand how the transmission risk would change if the action was not performed.

[0080] Through this "actual state - counterfactual baseline - difference feedback" mechanism, this invention not only achieves a quantitative display of the effects of interactive actions, but also ensures the immediacy and scientific nature of feedback. Users can clearly see the value of their actions during interaction, thereby enhancing the intuitiveness and persuasiveness of epidemic prevention education.

[0081] This embodiment also provides an AR interactive system for popularizing health and epidemic prevention science, including: The collection unit analyzes and extracts parameters from relevant documents on epidemic prevention to form a parameter card.

[0082] The analysis unit generates the set of equations for the disease transmission mechanism based on the retrieval of the relevant literature and the parameter card.

[0083] The control unit inputs interactive instructions through a computer interface. The interactive instructions correspond to the selection or switching of protective actions and are mapped into control quantities of the epidemic mechanism equation.

[0084] The feedback unit injects the control quantity into the disease transmission mechanism equation group, calculates the infection probability, exposure dose, air concentration and surface pathogen load in real time, and generates counterfactual comparison results and action contribution based on the interactive instructions; displays the interactive action state and mechanism evolution curve in a dual-track manner in the computer interface, and provides instant feedback based on the counterfactual difference.

[0085] This embodiment also provides a computer device suitable for the AR interaction method for health and epidemic prevention science popularization, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the AR interaction method for health and epidemic prevention science popularization proposed in the above embodiment.

[0086] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0087] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the AR interaction method for health and epidemic prevention popularization proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0088] To sum up, the present application converts unstructured knowledge into computable data base by analyzing micro and macro mechanisms from epidemic prevention documents and constructing parameter cards containing numerical ranges, evidence levels and coupling relationships. Based on parameter cards and action evidence anchors, the present application establishes epidemic disease transmission mechanism equation sets and learns the relationship between actions and multi-dimensional parameters with “interactive action operator-action node offset characteristics-biaxial evolution knowledge graph network” to ensure the explainable mapping of actions to mechanisms. The present application normalizes selection signals into interactive signals as control variables and directly injects them into mechanism equations to realize real-time adjustment of source, ventilation, sedimentation and contact control sites. The present application normalizes mechanism outputs into four indexes of air concentration, surface pathogen load, exposure dose and infection probability and generates instant feedback and action contribution degree with “actual state-counterfactual baseline-difference” mechanism to intuitively reveal the protection value of single or combined actions. When detecting complex environmental conditions such as heterogeneous ventilation, air flow channel and partition barriers, the present application triggers a hybrid computing module to generate a global flow field by scaling and superimposing local flow field templates and global attention distribution to ensure accuracy and stability in complex scenarios. The present application synchronously displays “action state trajectory” and “mechanism evolution curve” with double-track visualization and highlights counterfactual difference with color scale and label prompt to form a public-oriented, interactive, quantifiable and traceable epidemic prevention popularization experience.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An AR interactive method for health and epidemic prevention science popularization, characterized by: This includes parsing relevant literature on epidemic prevention and extracting parameters to form parameter cards; Based on the search of the relevant literature and the parameter card, a set of equations for the disease transmission mechanism is generated; Input interactive instructions through the computer interface, the interactive instructions corresponding to the selection or switching of protective actions, and mapped to the control quantity of the epidemic mechanism equation; Injecting the control variables into the disease transmission mechanism equations, calculating the infection probability, exposure dose, air concentration, and surface pathogen load in real time, and generating counterfactual comparison results and action contributions based on the interactive instructions; The interactive action status and mechanism evolution curve are displayed in a dual-track manner in the computer interface, and instant feedback is provided based on the counterfactual difference; During the calculation process of the disease transmission mechanism equation group, when complex environmental conditions are detected, the hybrid calculation module is triggered, and the calculation results are used to calibrate the parameters of the disease transmission mechanism equation group.

2. The AR interactive method for health and epidemic prevention science popularization according to claim 1, characterized in that: The relevant literature includes articles on the target disease and a collection of credible sources; Analyze the contents of the relevant literature to obtain the micro-mechanisms and macro-mechanisms of the disease; According to the microscopic mechanism and the macroscopic mechanism, a directly corresponding parameter card is generated; The coupling relationship between microscopic mechanisms, between macroscopic mechanisms, and between microscopic mechanisms and macroscopic mechanisms is used to supplement the coupling relationship of the parameter card.

3. The AR interactive method for health and epidemic prevention science popularization according to claim 2, characterized in that: The microscopic mechanism is the individual properties of pathogenic microorganisms, including the following elements: the morphological characteristics of the pathogenic microorganisms, their sensitivity to different daily detergents, their individual size and ability to penetrate different isolation barriers, their survival and decay processes in different environments, and their individual pathogenic processes; The macro-mechanisms include: the spatial spread of the disease, the intensity of the disease response in different populations, and the susceptible population; The generation process of the parameter card is as follows: Step 1: Conduct the first stage of learning by searching the relevant literature. During the first stage of learning, only any mechanism elements in the microscopic mechanism or the macroscopic mechanism and the multidimensional parameters corresponding to the mechanism elements are retrieved. Each retrieval result is used as an evidence, and an anchor point is embedded at each evidence position. Use the similarity of the retrieval as the weight of evidence; Generate a parameter file based on the j-dimensional parameters of the mechanism element in each piece of evidence i; integrate all evidence to obtain parameters containing the J dimensions of all evidence; map each parameter file in the J-dimensional parameter space generated by all parameter files, and mark the evidence weight to obtain the parameter card; Where i represents the evidence index, j represents the parameter dimension reflected in the evidence, and J represents the dimension after taking the union of all dimensions of all evidence; Step 2: After locating the anchor points of each piece of evidence during the first phase of learning by searching the relevant literature, we proceed to the second phase of learning: analyzing the coupling relationships between the anchor points and generating candidate edges for rules. The rules include: co-occurrence rules, causal triggering rules, conditional rules, and mutually exclusive or collaborative rules. At the same time, the confidence of each edge is used as the corresponding weight.

4. The AR interactive method for health and epidemic prevention science popularization according to claim 3, characterized in that: The disease transmission mechanism equations include obtaining interactive actions with inputs on a computer interface and encoding them into interactive action operators; After locating the anchor point in the relevant literature, the third stage of learning is carried out. In the relevant literature, actions similar to the interactive action d are retrieved, and control processes or interactive actions similar to the working principle of the interactive action are retrieved through transfer learning as action nodes. ; And according to the expression of each action node in the corresponding document, it is encoded as an action node The action operator generates the action node The vector difference between the action operator and the interactive action operator encoded by the interactive action d is used as the vector difference between each action node Offset feature ; Among them, e represents the index of the action node, represents the e-th action node of the interaction action d; The disease transmission mechanism equation group is constructed through a dual-axis evolutionary knowledge graph network to generate the interactive relationship between interactive actions and each dimensional parameter in each mechanism element; the dual-axis evolutionary knowledge graph network consists of two channels of longitudinal flow and lateral flow and a cross-axis fusion head, and the two channels are connected by a modulation elimination layer.

5. The AR interactive method for health and epidemic prevention science popularization according to claim 4, characterized in that: The lateral flow consists of a two-layer graph convolutional network with an attention mechanism. It takes a bipartite graph of action nodes and parameter card multidimensional parameter nodes as input. During the information transmission process, attention is modulated by the direction and size of the offset vector, the edge type, and the confidence of the evidence. It is used to learn the structural differences between the interactive actions and the multidimensional parameters of the parameter card and output dimension-by-dimension lateral adjustment clues. The vertical flow is composed of a fully connected neural network with monotonic constraints. It takes action depth as input and can generate a monotonic influence curve between action depth and multidimensional parameters of the parameter card. It also introduces evidence confidence and condition labels during aggregation, so that evidence that does not meet the applicable conditions is automatically downgraded or blocked: if the evidence only partially matches the conditions, the automatic downgrade operation is performed; if the evidence does not match the conditions at all, the automatic blocking operation is performed. The modulation elimination layer is placed between the horizontal flow and the vertical flow, and realizes the mutual elimination of horizontal and vertical features through residual orthogonalization and decorrelation constraints, thereby avoiding the coupling interference of horizontal and vertical channels in information learning; The cross-axis fusion head consists of a two-layer fully connected network with physical feasible domain projection and vertical and horizontal bidirectional smoothing mechanism at the output stage.

6. The AR interactive method for health and epidemic prevention science popularization according to claim 5, characterized in that: The control variable is injected into the disease transmission mechanism equation group to generate the corresponding value of each dimension parameter on the parameter card; and according to the corresponding value, it is mapped into: infection probability, exposure dose, air concentration and surface pathogen load through a preset mapping function; The actual state of the interaction action and the corresponding counterfactual baseline state are simultaneously calculated; the counterfactual baseline state is generated by setting the interaction action corresponding to the interaction instruction to a virtual state of not being executed or having the minimum execution intensity, while keeping other actions unchanged; By comparing the calculation results of the disease transmission mechanism equation group under the actual state and the counterfactual baseline state, the difference in infection probability, exposure dose, air concentration and surface pathogen load is obtained as the counterfactual comparison result; Based on the counterfactual comparison results and the weight of evidence of each dimension parameter in the parameter card, the action contribution of each interactive action is calculated.

7. The AR interactive method for health and epidemic prevention science popularization according to claim 6, characterized in that: The complex environmental conditions include the presence of multiple locally different air flow fields during the simulation of disease transmission; The hybrid calculation module includes scaling and superimposing a pre-built local flow field template to make the global flow field conform to the current environment, allocating global attention parameters according to the scaling and superimposition process of the local flow field template, and obtaining a global attention weight through global normalization; The output results of injecting the control amount into the disease propagation mechanism equation group are globally distributed according to attention; The local flow field template includes: assuming that the global space is composed of multiple fixed unit minimum spaces, and obtaining a local typical flow field space by simulating the typical space; each space unit is equipped with an attention parameter, which is proportional to the overall strength of the local flow field template.

8. An AR interactive system for popularizing health and epidemic prevention science, based on the AR interactive method for popularizing health and epidemic prevention science according to any one of claims 1 to 7, characterized in that: Including, a collection unit, which uses relevant documents on epidemic prevention to parse and extract parameters to form a parameter card; An analysis unit generates the disease transmission mechanism equations based on the search of the relevant literature and the parameter card; A control unit inputs interactive instructions through a computer interface, wherein the interactive instructions correspond to the selection or switching of protective actions and are mapped into control quantities of the disease mechanism equation; The feedback unit injects the control quantity into the disease transmission mechanism equation group, calculates the infection probability, exposure dose, air concentration and surface pathogen load in real time, and generates counterfactual comparison results and action contribution based on the interactive instructions; displays the interactive action state and mechanism evolution curve in a dual-track manner in the computer interface, and provides instant feedback based on the counterfactual difference.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the AR interaction method for health and epidemic prevention science popularization according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AR interaction method for health and epidemic prevention science popularization according to any one of claims 1 to 7 are implemented.

Images