Computer vision assisted intelligent campus security situation intelligent perception method and system

Abstract

This application discloses a computer vision-assisted intelligent perception method and system for smart campus security situation, relating to the field of image data processing technology. The method includes: collecting temporal visual data from multiple monitoring areas within the campus, identifying abnormal behavior patterns, and generating a behavioral semantic graph; constructing a dynamic security influence factor graph network based on the behavioral semantic graph and the topology of the monitoring areas; dividing security communities and analyzing key nodes according to the dynamic security influence factor graph network, identifying core influence nodes and corresponding security situation indicators in multiple security communities; integrating the security situation indicators of the core influence nodes with a pre-set campus security strategy library, and using a multi-objective optimization algorithm to make adaptive scheduling decisions for security resources, outputting a real-time security early warning scheme. This solves the technical problems of limited monitoring range, slow response speed, and insufficient data analysis capabilities in existing campus security management models.

Description

Technical Field

[0001] This application relates to the field of image data processing technology, specifically to a computer vision-assisted intelligent perception method and system for smart campus security situation. Background Technology

[0002] With the continuous expansion of campuses and the increasing number of students, campus safety has become an increasingly important concern. However, traditional campus safety management models mainly rely on manual patrols and fixed monitoring equipment, which have problems such as limited monitoring range, slow response speed, and insufficient data analysis capabilities, making it difficult to meet the refined and intelligent safety management needs of modern smart campuses. Summary of the Invention

[0003] This application provides a computer vision-assisted intelligent perception method and system for smart campus security situation, which solves the technical problems of limited monitoring range, slow response speed and insufficient data analysis capabilities in the existing campus security management model.

[0004] The technical solution to the above-mentioned technical problems in this application is as follows:

[0005] Firstly, this application provides a computer vision-assisted intelligent perception method for security situation in smart campuses, the method comprising:

[0006] Collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on a dynamic scene semantic segmentation model, identify abnormal behavior patterns, and generate a behavior semantic map.

[0007] Based on the behavioral semantic graph and the topology of the monitoring area, a dynamic security influence factor graph network is constructed.

[0008] Based on the dynamic security impact factor graph network, security communities are divided and key nodes are analyzed to identify core impact nodes and corresponding security status indicators in multiple security communities.

[0009] By integrating the security status indicators of the core impact nodes and the preset campus security strategy library, a multi-objective optimization algorithm is used to make adaptive scheduling decisions for security resources and output a real-time security early warning scheme.

[0010] Secondly, this application provides a computer vision-assisted intelligent perception system for smart campus security situation, including:

[0011] The information acquisition module is used to collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on a dynamic scene semantic segmentation model, identify abnormal behavior patterns, and generate a behavior semantic map.

[0012] The network construction module is used to construct a dynamic security influence factor graph network based on the behavioral semantic graph and the topology of the monitoring area;

[0013] The node analysis module is used to divide security communities and analyze key nodes based on the dynamic security impact factor graph network, and identify the core impact nodes and corresponding security status indicators in multiple security communities.

[0014] The solution output module is used to integrate the security status indicators of the core impact nodes and the preset campus security strategy library, and to make adaptive scheduling decisions for security resources through a multi-objective optimization algorithm, and output a real-time security early warning solution.

[0015] This application provides one or more technical solutions, which have at least the following technical effects or advantages:

[0016] This application provides a computer vision-assisted intelligent perception method and system for smart campus security situation. First, it utilizes a high-definition camera array to synchronously collect multi-view temporal visual data. Combined with a dynamic scene semantic segmentation model, it achieves pixel-level semantic parsing, extracting individual behavioral trajectories and group interaction features. This effectively identifies abnormal behavior patterns and constructs a behavioral semantic graph, overcoming the limitations of traditional monitoring systems. Second, it constructs a dynamic security influence factor graph network based on a graph attention network architecture. Through an attention mechanism, it dynamically learns the security influence propagation weights between regions. Combined with a modularity optimization algorithm, it achieves security community division and key node analysis, identifying core influence nodes and corresponding security situation indicators, thus improving risk propagation prediction capabilities. Finally, by integrating the security situation indicators of core influence nodes with a preset campus security strategy library, it employs a multi-objective optimization algorithm for adaptive scheduling decisions of security resources. This enables real-time generation of dynamic early warning schemes that include risk propagation capabilities, impact range, and emergency response priorities. This solves the problems of slow response speed and insufficient data analysis capabilities in traditional management models, providing a full-process intelligent solution for smart campus security management.

[0017] Through the above technical solutions, this application can improve the level of intelligence in campus safety management, increase the accuracy of abnormal behavior identification, shorten the response time of safety incidents, and effectively reduce the incidence of campus safety accidents. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is a flowchart illustrating the computer vision-assisted intelligent perception method for smart campus security situation provided in the embodiments of this application;

[0020] Figure 2 This is a schematic diagram of the structure of the computer vision-assisted smart campus security situation intelligent perception system provided in the embodiments of this application.

[0021] The components represented by each number in the attached diagram are explained below:

[0022] Information acquisition module 11, network construction module 12, node analysis module 13, and solution output module 14. Detailed Implementation

[0023] This application provides a computer vision-assisted intelligent perception method and system for smart campus security situation, which addresses the technical problems of limited monitoring range, slow response speed, and insufficient data analysis capabilities in existing campus security management models.

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0026] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0027] Example 1, as Figure 1As shown in the embodiments of this application, a computer vision-assisted intelligent perception method for smart campus security situation is provided, including:

[0028] S10: Collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on a dynamic scene semantic segmentation model, identify abnormal behavior patterns, and generate a behavior semantic map.

[0029] In this embodiment, firstly, a high-definition camera array is deployed in various areas of the campus. This camera array supports synchronous triggering and data alignment, ensuring time consistency across different monitored areas. The collected raw visual data first undergoes a preprocessing module for denoising, enhancement, and standardization to improve the accuracy of subsequent analysis.

[0030] Secondly, a dynamic scene semantic segmentation model is used to perform multi-scale behavior analysis on the preprocessed data. This model employs a deep convolutional neural network architecture combined with an attention mechanism, enabling it to adaptively identify behavioral features in different scenarios. Through pixel-level semantic segmentation, individual behavioral trajectories and group interaction features are extracted, thereby identifying abnormal behavioral patterns such as running, fighting, and gathering.

[0031] Finally, the identified abnormal behavior patterns and their spatiotemporal information are integrated into a behavioral semantic graph, providing basic data support for subsequent security situation analysis.

[0032] Specifically, step S10 in the method includes:

[0033] Multi-view temporal visual data is collected simultaneously using a high-definition camera array deployed on campus.

[0034] Based on a dynamic scene semantic segmentation model, pixel-level semantic parsing is performed on the temporal visual data to extract individual behavior trajectories and group interaction features. The individual behavior trajectories include movement speed and direction of movement, and the group interaction features include individual spacing and clustering density.

[0035] Based on a pre-defined abnormal behavior knowledge base, the individual behavior trajectories and group interaction features are matched to mark abnormal behavior events and construct a behavioral semantic graph.

[0036] In this embodiment, firstly, in selecting the high-definition camera array, the environmental characteristics and monitoring needs of different areas of the campus are considered. For example, in open areas such as playgrounds and squares, cameras with wide viewing angles and high resolution are selected to ensure that they can cover a large area and clearly capture people's behavior and actions. In indoor areas such as teaching buildings and libraries, cameras with low illumination and wide dynamic range are selected to adapt to changes in indoor light and ensure that high-quality visual data can be obtained under different lighting conditions.

[0037] Multi-view time-series visual data is synchronously acquired using clock synchronization technology and data transmission protocols to ensure strict time alignment of data collected by each camera, avoiding behavioral analysis errors caused by time differences. Simultaneously, to guarantee the real-time performance and stability of the data, high-speed network transmission equipment is used to rapidly transmit the acquired data to the data processing center.

[0038] Furthermore, when performing pixel-level semantic parsing based on the dynamic scene semantic segmentation model, the model continuously learns and adapts to changes in various scenes within the campus. For example, with the changing seasons, the campus landscape and people's clothing will differ; different teaching activities and campus cultural activities will also bring about changes in people's behavioral patterns. The model dynamically adjusts its parameters and feature extraction methods according to these changes to improve the accuracy of recognizing individual behavioral trajectories and group interaction features.

[0039] When extracting individual behavioral trajectories, in addition to focusing on movement speed and direction, it is important to analyze the coherence and periodicity of the behavior. For example, normal walking behavior usually exhibits certain regularity and coherence, while abnormal behavior may manifest as sudden acceleration, deceleration, or abrupt changes in direction. For group interaction characteristics, in addition to individual spacing and clustering density, the dynamic changes of the group should be considered, such as the group's tendency to disperse and aggregate.

[0040] By comparing real-time extracted individual behavioral trajectories and group interaction features with predefined rules in an abnormal behavior knowledge base, when the feature data matches a specific abnormal pattern, such as exceeding speed limits or excessive density, it is marked as an abnormal behavior event. Using the marked abnormal events as nodes and the temporal, spatial, and logical connections between nodes as edges, a behavioral semantic graph representing the distribution and transmission path of safety hazards is constructed.

[0041] Specifically, when labeling anomalous behavioral events, they are graded according to their severity and potential harm. For example, minor pushing or shoving is labeled as low-level anomalous behavior, while violent fighting is labeled as high-level anomalous behavior. Different levels of anomalous behavior are distinguished by different identifiers and parameters in the behavioral semantic map.

[0042] In the process of constructing a behavioral semantic graph, abnormal behavior events are also associated with corresponding monitoring areas, times, and other information to improve the accuracy of abnormal behavior event labeling.

[0043] The construction steps of the dynamic scene semantic segmentation model include:

[0044] Historical time-series visual data from multiple monitoring areas on campus were collected, and the corresponding ground truth values ​​of individual behavioral trajectories and group interaction features were labeled to construct a sample dataset.

[0045] A dynamic scene semantic segmentation model is constructed based on a deep learning network with an encoder-decoder architecture. The encoder is used to extract multi-scale features from temporal visual data, and the decoder is used to achieve pixel-level semantic parsing.

[0046] Using the historical time-series visual data as input, and with the corresponding individual behavior trajectory ground truth and group interaction feature ground truth as supervision targets, the dynamic scene semantic segmentation model is iteratively trained until convergence.

[0047] In this embodiment, firstly, historical time-series visual data from multiple monitoring areas within the campus are collected, covering data from different time periods, weather conditions, and activity scenarios. For example, data from daytime teaching activities, evening study and rest periods, different weather conditions such as sunny, rainy, and snowy days, and special event scenarios such as sports meets and cultural performances. Simultaneously, a sample dataset is constructed by annotating the ground truth values ​​of individual behavioral trajectories and group interaction features.

[0048] For example, the true value of an individual's behavioral trajectory is obtained by parsing the movement speed and direction of motion from the video. The movement speed value is the accurate speed obtained by a high-precision positioning algorithm, and the unit is meters per second. The movement direction value is the precise movement angle calculated from the trajectory coordinates.

[0049] The ground truth of group interaction features is extracted from the video to obtain the inter-individual spacing and cluster density. The inter-individual spacing matrix records the actual distance between any two individuals in the group, which has been precisely measured; the cluster density value represents the number of individuals per unit area within a specific region, calculated based on the accurate location of the individuals.

[0050] Secondly, a dynamic scene semantic segmentation model is constructed based on an encoder-decoder architecture. The encoder part adopts a multi-layer convolutional neural network structure, extracting multi-scale features from temporal visual data through convolutional kernels of different scales. Multi-scale features not only contain local details but also encompass global contextual information, which helps to identify behavioral features in different scenes. The decoder part maps the features extracted by the encoder back to the pixel level of the original image through operations such as deconvolution and upsampling, achieving pixel-level semantic parsing.

[0051] Furthermore, during the iterative training of the dynamic scene semantic segmentation model, optimization algorithms such as stochastic gradient descent are used to update the model's parameters. Simultaneously, to prevent overfitting, regularization methods such as L1 and L2 regularization are employed. During training, historical temporal visual data is divided into training, validation, and test sets. The training set is used for updating model parameters, the validation set is used to adjust hyperparameters such as learning rate and batch size, and the test set is used to evaluate the model's final performance.

[0052] For example, the specific steps for building and training a dynamic scene semantic segmentation model based on a deep learning network are as follows:

[0053] First, data preparation: the input nodes of the dynamic scene semantic segmentation model are historical time-series visual data, collected from multiple monitoring areas within the campus.

[0054] Secondly, in model construction, the number of nodes in the input layer is equal to the dimension of the input features. For example, if there are three features such as time period, weather conditions, and activity scene, the input layer contains three nodes. Set 1-3 hidden layers, and adjust the number of nodes in each layer through experiments, such as 64 or 32. The activation function is ReLU. The output layer generally does not use an activation function. For example, if the output time is 2 nodes, directly output continuous values.

[0055] Next, model training uses pixel-level semantic parsing as output. The training framework is constructed using the ground truth values ​​of individual behavioral trajectories and group interaction features as supervision, employing the Adam optimizer and mean squared error loss function. The batch size is set to 32, the total training epochs to 50, and an early stopping mechanism (patience=5) is introduced. If the validation set loss does not decrease for five consecutive epochs, the training process is automatically terminated, resulting in a trained dynamic scene semantic segmentation model. This effectively avoids overfitting while ensuring the model reaches convergence.

[0056] Furthermore, after the model training converges, the model is further optimized and adjusted. For example, multiple trained models can be combined using model fusion to improve the model's generalization ability and accuracy. Simultaneously, the model is monitored and updated in real time, and its parameters are adjusted promptly as the campus environment and personnel behavior patterns change to ensure the model can accurately identify abnormal behavior patterns.

[0057] By constructing and optimizing the dynamic scene semantic segmentation model in the above manner, we can provide more accurate and reliable basic data support for subsequent campus security situation analysis, and further improve the level of intelligence in campus security management.

[0058] S20: Based on the behavioral semantic graph and the topology of the monitoring area, construct a dynamic security influence factor graph network;

[0059] In this embodiment of the application, a dynamic security influence factor graph network is constructed based on the behavioral semantic graph and the topology of the monitoring area. The network uses the monitoring area as the node and abnormal behavioral events and their potential impact on the surrounding area as the edge.

[0060] First, the topology between nodes is determined based on the physical location and connectivity of the monitored area. Then, by combining the abnormal behavior event information from the behavioral semantic graph, each edge is assigned a corresponding weight.

[0061] In constructing a dynamic security impact factor graph network, the influence of time is considered. The scope and extent of the impact of anomalous behavior may change over time.

[0062] Specifically, step S20 in the method includes:

[0063] Based on the behavioral semantic graph and the corresponding monitoring area topology, the ground values ​​of the security impact propagation weights between different monitoring areas are labeled, and a graph network training dataset is constructed.

[0064] Based on the graph attention network architecture, a dynamic security influence factor graph network is constructed, wherein the graph attention network uses the monitoring area as the node and the security influence relationship between the areas as the edge, and dynamically learns the security influence propagation weight through the attention mechanism;

[0065] Using the behavioral semantic graph and the topology of the monitoring area as input, and the corresponding ground truth values ​​of the security impact propagation weights as the supervision target, the dynamic security impact factor graph network is iteratively trained until convergence.

[0066] In this embodiment, information is first extracted from the behavioral semantic graph and the topology of the monitoring area to label the ground truth values ​​of the security impact propagation weights between different monitoring areas. During the labeling process, factors such as the type, severity, and location of the abnormal behavior, as well as the spatial distance and personnel flow between monitoring areas, are considered.

[0067] For example, in adjacent classrooms within a teaching building, if abnormal behavior such as fighting occurs, the propagation weight of the safety impact may be relatively high; while for areas that are far apart and have less personnel flow, the propagation weight of the safety impact is relatively low. Through analysis of extensive historical data and real-world cases, weighting rules were determined, and a graph network training dataset was constructed.

[0068] Secondly, a dynamic security impact factor graph network is constructed based on a graph attention network architecture. The graph attention network uses monitored areas as nodes, with each node containing relevant feature information about that area, such as whether abnormal behavior has occurred and the level of abnormal behavior. Edges represent the security impact relationships between areas, with the edge weights indicating the degree of security impact between different monitored areas. Through an attention mechanism, the graph attention network can adaptively learn the importance of different nodes and edges, thereby dynamically adjusting the security impact propagation weights.

[0069] Then, using behavioral semantic graphs and the topology of monitored areas as inputs, and the corresponding ground truth weights for security impact propagation as the supervision target, the dynamic security impact factor graph network is iteratively trained. During training, optimization algorithms such as stochastic gradient descent are used to update the network parameters, so that the network output gradually approximates the true security impact propagation weights. Simultaneously, to prevent overfitting, regularization methods such as L1 and L2 regularization are employed. The training dataset is divided into training, validation, and test sets. The training set is used for updating model parameters, the validation set is used to adjust model hyperparameters such as learning rate and batch size, and the test set is used to evaluate the final performance of the model.

[0070] During training, the model's performance metrics, such as mean squared error and accuracy, are monitored in real time. When the model's performance metrics reach certain requirements or the validation set loss does not decrease for several consecutive rounds, the model is considered to have converged, and the training process is terminated. This results in a dynamically trained security impact factor graph network. This network can predict the propagation of security impacts between different monitoring areas based on real-time behavioral semantic graphs and the topology of the monitoring area, providing crucial support for intelligent perception of campus security situations.

[0071] Furthermore, the trained dynamic security impact factor graph network is evaluated and optimized. By analyzing past campus security incident records, it is observed how adjacent areas are affected when an anomaly occurs in a certain area. Weights are assigned to the impact relationships between every two adjacent areas based on the frequency and severity of consecutive events, ultimately forming a graph network training dataset containing all regional relationships and impact weights. Simultaneously, as the campus environment and personnel behavior patterns continuously change, the model is regularly updated and maintained to ensure it can always adapt to new security situations and provide reliable decision-making support for campus security management.

[0072] S30: Based on the dynamic security impact factor graph network, perform security community division and key node analysis, and identify the core impact nodes and corresponding security status indicators in multiple security communities;

[0073] In this embodiment, a graph partitioning algorithm is used to divide security communities based on a dynamic security influence factor graph network. A security community refers to a set of monitoring areas with similar security characteristics and mutual influence relationships in the campus security situation. By analyzing the edge weights between nodes, nodes that are closely connected and have a significant mutual influence are grouped into the same security community. For example, adjacent classrooms within a teaching building area with high security influence propagation weights are grouped into one security community.

[0074] Secondly, after dividing the security communities, a critical node analysis is performed on each security community. Critical nodes refer to the monitoring areas within a security community that have a significant impact on the security posture. Core influencing nodes are identified by calculating indicators such as degree centrality and betweenness centrality. Degree centrality measures the number of connections a node has with other nodes, while betweenness centrality reflects the importance of a node as a bridge in the network.

[0075] For each core impact node, the corresponding security situation indicators are identified, including but not limited to the frequency, severity, and scope of security impact of abnormal behavior events. For example, in the monitoring area where a core impact node is located, if the frequency of abnormal behavior events is high and the severity is high, then the security situation indicators corresponding to that node show a relatively severe state.

[0076] By identifying key impact nodes and security status indicators across multiple security communities, the security situation in different areas of the campus can be understood. For areas with weaker security at key impact nodes, targeted security measures can be prioritized, such as increasing patrol frequency and strengthening the deployment of monitoring equipment. Simultaneously, continuous monitoring of changes in security status indicators at key impact nodes allows for timely adjustments to security management strategies, achieving dynamic management and precise control of campus security, and further improving the efficiency and effectiveness of campus security management.

[0077] Specifically, step S30 in the method includes:

[0078] Based on the node connection relationships and security impact propagation weights in the dynamic security impact factor graph network, a modularity optimization algorithm is used to divide security communities and identify security communities with tight internal connections.

[0079] Calculate the degree centrality and betweenness centrality of nodes within each security community, and assess the criticality of nodes in the process of security impact propagation;

[0080] Based on the ranking of the criticality of the nodes, the top-ranked critical nodes with a predetermined proportion are selected as core influencing nodes, and the corresponding security situation indicators are extracted. The security situation indicators include risk propagation capability, scope of influence, and emergency response priority.

[0081] In this embodiment, firstly, based on the node connection relationships and security impact propagation weights in the dynamic security impact factor graph network, a modularity optimization algorithm is used to perform security community partitioning. The modularity optimization algorithm can measure the quality of graph partitioning. By continuously adjusting the partitioning method, it makes the connections between nodes within each security community tight, while the connections between nodes in different security communities are relatively sparse, thereby identifying security communities with tight internal connections.

[0082] In practice, based on the weight of security impact propagation between monitored areas, areas with significant mutual influence and close connections are grouped into the same security community. For example, a library and its surrounding study rooms are grouped into a security community because the flow of people in the area is highly interconnected, and the propagation of security impact is relatively direct.

[0083] Next, the degree centrality and betweenness centrality of nodes within each security community are calculated. Degree centrality is calculated by counting the number of connections each node has with other nodes. The more connections a node has, the more active it is in the security community, and the greater its potential impact on the security posture. Betweenness centrality measures the importance of a node as a bridge in the network, determined by calculating the number of shortest paths passing through that node. A high betweenness centrality of a node means it plays a crucial transit role in the propagation of security influences; if this node encounters a problem, it could significantly impact the security posture of the entire security community.

[0084] Then, based on the ranking of node criticality, the top-ranked nodes (by a predetermined percentage) are selected as core influencing nodes. This predetermined percentage can be adjusted according to the actual situation of the campus and security management needs; for example, the top 10% of nodes could be selected. For core influencing nodes, corresponding security situation indicators are extracted, including risk propagation capability, scope of impact, and emergency response priority.

[0085] Among them, risk propagation capability reflects the possibility and speed at which security issues caused by the node spread to the surrounding areas; the scope of impact clarifies the range of monitored areas that the security issues of the node may affect; and the emergency response priority is determined according to the importance of the node and the severity of the security issues, so as to be able to respond quickly when security issues occur.

[0086] Specifically, risk propagation capability is the sum of the security impact propagation weights of the nodes; the impact range is defined by calculating the number of nodes reachable from the node within a specific number of steps / area; emergency response priority is calculated by combining risk propagation capability, impact range, and the inherent risk level of the area. Emergency Response Priority = (Risk Propagation Capability × Weight A) + (Impact Range × Weight B) + (Inherent Risk Level of the Area × Weight C). The inherent risk level of the area is pre-set by a team of security experts by integrating campus security management data and regional functional attributes.

[0087] For example, the risk propagation capability of a campus library node is 50, which is the sum of the security impact propagation weights around this node. The impact range is within 3 steps, reaching 5 surrounding areas, corresponding to an area of ​​200 square meters, with an inherent risk level of 3 for each area. We set weight A to 0.4, weight B to 0.3, and weight C to 0.3. Therefore, the risk propagation capability of this library node is 50, and the calculated impact range is 5 areas or 200 square meters. The emergency response priority is (50 × 0.4) + (200 × 0.3) + (3 × 0.3) = 20 + 60 + 0.9 = 80.9.

[0088] By analyzing and monitoring key influencing nodes and their security status indicators, we can grasp the security situation on campus and provide strong support for campus security management.

[0089] This includes calculating the degree centrality and betweenness centrality of nodes within each security community, and assessing the criticality of nodes in the propagation of security impacts, including:

[0090] Based on the dynamic security influence factor graph network, the degree centrality of each node is calculated, wherein the degree centrality is the ratio of the number of connections between the current node and other nodes to the maximum possible number of connections.

[0091] By analyzing the shortest paths between all node pairs in the dynamic security influence factor graph network, the betweenness centrality of each node is calculated, where the betweenness centrality is the number of times the current node appears on the shortest path of other node pairs.

[0092] The degree centrality and betweenness centrality are standardized and then weighted and fused using preset weight coefficients to obtain the node criticality evaluation value.

[0093] In this embodiment, the degree centrality of each node is first calculated. In a dynamic security influence factor graph network, degree centrality reflects the activity level of a node and the tightness of its connections within the network. The degree centrality is the ratio obtained by dividing the number of connections of the current node to other nodes by the maximum possible number of connections.

[0094] For example, in a secure community with 10 nodes, if a node is connected to 5 other nodes, and the maximum possible number of connections is 9, then the degree centrality of that node is 5 / 9.

[0095] Next, the betweenness centrality of each node is calculated by analyzing the shortest paths between all node pairs in the dynamic security influence factor graph network. Betweenness centrality measures the importance of a node as a bridge in the network. It is obtained by counting the number of times the current node appears on the shortest paths of other node pairs.

[0096] For example, if in a certain analysis it is found that node A appears on the shortest path of 10 pairs of nodes, then the betweenness centrality of node A is 10.

[0097] Taking into account the impact of degree centrality and betweenness centrality on the criticality of nodes, degree centrality and betweenness centrality are standardized. Standardization eliminates the dimensional differences between different indicators. Then, the standardized degree centrality and betweenness centrality are weighted and merged using preset weighting coefficients.

[0098] For example, if the weight coefficient of degree centrality is preset to 0.6 and the weight coefficient of betweenness centrality is 0.4, and the standardized degree centrality of a node is 0.8 and the betweenness centrality is 0.7, then the criticality evaluation value of the node is 0.8×0.6+0.7×0.4=0.76.

[0099] By using the above methods, the criticality of each node in the process of security impact propagation can be assessed, providing a basis for subsequent selection of core impact nodes and security situation analysis.

[0100] S40: Integrate the security situation indicators of the core impact nodes and the preset campus security strategy library, and make adaptive scheduling decisions for security resources through a multi-objective optimization algorithm to output a real-time security early warning scheme.

[0101] In this embodiment, the security posture indicators of core impact nodes are integrated with a pre-defined campus security strategy library. The campus security strategy library contains various response strategies for different security postures, such as increasing security personnel patrols, strengthening monitoring equipment, and conducting security awareness and education activities. During the integration process, based on security posture indicators such as the risk propagation capability, impact scope, and emergency response priority of the core impact nodes, matching security strategies are selected from the campus security strategy library.

[0102] Next, a multi-objective optimization algorithm is used for adaptive scheduling decisions of security resources, considering multiple objectives, such as minimizing the cost of security resource usage and improving the efficiency of security management while ensuring campus safety. Security resources are allocated and scheduled based on the security situation indicators of core impact nodes and the selected security strategies. For example, for areas with core impact nodes that have high emergency response priority, strong risk propagation capabilities, and a wide impact range, more security personnel and emergency supplies are prioritized, and the monitoring frequency of surveillance equipment in that area is increased.

[0103] During the decision-making process, the real-time status and availability of campus security resources are considered. If security personnel in certain areas are performing other tasks, or if certain monitoring equipment malfunctions, corresponding adjustments need to be made during dispatch decisions. Simultaneously, the dispatch plan for security resources is further optimized by considering the actual situation on campus, such as personnel flow at different times and the functional attributes of different areas.

[0104] Through iterative calculations using a multi-objective optimization algorithm, the scheduling scheme for security resources is continuously adjusted until the optimal scheduling strategy is found. The final output is a real-time security early warning scheme, which specifies concrete security measures and resource scheduling arrangements for areas with different core impact nodes. For example, the scheme might include adding two security personnel to patrol a teaching building area, or increasing the monitoring time of surveillance equipment in a library area.

[0105] Specifically, step S40 in the method includes:

[0106] Integrate the security situation indicators of the core impact nodes to construct a set of security situation indicators;

[0107] Based on the campus security policy library, the emergency response rules and resource constraints corresponding to the current security situation are matched to determine the types and configuration parameters of available security resources.

[0108] Based on the aforementioned set of security situation indicators, security resource types, and configuration parameters, a multi-objective optimization function is constructed.

[0109] Based on the multi-objective optimization function, a non-dominated sorting genetic algorithm is used to search for the Pareto optimal solution set and solve for the optimal safe resource scheduling scheme.

[0110] A real-time security early warning scheme is generated based on the optimal security resource scheduling scheme.

[0111] In this embodiment, the security situation indicators of core impact nodes are first integrated to construct a set of security situation indicators. Indicators such as the risk propagation capability, impact scope, and emergency response priority of each core impact node are summarized to form a set of indicators reflecting the campus security situation.

[0112] Next, based on the campus security policy database, the emergency response rules and resource constraints corresponding to the current security situation are matched to determine the types and configuration parameters of available security resources. The campus security policy database stores response rules and resource usage specifications for different security situations. By comparing the set of security situation indicators with the policy database, emergency response rules that match the current campus security situation are identified. Simultaneously, the types of available security resources, such as security personnel, monitoring equipment, and emergency supplies, are determined, and their configuration parameters are specified, such as patrol routes for security personnel and the monitoring range and frequency of monitoring equipment.

[0113] Then, based on the set of security situation indicators and the types and configuration parameters of security resources, a multi-objective optimization function is constructed. This function needs to consider multiple objectives, such as ensuring campus safety, reducing the cost of security resource usage, and improving security management efficiency. Through mathematical modeling, the objectives are transformed into functional expressions, enabling quantitative analysis and optimization of security resource scheduling.

[0114] Subsequently, based on the multi-objective optimization function, a non-dominated sorting genetic algorithm is used to search for the Pareto optimal solution set to find the optimal safe resource scheduling scheme. The non-dominated sorting genetic algorithm is an effective multi-objective optimization algorithm that, by simulating the biological evolution process, iteratively searches to find a set of non-dominated solutions, i.e., the Pareto optimal solution set. In this solution set, each solution cannot improve any objective without harming other objectives. By evaluating and comparing the solutions, the optimal safe resource scheduling scheme is finally determined.

[0115] Finally, a real-time security early warning plan is generated based on the optimal security resource scheduling scheme. This plan will specify the concrete security measures and security resource scheduling arrangements for the areas where different core impact nodes are located.

[0116] For example, for areas with high emergency response priority, the plan will detail how many additional security personnel will be added, how the monitoring strategies of surveillance equipment will be adjusted, and what emergency supplies will be allocated. Simultaneously, the real-time security early warning plan should also have dynamic adjustment capabilities, able to be updated and optimized in a timely manner according to real-time changes in the campus security situation, to ensure continuous and effective protection of campus security.

[0117] Furthermore, based on the aforementioned set of security situation indicators and the types and configuration parameters of security resources, a multi-objective optimization function is constructed, including:

[0118] The risk propagation capability, impact range, and emergency response priority are extracted from the set of security situation indicators and set as the first optimization objective, the second optimization objective, and the third optimization objective, respectively. The first optimization objective is to minimize the risk propagation capability, the second optimization objective is to maximize the impact range coverage, and the third optimization objective is to optimize the emergency response priority matching degree.

[0119] Based on the security resource type and configuration parameters, determine the total resource limit and response time threshold, and construct a set of constraints.

[0120] By combining the first optimization objective, the second optimization objective, the third optimization objective, and the set of constraints, a multi-objective optimization function containing multi-dimensional optimization indicators is constructed, wherein the weight coefficients of each optimization objective are dynamically adjusted according to the priority rules in the campus security policy library.

[0121] In this embodiment of the application, firstly, the risk propagation capability, impact range and emergency response priority are extracted from the set of security situation indicators, and set as the first optimization objective, the second optimization objective and the third optimization objective, respectively.

[0122] Furthermore, since reducing the likelihood and speed of the spread of security issues can effectively control the scale of the impact of security incidents and prevent security issues from spreading widely on campus, minimizing the risk propagation capability is taken as the first optimization objective; maximizing the coverage of the impact range is taken as the second optimization objective, ensuring that all potentially affected areas on campus can be effectively monitored and protected, reducing blind spots in security protection; and optimizing the matching degree of emergency response priorities is taken as the third optimization objective, ensuring that when a security issue occurs, a rapid and accurate response can be made according to the importance of the node and the severity of the security issue, thereby improving the efficiency of emergency response.

[0123] Secondly, based on the types and configuration parameters of security resources, resource quantity constraints and response time thresholds are determined, constructing a set of constraints. Resource quantity constraints refer to the limited availability of security resources on campus, such as the number of security personnel, monitoring equipment, and emergency supplies reserves; the total amount of resources cannot be exceeded during security resource allocation. The response time threshold specifies the maximum time required for security resources to arrive on-site and begin processing after a security incident occurs, ensuring timely response to security events.

[0124] Then, by integrating the first, second, and third optimization objectives and the set of constraints, a multi-objective optimization function with multi-dimensional optimization indicators is constructed. This function needs to balance the relationships between the various optimization objectives to achieve overall optimal scheduling. The weight coefficients of each optimization objective are dynamically adjusted according to the priority rules in the campus security policy library.

[0125] For example, during large-scale events held on campus, the large flow of people increases the safety risks, so the weighting coefficient of risk transmission capability can be increased to strengthen the control of the spread of safety issues; while in daily campus management, more attention is paid to the coverage of the scope of impact and the matching degree of emergency response priority, and the weighting coefficients of these two optimization objectives are adjusted accordingly.

[0126] By dynamically adjusting the weight coefficients, the multi-objective optimization function can better adapt to different campus security situations, thereby formulating a more reasonable and effective security resource scheduling plan.

[0127] In summary, compared to existing technologies, this application, by constructing a dynamic security impact factor graph network, conducts in-depth analysis and monitoring of core impact nodes and their security status indicators, enabling accurate understanding of the campus security situation. After assessing the criticality of nodes in the security impact propagation process, adaptive scheduling decisions for security resources are made based on the security status indicators of core impact nodes, outputting real-time security early warning schemes. By comprehensively considering multiple factors related to campus security, from the risk propagation capabilities and impact scope of nodes to emergency response priorities and the rational allocation of security resources, a complete campus security management system is formed.

[0128] In summary, the embodiments of this application have at least the following technical effects:

[0129] This application provides a computer vision-assisted intelligent perception method for smart campus security situation. First, it utilizes a high-definition camera array to synchronously collect multi-view temporal visual data. Combined with a dynamic scene semantic segmentation model, it achieves pixel-level semantic parsing, extracting individual behavioral trajectories and group interaction features. This effectively identifies abnormal behavior patterns and constructs a behavioral semantic graph, overcoming the limitations of traditional monitoring systems. Second, it constructs a dynamic security influence factor graph network based on a graph attention network architecture. Through an attention mechanism, it dynamically learns the security influence propagation weights between regions. Combined with a modularity optimization algorithm, it achieves security community division and key node analysis, identifying core influence nodes and corresponding security situation indicators, thus improving risk propagation prediction capabilities. Finally, by integrating the security situation indicators of core influence nodes with a preset campus security strategy library, it employs a multi-objective optimization algorithm for adaptive scheduling decisions of security resources. This enables real-time generation of dynamic early warning schemes that include risk propagation capabilities, impact range, and emergency response priorities. This solves the problems of slow response speed and insufficient data analysis capabilities in traditional management models, providing a full-process intelligent solution for smart campus security management. Through the above technical solutions, this application can improve the intelligence level of campus security management, increase the accuracy of abnormal behavior identification, shorten security incident response time, and effectively reduce the incidence of campus security accidents.

[0130] Example 2, as Figure 2 As shown, based on the same inventive concept as the computer vision-assisted intelligent perception method for smart campus security situation provided in Embodiment 1, this application also provides a computer vision-assisted intelligent perception system for smart campus security situation, including:

[0131] Information acquisition module 11 is used to collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on dynamic scene semantic segmentation model, identify abnormal behavior patterns and generate behavior semantic map;

[0132] Network construction module 12 is used to construct a dynamic security influence factor graph network based on the behavioral semantic graph and the monitoring area topology.

[0133] Node analysis module 13 is used to divide security communities and analyze key nodes based on the dynamic security impact factor graph network, and identify the core impact nodes and corresponding security status indicators in multiple security communities.

[0134] The solution output module 14 is used to integrate the security status indicators of the core impact nodes and the preset campus security strategy library, and to make adaptive scheduling decisions for security resources through a multi-objective optimization algorithm, and output a real-time security early warning solution.

[0135] In one embodiment, the information acquisition module 11 is specifically used for:

[0136] Multi-view temporal visual data is collected simultaneously using a high-definition camera array deployed on campus.

[0137] Based on a dynamic scene semantic segmentation model, pixel-level semantic parsing is performed on the temporal visual data to extract individual behavior trajectories and group interaction features. The individual behavior trajectories include movement speed and direction of movement, and the group interaction features include individual spacing and clustering density.

[0138] Based on a pre-defined abnormal behavior knowledge base, the individual behavior trajectories and group interaction features are matched to mark abnormal behavior events and construct a behavioral semantic graph.

[0139] Furthermore, in one embodiment of the application, the construction steps of the dynamic scene semantic segmentation model include:

[0140] Historical time-series visual data from multiple monitoring areas on campus were collected, and the corresponding ground truth values ​​of individual behavioral trajectories and group interaction features were labeled to construct a sample dataset.

[0141] A dynamic scene semantic segmentation model is constructed based on a deep learning network with an encoder-decoder architecture. The encoder is used to extract multi-scale features from temporal visual data, and the decoder is used to achieve pixel-level semantic parsing.

[0142] Using the historical time-series visual data as input, and with the corresponding individual behavior trajectory ground truth and group interaction feature ground truth as supervision targets, the dynamic scene semantic segmentation model is iteratively trained until convergence.

[0143] In one embodiment, network building module 12 is specifically used for:

[0144] Based on the behavioral semantic graph and the corresponding monitoring area topology, the ground values ​​of the security impact propagation weights between different monitoring areas are labeled, and a graph network training dataset is constructed.

[0145] Based on the graph attention network architecture, a dynamic security influence factor graph network is constructed, wherein the graph attention network uses the monitoring area as the node and the security influence relationship between the areas as the edge, and dynamically learns the security influence propagation weight through the attention mechanism;

[0146] Using the behavioral semantic graph and the topology of the monitoring area as input, and the corresponding ground truth values ​​of the security impact propagation weights as the supervision target, the dynamic security impact factor graph network is iteratively trained until convergence.

[0147] In one embodiment, the node analysis module 13 is specifically used for:

[0148] Based on the node connection relationships and security impact propagation weights in the dynamic security impact factor graph network, a modularity optimization algorithm is used to divide security communities and identify security communities with tight internal connections.

[0149] Calculate the degree centrality and betweenness centrality of nodes within each security community, and assess the criticality of nodes in the process of security impact propagation;

[0150] Based on the ranking of the criticality of the nodes, the top-ranked critical nodes with a predetermined proportion are selected as core influencing nodes, and the corresponding security situation indicators are extracted. The security situation indicators include risk propagation capability, scope of influence, and emergency response priority.

[0151] Furthermore, in one embodiment of the application, the degree centrality and betweenness centrality of nodes within each security community are calculated, and the node criticality in the security impact propagation process is evaluated, including:

[0152] Based on the dynamic security influence factor graph network, the degree centrality of each node is calculated, wherein the degree centrality is the ratio of the number of connections between the current node and other nodes to the maximum possible number of connections.

[0153] By analyzing the shortest paths between all node pairs in the dynamic security influence factor graph network, the betweenness centrality of each node is calculated, where the betweenness centrality is the number of times the current node appears on the shortest path of other node pairs.

[0154] The degree centrality and betweenness centrality are standardized and then weighted and fused using preset weight coefficients to obtain the node criticality evaluation value.

[0155] In one embodiment, the solution output module 14 is specifically used for:

[0156] Integrate the security situation indicators of the core impact nodes to construct a set of security situation indicators;

[0157] Based on the campus security policy library, the emergency response rules and resource constraints corresponding to the current security situation are matched to determine the types and configuration parameters of available security resources.

[0158] Based on the aforementioned set of security situation indicators, security resource types, and configuration parameters, a multi-objective optimization function is constructed.

[0159] Based on the multi-objective optimization function, a non-dominated sorting genetic algorithm is used to search for the Pareto optimal solution set and solve for the optimal safe resource scheduling scheme.

[0160] A real-time security early warning scheme is generated based on the optimal security resource scheduling scheme.

[0161] Furthermore, in one embodiment, a multi-objective optimization function is constructed based on the set of security situation indicators and the types and configuration parameters of security resources, including:

[0162] The risk propagation capability, impact range, and emergency response priority are extracted from the set of security situation indicators and set as the first optimization objective, the second optimization objective, and the third optimization objective, respectively. The first optimization objective is to minimize the risk propagation capability, the second optimization objective is to maximize the impact range coverage, and the third optimization objective is to optimize the emergency response priority matching degree.

[0163] Based on the security resource type and configuration parameters, determine the total resource limit and response time threshold, and construct a set of constraints.

[0164] By combining the first optimization objective, the second optimization objective, the third optimization objective, and the set of constraints, a multi-objective optimization function containing multi-dimensional optimization indicators is constructed, wherein the weight coefficients of each optimization objective are dynamically adjusted according to the priority rules in the campus security policy library.

[0165] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0166] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0167] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications within the scope of this application.

[0168] Variations, combinations, or equivalents. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope.

[0169] The scope of this application. Thus, if these modifications and variations of this application fall within the scope of this application and its equivalents,

[0170] Therefore, this application is intended to include these modifications and variations.

Claims

1. A computer vision-assisted intelligent perception method for security situation in smart campuses, characterized in that, The method includes: Collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on a dynamic scene semantic segmentation model, identify abnormal behavior patterns, and generate a behavior semantic map. Based on the aforementioned behavioral semantic graph and the topology of the monitoring area, a dynamic security influence factor graph network is constructed, including: Based on the behavioral semantic graph and the corresponding monitoring area topology, the ground values ​​of the security impact propagation weights between different monitoring areas are labeled, and a graph network training dataset is constructed. Based on the graph attention network architecture, a dynamic security influence factor graph network is constructed, wherein the graph attention network uses the monitoring area as the node and the security influence relationship between the areas as the edge, and dynamically learns the security influence propagation weight through the attention mechanism; Using the behavioral semantic graph and the topology of the monitoring area as input, and the corresponding security impact propagation weights as the supervision target, the dynamic security impact factor graph network is iteratively trained until convergence. Based on the dynamic security impact factor graph network, security communities are divided and key nodes are analyzed to identify core impact nodes and corresponding security posture indicators in multiple security communities, including: Based on the node connection relationships and security impact propagation weights in the dynamic security impact factor graph network, a modularity optimization algorithm is used to divide security communities and identify security communities with tight internal connections. Calculate the degree centrality and betweenness centrality of nodes within each security community, and assess the criticality of nodes in the process of security impact propagation; Based on the ranking of the criticality of the nodes, the top-ranked critical nodes with a predetermined proportion are selected as core influencing nodes, and the corresponding security situation indicators are extracted. The security situation indicators include risk propagation capability, scope of influence, and emergency response priority. By integrating the security status indicators of the core impact nodes and the preset campus security strategy library, a multi-objective optimization algorithm is used to make adaptive scheduling decisions for security resources and output a real-time security early warning scheme.

2. The computer vision-assisted intelligent perception method for smart campus security situation as described in claim 1, characterized in that, Temporal visual data from multiple monitoring areas on campus were collected. Multi-scale behavior analysis was performed based on a dynamic scene semantic segmentation model to identify abnormal behavior patterns and generate a behavioral semantic map, including: Multi-view temporal visual data is collected simultaneously using a high-definition camera array deployed on campus. Based on a dynamic scene semantic segmentation model, pixel-level semantic parsing is performed on the temporal visual data to extract individual behavior trajectories and group interaction features. The individual behavior trajectories include movement speed and direction of movement, and the group interaction features include individual spacing and clustering density. Based on a pre-defined abnormal behavior knowledge base, the individual behavior trajectories and group interaction features are matched to mark abnormal behavior events and construct a behavioral semantic graph.

3. The computer vision-assisted intelligent perception method for smart campus security situation as described in claim 2, characterized in that, The construction steps of the dynamic scene semantic segmentation model include: Historical time-series visual data from multiple monitoring areas on campus were collected, and the corresponding ground truth values ​​of individual behavioral trajectories and group interaction features were labeled to construct a sample dataset. A dynamic scene semantic segmentation model is constructed based on a deep learning network with an encoder-decoder architecture. The encoder is used to extract multi-scale features from temporal visual data, and the decoder is used to achieve pixel-level semantic parsing. Using the historical time-series visual data as input, and with the corresponding individual behavior trajectory ground truth and group interaction feature ground truth as supervision targets, the dynamic scene semantic segmentation model is iteratively trained until convergence.

4. The computer vision-assisted intelligent perception method for smart campus security situation as described in claim 1, characterized in that, Calculate the degree centrality and betweenness centrality of nodes within each security community, and assess the node criticality in the propagation of security impacts, including: Based on the dynamic security influence factor graph network, the degree centrality of each node is calculated, wherein the degree centrality is the ratio of the number of connections between the current node and other nodes to the maximum possible number of connections. By analyzing the shortest paths between all node pairs in the dynamic security influence factor graph network, the betweenness centrality of each node is calculated, where the betweenness centrality is the number of times the current node appears on the shortest path of other node pairs. The degree centrality and betweenness centrality are standardized and then weighted and fused using preset weight coefficients to obtain the node criticality evaluation value.

5. The computer vision-assisted intelligent perception method for smart campus security situation as described in claim 1, characterized in that, By integrating the security situation indicators of the core impact nodes and a pre-set campus security policy library, a multi-objective optimization algorithm is used to make adaptive scheduling decisions for security resources, and a real-time security early warning scheme is output, including: Integrate the security situation indicators of the core impact nodes to construct a set of security situation indicators; Based on the campus security policy library, the emergency response rules and resource constraints corresponding to the current security situation are matched to determine the types and configuration parameters of available security resources. Based on the aforementioned set of security situation indicators, security resource types, and configuration parameters, a multi-objective optimization function is constructed. Based on the multi-objective optimization function, a non-dominated sorting genetic algorithm is used to search for the Pareto optimal solution set and solve for the optimal safe resource scheduling scheme. A real-time security early warning scheme is generated based on the optimal security resource scheduling scheme.

6. The computer vision-assisted intelligent perception method for smart campus security situation as described in claim 5, characterized in that, Based on the aforementioned set of security situation indicators and the types and configuration parameters of security resources, a multi-objective optimization function is constructed, including: The risk propagation capability, impact range, and emergency response priority are extracted from the set of security situation indicators and set as the first optimization objective, the second optimization objective, and the third optimization objective, respectively. The first optimization objective is to minimize the risk propagation capability, the second optimization objective is to maximize the impact range coverage, and the third optimization objective is to optimize the emergency response priority matching degree. Based on the security resource type and configuration parameters, determine the total resource limit and response time threshold, and construct a set of constraints. By combining the first optimization objective, the second optimization objective, the third optimization objective, and the set of constraints, a multi-objective optimization function containing multi-dimensional optimization indicators is constructed, wherein the weight coefficients of each optimization objective are dynamically adjusted according to the priority rules in the campus security policy library.

7. A computer vision-assisted intelligent perception system for smart campus security situation, characterized in that, The method for performing the computer vision-assisted intelligent perception method for smart campus security situation according to any one of claims 1-6 includes: The information acquisition module is used to collect time-series visual data from multiple monitoring areas on campus, perform multi-scale behavior analysis based on a dynamic scene semantic segmentation model, identify abnormal behavior patterns, and generate a behavior semantic map. The network construction module is used to construct a dynamic security influence factor graph network based on the behavioral semantic graph and the monitoring area topology, including: Based on the behavioral semantic graph and the corresponding monitoring area topology, the ground values ​​of the security impact propagation weights between different monitoring areas are labeled, and a graph network training dataset is constructed. Based on the graph attention network architecture, a dynamic security influence factor graph network is constructed, wherein the graph attention network uses the monitoring area as the node and the security influence relationship between the areas as the edge, and dynamically learns the security influence propagation weight through the attention mechanism; Using the behavioral semantic graph and the topology of the monitoring area as input, and the corresponding security impact propagation weights as the supervision target, the dynamic security impact factor graph network is iteratively trained until convergence. The node analysis module is used to perform security community segmentation and key node analysis based on the dynamic security impact factor graph network, identifying core impact nodes and corresponding security posture indicators in multiple security communities, including: Based on the node connection relationships and security impact propagation weights in the dynamic security impact factor graph network, a modularity optimization algorithm is used to divide security communities and identify security communities with tight internal connections. Calculate the degree centrality and betweenness centrality of nodes within each security community, and assess the criticality of nodes in the process of security impact propagation; Based on the ranking of the criticality of the nodes, the top-ranked critical nodes with a predetermined proportion are selected as core influencing nodes, and the corresponding security situation indicators are extracted. The security situation indicators include risk propagation capability, scope of influence, and emergency response priority. The solution output module is used to integrate the security status indicators of the core impact nodes and the preset campus security strategy library, and to make adaptive scheduling decisions for security resources through a multi-objective optimization algorithm, and output a real-time security early warning solution.

Images