Data analysis and behavior prediction management method based on campus security network

Abstract

The invention provides a data analysis and behavior prediction management method based on a campus security network, and relates to the technical field of network security, and the method comprises the steps: carrying out the analysis of obtained campus scene data, so as to recognize and obtain a special event; determining a target object associated with the special event according to the special event; combining the target object with the campus scene data, and performing analysis through a preset behavior prediction model to identify and obtain a behavior prediction result of the target object; and comparing a preset threshold value with the behavior prediction result, and performing abnormal management analysis on the behavior prediction result exceeding the preset threshold value to obtain an abnormal behavior management analysis result.

Description

Technical Field

[0001] The present application belongs to the field of network security technology, and specifically relates to a data analysis and behavior prediction management method based on a campus security network. Background Art

[0002] With the continuous advancement of educational informatization and the construction of smart campuses, schools have gradually introduced a large number of intelligent sensing devices in daily management, such as cameras, access control systems, consumer terminals, and sensors, to achieve comprehensive monitoring and data collection of the campus environment and personnel behavior. This scene data provides a foundation for improving campus management, but many challenges still exist in practical application.

[0003] Currently, campus security management relies primarily on manual inspections, post-event alerts, or simple rule-based anomaly detection mechanisms. These methods struggle to detect potentially risky behaviors and lack the ability to dynamically predict individual behavior patterns. Furthermore, faced with massive amounts of heterogeneous scene data, efficiently identifying noteworthy incidents and identifying relevant targets presents a pressing technical challenge.

[0004] Traditional abnormal behavior identification methods are usually limited to single-dimensional data analysis. For example, students' attendance is judged solely through attendance records, or their learning status is assessed solely based on changes in grades. They lack the ability to integrate multi-source information and conduct intelligent modeling, resulting in low recognition accuracy, delayed response, and high false alarm and missed alarm rates. Summary of the Invention

[0005] This application provides a data analysis and behavior prediction management method based on a campus security network to solve one of the above technical problems.

[0006] The technical solutions adopted in this application are:

[0007] The present application embodiment provides a data analysis and behavior prediction management method based on a campus security network, comprising:

[0008] Analyze the acquired campus scene data to identify special events;

[0009] Determine target objects associated with special events based on special events;

[0010] The target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the target object's behavior prediction results;

[0011] The behavior prediction results are compared with the preset threshold, and the behavior prediction results that exceed the preset threshold are analyzed through abnormal management to obtain abnormal behavior management analysis results.

[0012] According to one embodiment of the present application, the acquired campus scene data is analyzed to identify special events, specifically:

[0013] Through the cameras, sensors and access control system equipment installed in the smart campus, scene data of multiple areas are collected in real time. The scene data includes image data, sensor data and positioning data, and the scene data is denoised, format converted and cleaned;

[0014] Analyze video data through image recognition algorithms to identify people's activity trajectories and gathering behavior patterns;

[0015] Perform time series analysis on sensor data to identify abnormal data points;

[0016] Use clustering algorithms to analyze abnormal clustering patterns within the region;

[0017] Based on the analysis results, the identified events are classified as special events.

[0018] According to an embodiment of the present application, determining the target object associated with the special event according to the special event is specifically:

[0019] Filter data related to special events from campus scene data;

[0020] If the relevant data is image data, identity information is extracted through a face recognition algorithm and compared with the campus database;

[0021] If the relevant data is access records or check-in information, identity information is extracted from it and matched against the campus database to determine the target object.

[0022] According to one embodiment of the present application, the preset behavior prediction model is specifically:

[0023] Load the basic behavior prediction model;

[0024] Using historical event record data similar to the special event to train the initial model to generate a first intermediate model;

[0025] Use the first intermediate model to predict the data of associated objects and associated events to generate pseudo labels;

[0026] Generate enhanced data versions by time shifting, time scaling, or adding noise;

[0027] Use the augmented data version to perform unsupervised training on the model to generate a second intermediate model;

[0028] The feature representation extracted by the second intermediate model is used as input, and supervised and unsupervised training are performed alternately until the model performance reaches the preset conditions.

[0029] According to one embodiment of the present application, the target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the behavior prediction result of the target object, specifically:

[0030] Integrate the target object's learning data and life data, and combine it with campus scene data;

[0031] The behavior of the target object is analyzed using a preset behavior prediction model, and potential abnormal behaviors that do not conform to the normal pattern are identified as the behavior prediction results of the target object.

[0032] According to one embodiment of the present application, the behavior prediction result is compared with the preset threshold, and the behavior prediction result exceeding the preset threshold is analyzed through abnormal management to obtain an abnormal behavior management analysis result, specifically:

[0033] The preset threshold is a risk score critical value derived from statistical analysis of historical behavior data, and is used to determine whether the target object's behavior deviates from the normal behavior pattern;

[0034] The behavior prediction result is a risk score for one or more dimensions, including academic performance risk score, attendance risk score, consumer behavior risk score, and entry and exit record risk score. The score of each dimension is compared with the corresponding preset threshold.

[0035] If the behavior prediction result does not exceed the preset threshold, the behavior is marked as normal behavior and the current monitoring status is maintained;

[0036] If the preset threshold is exceeded, the behavior will be marked as abnormal and the abnormal management process will be triggered;

[0037] Determine the abnormality type based on the behavior prediction result exceeding the threshold;

[0038] Extracting contextual information related to the anomaly type, including student profiles, historical behavioral data, and social relationship data;

[0039] According to the abnormality type and the corresponding background information, an abnormal behavior management analysis result is formed, and the abnormal behavior management analysis result includes personalized management and graded intervention measures.

[0040] According to one embodiment of the present application, the personalized management is specifically:

[0041] Provide different management methods according to different exception types;

[0042] Dynamically adjust management methods based on the severity of abnormal behavior and the student's background information.

[0043] According to one embodiment of the present application, the graded intervention measures are specifically:

[0044] Conduct time series analysis on identified abnormal behaviors to obtain development trends of abnormal behaviors;

[0045] Risk levels are divided according to the severity of abnormal behavior and its development trend;

[0046] Trigger different intervention measures based on different risk levels;

[0047] If risky behavior is not intervened in time and the risk escalates, the intervention measures will be automatically upgraded from automatic alarm to manual verification.

[0048] A second aspect of the present application provides a computer-readable storage medium having a program stored thereon, which implements the steps in the method when executed by a processor.

[0049] In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor implements the steps in the method described above when executing the program.

[0050] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows:

[0051] This application integrates multiple data types, including image data, sensor data, and location data, and uses techniques such as image recognition, time series analysis, and clustering algorithms to automatically identify special events such as unusual gatherings, unusual entry and exit, and sudden behavioral changes on campus. This improves the accuracy and real-time nature of event recognition, avoiding the inefficiency and subjectivity of traditional methods that rely on manual judgment.

[0052] After identifying a special event, this application can extract target information from relevant data and quickly identify the target object related to the event by combining technologies such as face recognition and identity feature matching. This achieves a precise mapping from macro events to specific individuals, providing a clear target basis for subsequent behavioral analysis.

[0053] This application utilizes a pre-defined behavioral prediction model, inputting multi-dimensional data such as the target subject's learning behavior and life trajectory into the model to output a prediction of their future behavioral trends. This improves the ability to foresee potential abnormal behaviors, facilitating early intervention measures and reducing the likelihood of risks.

[0054] This application compares behavioral predictions with pre-set thresholds to identify targets exceeding risk thresholds. It then conducts in-depth anomaly management analysis, including classification and identification, cause inference, and management strategy recommendations. This achieves a closed-loop management system from prediction to intervention, enhancing the intelligence of the campus safety management system and strengthening the ability to respond to emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0056] Figure 1 A flow chart of a data analysis and behavior prediction management method based on a campus security network provided in an embodiment of the present application;

[0057] Figure 2 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0058] Reference numerals:

[0059] 810 , processor; 820 , communication interface; 830 , memory; 840 , communication bus. DETAILED DESCRIPTION

[0060] In order to more clearly illustrate the overall concept of the present application, a detailed description is given below in an illustrative manner in conjunction with the accompanying drawings.

[0061] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application may also be implemented in other ways than those described herein, and therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below. It should be noted that the embodiments of the present application and the features of each embodiment may be combined with each other unless there is a conflict.

[0062] In this application, unless otherwise expressly specified and limited, a first feature "above" or "below" a second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0063] Example 1

[0064] like Figure 1 As shown, a data analysis and behavior prediction management method based on a campus safety network includes:

[0065] S100: Analyze the acquired campus scene data to identify special events.

[0066] As mentioned above, real-time data is collected from various sensors and devices deployed on campus (such as cameras, temperature and humidity sensors, and access control systems). This data can include image data (such as video streams), environmental data (such as temperature and humidity), and location data (such as tracking personnel locations through Wi-Fi or RFID). By integrating multiple data sources, the system can obtain a comprehensive and multi-dimensional view of campus activities.

[0067] Once enough scene data has been collected, the next step is to conduct in-depth analysis of it. This step aims to extract meaningful information, the so-called "special events", from a large amount of raw data. For example, abnormal gathering behavior, frequent entry and exit of the campus at abnormal times, sudden decline in grades, etc. may be regarded as events that require special attention. To achieve this, the system may use a series of advanced data analysis technologies, such as image recognition algorithms (for processing video data), time series analysis (for monitoring changing trends in sensor data), and clustering algorithms (for discovering patterns and anomalies).

[0068] For example, consider a particular classroom building where typically only a handful of students gather during breaks. However, over a specific period of time, the system detects a significant increase in the number of people in the area, exceeding the normal range. Using image recognition technology from video surveillance data and data analysis from human body sensors, the system identifies this as an unusual gathering and marks it as a special event.

[0069] Campus security policy requires all students to return to their dormitories within a specified timeframe during the night. If a student's access control records show them entering and leaving campus multiple times late at night, this will be considered a potential incident by the system. By analyzing access control system data and combining it with other relevant data (such as the student's course schedule), the system can further determine whether any anomalies exist.

[0070] A student's academic performance is a key indicator of their learning status. If the system detects a significant decline in a student's grades over the course of a semester, especially if a similar trend persists across multiple subjects, this could indicate a problem. By analyzing time series data of past exam scores, the system can identify such abnormal performance as a specific event.

[0071] It should be noted that in specific implementation scenarios, based on the above solution, the system can not only rely on existing physical sensors (such as cameras and access control systems), but also integrate more types of sensors and technologies, such as voice recognition systems (for monitoring the sound environment inside and outside the classroom) and air quality monitors (for assessing indoor air quality and health impacts). These additional data sources can help the system more comprehensively understand the various factors in the campus environment and their interrelationships.

[0072] In specific implementation scenarios, building on the above solutions, more complex machine learning models can be introduced to enhance data analysis. For example, deep learning networks can be used to improve image recognition accuracy; natural language processing techniques can be used to analyze information on social media to better understand students' emotional states and social interactions. Furthermore, reinforcement learning algorithms can dynamically adjust analysis strategies based on different campus environments, improving adaptability to emerging anomalies.

[0073] In specific implementation scenarios, this solution can be integrated with other campus management systems (such as the academic affairs management system and library management system) based on the above solution. This allows the system to obtain more contextual information and more accurately identify special events. For example, when connected to the academic affairs system, the system can understand students' course selection status and help better explain the causes of certain abnormal behaviors. This cross-platform data sharing and collaborative working method not only expands the scope of data application, but also enhances the functionality and practicality of the overall system.

[0074] S200: Determine a target object associated with a special event according to the special event.

[0075] As mentioned above, after identifying special events, the next step is to determine the specific individuals or objects (i.e., target objects) directly related to these events. This step is crucial for subsequent behavioral analysis and the formulation of management strategies. The system extracts target data related to special events from the scene data and uses different methods to identify target objects based on their type (such as image data or text data). For example, if the data is image data, facial recognition technology can be used to identify specific people; if it is text data, identity features (such as name, student number, etc.) can be extracted from it and matched with the information in the database.

[0076] For example, imagine a campus event where the system detects unusual gatherings in a certain area (a special event). To identify the specific individuals participating in the gathering (the target individuals), the system first obtains a live video stream of the area from a camera. It then uses a facial recognition algorithm to identify the faces in the video and compares the results with photos of students or faculty in the campus database to confirm their identities.

[0077] If the specific incident involves a student's unusual attendance (e.g., frequent absences), the system can extract relevant information about the student (e.g., student ID, name) from text data such as access control records and classroom sign-in records. The system then matches this information with the student's profile in the campus management system to ultimately determine the target's identity. Furthermore, the system can further analyze the student's other activity records (e.g., library visits, club participation) to fully understand their behavioral patterns.

[0078] In some complex situations, it may be necessary to combine multiple types of data to identify the target. For example, when the system detects a suspected theft on campus, in addition to using surveillance video for facial recognition, it also needs to refer to the location information of the stolen item (such as the RFID tag), the time and location of the owner's report, and other information to comprehensively determine and identify the suspect.

[0079] It should be noted that in specific implementation scenarios, the system can further integrate more types of sensor data based on the above solution, such as sound sensors (for capturing unusual sounds) and environmental sensors (for monitoring temperature and humidity changes), thereby providing richer contextual information and helping to more accurately identify the target object. For example, in an emergency, by combining video data and environmental sensor data, we can better understand the context of the incident and help accurately identify the relevant personnel.

[0080] In specific implementation scenarios, building on the above solutions, as technology advances, more advanced recognition technologies can be introduced to improve the accuracy of target identification. For example, deep learning models can be used to optimize facial recognition algorithms, or natural language processing techniques can be used to extract useful identity information from social media or other text sources. Furthermore, multimodal recognition methods can be explored, combining multiple types of data—images, sounds, text—for comprehensive analysis to obtain more comprehensive information about the target.

[0081] In specific implementation scenarios, the system can dynamically adjust its recognition strategy based on the above solutions, tailored to the needs of different time periods or specific situations. For example, during exams, the system can strengthen monitoring of people inside and outside the examination room and adopt stricter facial recognition standards; while during daily teaching activities, the requirements can be relaxed appropriately based on actual circumstances. This flexibility not only improves the system's adaptability but also ensures its effectiveness in various environments.

[0082] In specific implementation scenarios, this system can also be integrated with other campus management platforms (such as the academic affairs management system and health management system) based on the above solution. This allows the system to obtain more contextual information and more accurately identify the target objects. For example, when connected to the academic affairs system, the system can understand students' course selection and help better explain the reasons for certain abnormal behaviors. This cross-platform data sharing and collaborative working method not only expands the scope of data application, but also enhances the functionality and practicality of the overall system.

[0083] S300: The target object is combined with campus scene data and analyzed using a preset behavior prediction model to identify and obtain a behavior prediction result of the target object.

[0084] As mentioned above, after identifying the target person associated with a special event, the system integrates their identity information with relevant campus data (such as learning records, entry and exit records, and consumer behavior) and inputs this information into a trained behavior prediction model. This model, based on historical data and multi-dimensional features, can predict potential abnormal behavior trends of the target person in the future.

[0085] The process of building a behavior prediction model typically involves training with extensive historical behavioral data, extracting key behavioral features, establishing a library of individual behavioral patterns, and optimizing model performance through a combination of supervised and unsupervised learning. In practical application, the system uses the model to determine the target's behavioral trends based on the currently acquired data and outputs a risk score or probability of behavioral change, which serves as an important basis for subsequent anomaly management analysis.

[0086] For example, a student may have recently experienced decreased class attendance, late assignment submissions, and decreased library visits.

[0087] The system inputs these behavioral data into the behavioral prediction model. The model integrates factors such as the student's past performance, course difficulty, and exam cycle to predict that the student's grades may decline in the next stage.

[0088] The output result is the “learning performance risk score”. If it exceeds the set threshold, the early warning mechanism will be triggered.

[0089] A student has been late three times in the past month and has not appeared in the dormitory registration in the past week.

[0090] The behavior prediction model analyzes the student’s previous attendance records, course schedules, holiday information, etc., and determines that there is a high possibility that the student will skip classes in the future.

[0091] The output result is the "attendance abnormality probability", which is used to evaluate whether an intervention process needs to be initiated.

[0092] A student's campus card spending dropped sharply for three consecutive days, even lower than his daily basic food expenses.

[0093] The system combines factors such as their historical consumption patterns, family financial status, and the timing of scholarship awards to predict that they may be facing financial difficulties or psychological problems.

[0094] The output result is the "Consumption Abnormality Index" for reference by the funding department to decide whether to intervene.

[0095] It should be noted that in specific implementation scenarios, the above solution can be further enhanced by introducing unstructured data such as speech recognition and social media text analysis to enrich the input dimensions of the behavior prediction model. For example, by analyzing students' social media posts, emotional fluctuations or negative emotional tendencies can be identified to assist in determining their mental health status.

[0096] In specific implementation scenarios, building on the above solution, the system can automatically adjust model parameters based on newly collected behavioral data during model execution to improve prediction accuracy. For example, behavioral patterns at the beginning and end of a semester differ significantly, so the model can dynamically optimize weight distribution based on the time period.

[0097] In specific implementation scenarios, the above solution can be expanded beyond single-target prediction to include group behavior prediction. For example, if multiple students in a class exhibit similar abnormal behavior, the system can predict whether there is overall academic pressure or teaching problems.

[0098] In specific implementation scenarios, the system can also build on the above solution by training dedicated sub-models for different types of abnormal behaviors (such as learning, life, and safety) to improve prediction targeting. For example, for high-risk students, a more refined model can be used for in-depth behavior prediction.

[0099] In specific implementation scenarios, building on the above solution, the behavior prediction model can be integrated with academic affairs systems, psychological counseling systems, financial aid systems, and other systems to enable cross-platform data analysis. For example, if the model predicts that a student has serious psychological issues, relevant information can be directly sent to the psychological counseling center for timely intervention.

[0100] S400 , comparing the behavior prediction results with a preset threshold, and analyzing the behavior prediction results exceeding the preset threshold through abnormal management to obtain abnormal behavior management analysis results.

[0101] As mentioned above, after predicting the target's behavior, the system generates predictions across one or more dimensions (such as risk score and anomaly probability), reflecting the likelihood of the target exhibiting abnormal behavior in the future. To determine whether further action is necessary, the system compares these predictions against pre-set thresholds.

[0102] If the prediction result does not exceed the preset threshold: it means that the target object's behavior is within the normal range, and the system can continue to monitor its subsequent behavior without immediate intervention.

[0103] If the prediction result exceeds the preset threshold, it indicates that the target object has a high probability of exhibiting abnormal behavior. At this time, the system will initiate the abnormal management analysis process, comprehensively evaluate the situation based on multi-source information, and output the corresponding management analysis results.

[0104] The “abnormal management analysis” includes but is not limited to: identifying abnormal types, analyzing possible causes, matching historical cases, recommending intervention measures, etc., and ultimately forming a set of analytical conclusions that can be used to guide actual management work.

[0105] For example, a student behavior prediction model outputs that his or her "academic performance risk score is 85 points", while the risk threshold set by the system is 80 points.

[0106] The system determines that the score exceeds the threshold and triggers the exception management analysis process.

[0107] The abnormal management analysis module retrieves the student's recent classroom performance, homework submission status, psychological assessment data and other information, and concludes that the decline in his grades may be caused by academic pressure and depression.

[0108] Output analysis results: Recommend arranging academic tutoring and providing psychological counseling support.

[0109] A student has been late for two consecutive weeks. The behavior prediction model outputs that his "abnormal attendance probability is 72%" and the system sets the threshold to 65%.

[0110] Once the threshold is exceeded, the system automatically notifies the class teacher to pay attention to the student's status and initiates abnormal management analysis.

[0111] The analysis results show that the student’s recent frequent absences are related to sudden changes in his family. It is recommended to adjust the attendance policy and strengthen humanistic care.

[0112] The final output includes recommended communication methods, follow-up plans and a list of responsible personnel.

[0113] The amount of spending on a student's campus card decreased significantly in the past week, and the behavior prediction model outputted that his "consumption abnormality index was 90 points", which was higher than the 85-point threshold set by the system.

[0114] The system triggers abnormal management analysis and makes a comprehensive judgment based on data such as financial support records and frequency of social activities.

[0115] The analysis results indicate possible financial difficulties and recommend notifying the student aid department for intervention and providing financial counseling resources.

[0116] It should be noted that in specific implementation scenarios, the system can also dynamically adjust the threshold standards for different subjects or scenarios based on factors such as time periods (such as the beginning of a semester, exam week), changes in group behavior patterns, and other factors based on the above solution to improve the accuracy and adaptability of judgments. For example, the threshold for abnormal learning behavior can be appropriately raised during exams to avoid misjudging temporary learning fluctuations.

[0117] In specific implementation scenarios, the above approach can be combined with the introduction of multiple independently trained behavior prediction sub-models to predict different types of behaviors (such as learning, life, and safety), and then use a fusion mechanism to comprehensively determine the overall risk level. This approach can improve the robustness and prediction accuracy of the system and is particularly suitable for the comprehensive assessment of complex behaviors.

[0118] In specific implementation scenarios, building on the above solution, the system can automatically analyze and push high-risk warning results to a manual review platform, where managers can make the final judgment based on the actual situation. Furthermore, review feedback can be used as training data to feed back into the model, enabling continuous optimization.

[0119] In specific implementation scenarios, building on the above solution, the analysis results of this step can be directly integrated into a tiered intervention mechanism, automatically triggering management measures at corresponding levels based on different risk levels. For example, a mild anomaly will only send a reminder; a moderate anomaly will trigger a special intervention; and a severe anomaly will establish an investigation team and a long-term tracking mechanism.

[0120] In specific implementation scenarios, building on the above solution, the results of abnormal management analysis can be linked with other campus management systems (such as the academic affairs system, mental health system, and security system) to achieve information sharing and collaborative processing. For example, if a student is marked as high-risk, the system can automatically synchronize their information with the psychological counseling center and dormitory administrators, forming a multi-party collaborative intervention mechanism.

[0121] According to one embodiment of the present application, the acquired campus scene data is analyzed to identify special events, specifically:

[0122] Through the cameras, sensors and access control system equipment installed in the smart campus, scene data of multiple areas are collected in real time. The scene data includes image data, sensor data and positioning data, and the scene data is denoised, format converted and cleaned;

[0123] Analyze video data through image recognition algorithms to identify people's activity trajectories and gathering behavior patterns;

[0124] Perform time series analysis on sensor data to identify abnormal data points;

[0125] Use clustering algorithms to analyze abnormal clustering patterns within the region;

[0126] Based on the analysis results, the identified events are classified as special events.

[0127] As mentioned above, a smart campus is equipped with a variety of monitoring and data collection devices, such as cameras, temperature and humidity sensors, human presence sensors, and access control systems. These devices are distributed across different areas of the campus and collect real-time raw data related to the campus environment and activities. Specifically, cameras primarily capture video images, sensors monitor environmental parameters (such as temperature, humidity, and human activity), and access control systems record the time and location of people entering and exiting.

[0128] Scenario data covers multiple types of data sources:

[0129] Image data: Video streams or static images from cameras, which can be used to identify people's activity trajectories, facial features, etc.

[0130] Sensor data: Environmental parameters provided by temperature, humidity, and human body sensors, etc., help understand the environmental conditions in a specific area.

[0131] Positioning data: Location information of people or objects obtained through Wi-Fi, RFID, or Bluetooth technology helps track their movement paths.

[0132] To ensure the effectiveness of subsequent analysis, raw data must first be preprocessed. This includes removing noise (such as interfering pixels in images and outliers in sensor data) and unifying data from different sources into a standard format for subsequent processing and analysis. For example, video data may require resolution or frame rate adjustment, while sensor data may require unit conversion or time synchronization.

[0133] Using advanced image recognition algorithms (such as convolutional neural networks (CNNs)), video data can be deeply analyzed. The specific steps are as follows:

[0134] First, face detection technology is used to extract all the faces of people appearing in the video, and their identities are determined through face recognition technology.

[0135] Then, the moving path of the person in the video is tracked to generate an activity trajectory map.

[0136] Finally, clustering algorithms are applied to identify behavioral patterns of people gathering and mark areas or time periods where anomalies may exist.

[0137] For sensor data, time series analysis methods are used to detect anomalous data points. Time series analysis can help identify trends and cyclical changes in data. For example, by comparing historical data, it is possible to identify a sudden increase in temperature or a significant increase in the frequency of a person's activity within a certain period of time. These anomalous data points may be early signs of potential problems and require further attention and investigation.

[0138] In addition to image and sensor data, clustering algorithms (such as K-means) can be used to analyze activity patterns across an entire area. This approach can help identify areas with unusual clustering. For example, if a large number of people gather on a certain floor of a building during non-school hours, this may indicate a potential safety hazard or other issue of concern. Using clustering algorithms, the system can automatically detect and flag these unusual clustering patterns.

[0139] After completing the aforementioned data analysis, the system categorizes identified events as "special events" based on the results. These events are typically those of potential significance, such as unusual gatherings of people, frequent entry and exit at unusual times, or a significant decline in academic performance. By categorizing these events, the system can take more targeted measures, such as further identifying target individuals or developing appropriate management strategies. This not only improves the system's response efficiency but also enhances its support for campus safety management.

[0140] According to an embodiment of the present application, determining the target object associated with the special event according to the special event is specifically:

[0141] Filter data related to special events from campus scene data;

[0142] If the relevant data is image data, identity information is extracted through a face recognition algorithm and compared with the campus database;

[0143] If the relevant data is access records or check-in information, identity information is extracted from it and matched against the campus database to determine the target object.

[0144] As mentioned above, after identifying a special event, the system will extract information directly related to the event from the previously collected multi-source scene data. For example, if the event of interest is "entering and exiting the campus at abnormal times," the system will retrieve the entry and exit records of the access control system, the video images captured by the camera, and any positioning information that may be involved during the corresponding time period. The system will then classify and determine the data to determine what type of data it belongs to, such as image data or text data (such as access records, sign-in information, etc.), so that different methods can be used to identify specific target objects.

[0145] When the target data is image data, the system uses facial recognition technology to extract facial features of people appearing in the video or picture and compares them with the student or faculty photo information in the campus database. This process usually includes three steps: face detection, feature extraction, and identity matching. First, the system locates the face area in the image; second, it extracts the key feature points of the face (such as the distance between the eyes, the shape of the nose, etc.) to form a feature vector; finally, it compares the feature vector with the known person information in the database to identify the specific individual identity. If the match is successful, the person can be confirmed as the target object.

[0146] When the target data is text-based, such as access control system card swipe records, class attendance sheets, and library borrowing records, the system extracts identity information such as student ID, work ID, and name pinyin. The system then matches these fields with personnel information in the campus database to confirm the corresponding student or faculty member's identity. For example, if an access control record shows a card number entering a teaching building late at night, the system will query the corresponding registrant information based on the card number and ultimately determine whether the person is a target of concern.

[0147] According to one embodiment of the present application, the preset behavior prediction model is specifically:

[0148] Load the basic behavior prediction model;

[0149] Using historical event record data similar to the special event to train the initial model to generate a first intermediate model;

[0150] Use the first intermediate model to predict the data of associated objects and associated events to generate pseudo labels;

[0151] Generate enhanced data versions by time shifting, time scaling, or adding noise;

[0152] Use the augmented data version to perform unsupervised training on the model to generate a second intermediate model;

[0153] The feature representation extracted by the second intermediate model is used as input, and supervised and unsupervised training are performed alternately until the model performance reaches the preset conditions.

[0154] As mentioned above, the system first needs to load a pre-trained basic behavior prediction model. This initial model can be pre-trained on a general dataset. Its purpose is to provide a preliminary behavior prediction framework as a foundation for subsequent training. This model typically includes some basic behavioral pattern recognition capabilities and feature extraction layers, but has not yet been trained on data specific to campus events.

[0155] After loading the base model, the system conducts supervised training on historical event records similar to the current event of interest (such as abnormal clustering or frequent absenteeism). These historical event records are labeled with specific types of abnormal behavior, such as a student's frequent tardiness or a sudden drop in grades over a certain period of time. Through this supervised learning approach, the system teaches the model how to distinguish between normal and abnormal behavior, adjusts model parameters based on these known patterns, and generates a preliminarily optimized first intermediate model.

[0156] Next, the system uses the newly generated first intermediate model to predict data about related objects (such as classmates and roommates) and related events (such as students' daily activity records and social interactions), generating pseudo-labels. Pseudo-labels are predictive classification labels that the model assigns to unlabeled data based on its existing knowledge. This method allows the system to provide preliminary classification information for more data in the absence of explicit labels, paving the way for further unsupervised training.

[0157] In order to improve the generalization and robustness of the model, the system will enhance the existing training data. Specific methods include:

[0158] Time shift: offset the time axis of the original data to simulate behavior patterns in different time periods.

[0159] Time scaling: Changing the time scale of the raw data allows the model to adapt to behavioral changes at different frequencies.

[0160] Adding noise: Adding random noise to the data helps the model better cope with the imperfect data in the real world.

[0161] These enhancement techniques can generate multiple variant versions of the data, allowing the model to be trained on a wider range of situations, thereby improving its performance.

[0162] With the enhanced data, the system begins unsupervised training of the model. In this phase, the system no longer relies on explicit labels, but instead draws on the structure and patterns of the data itself to learn. Through unsupervised learning, the model can discover underlying patterns and complex patterns in the data, further improving its understanding and predictive capabilities. Ultimately, this process produces a more optimized second intermediate model.

[0163] The final step is to combine the advantages of supervised and unsupervised learning and further optimize the model through alternating training. The specific approach is:

[0164] The feature representation extracted from the augmented data version by the second intermediate model is used as the new input data.

[0165] Alternate supervised and unsupervised training, that is, first use supervised data to update the model parameters, and then use unsupervised data to continue optimizing the model.

[0166] During each iteration, the system evaluates the model's performance indicators (such as accuracy and loss function value) and continuously adjusts the model until the performance meets the preset standards (such as reaching a specific accuracy threshold or the loss function converges).

[0167] According to one embodiment of the present application, the target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the behavior prediction result of the target object, specifically:

[0168] Integrate the target object's learning data and life data, and combine it with campus scene data;

[0169] The behavior of the target object is analyzed using a preset behavior prediction model, and potential abnormal behaviors that do not conform to the normal pattern are identified as the behavior prediction results of the target object.

[0170] As mentioned above, after determining the target object related to the special event, the system will collect various types of data on the object, including its learning behavior data and daily life behavior data during its time at school, and integrate these data with real-time or historical scene data collected from the campus environment.

[0171] Learning data: mainly includes course grades, homework submission status, class attendance records, exam performance, course selection information, etc., reflecting students' academic performance.

[0172] Life data: including but not limited to campus card spending records, dormitory entry and exit times, library usage frequency, physical exercise status, social activity participation, etc., used to characterize students' daily behavior patterns.

[0173] Campus scene data: multi-source heterogeneous data from cameras, access control systems, sensors and other devices, such as image data (used to identify behavioral trajectories), positioning data (used to determine people's movement paths), sensor data (used to monitor environmental changes), etc.

[0174] By integrating the above three types of data, a comprehensive portrait of the target object's behavior is formed, providing a rich data foundation for subsequent behavior modeling and anomaly identification.

[0175] Analyze the behavior of the target object using the preset behavior prediction model and identify potential abnormal behaviors that do not conform to the normal pattern as the behavior prediction results of the target object:

[0176] After data integration, the system feeds this multi-dimensional data into a trained behavior prediction model. This model, trained on a vast amount of historical data, is able to identify normal behavior patterns and issue warnings for deviations from them.

[0177] Functions of behavior prediction models:

[0178] Establish an individual behavior baseline based on the target object's historical behavior;

[0179] Analyze the differences between current behavior and historical behavior;

[0180] Determine whether there are trends or sudden changes that deviate significantly from the normal pattern;

[0181] Output behavior prediction results, usually expressed as risk scores, abnormality probabilities, or behavior trend prediction values.

[0182] Contents of behavior prediction results:

[0183] This may include outputs from multiple dimensions, such as the risk of academic performance decline, the probability of abnormal attendance, the consumer behavior fluctuation index, and the abnormal campus entry and exit score.

[0184] Each result corresponds to a specific type of abnormal behavior, which makes it easier for the subsequent management and analysis module to take appropriate intervention measures based on different abnormality types.

[0185] Ways to identify potentially abnormal behavior:

[0186] The system compares the prediction results with the set benchmark value or dynamic threshold;

[0187] If an indicator is significantly higher than the baseline value, it is determined to be a potential abnormal behavior;

[0188] At the same time, combined with time series analysis, it is possible to determine whether the anomaly is persistent or recurring, and further confirm its severity.

[0189] Through the above process, the system can predict the abnormal behaviors that may occur in the future of the target object and output structured behavior prediction results, providing a decision-making basis for subsequent risk assessment and management intervention.

[0190] According to one embodiment of the present application, the behavior prediction result is compared with the preset threshold, and the behavior prediction result exceeding the preset threshold is analyzed through abnormal management to obtain an abnormal behavior management analysis result, specifically:

[0191] The preset threshold is a risk score critical value derived from statistical analysis of historical behavior data, and is used to determine whether the target object's behavior deviates from the normal behavior pattern;

[0192] The behavior prediction result is a risk score for one or more dimensions, including academic performance risk score, attendance risk score, consumer behavior risk score, and entry and exit record risk score. The score of each dimension is compared with the corresponding preset threshold.

[0193] If the behavior prediction result does not exceed the preset threshold, the behavior is marked as normal behavior and the current monitoring status is maintained;

[0194] If the preset threshold is exceeded, the behavior will be marked as abnormal and the abnormal management process will be triggered;

[0195] Determine the abnormality type based on the behavior prediction result exceeding the threshold;

[0196] Extracting contextual information related to the anomaly type, including student profiles, historical behavioral data, and social relationship data;

[0197] According to the abnormality type and the corresponding background information, an abnormal behavior management analysis result is formed, and the abnormal behavior management analysis result includes personalized management and graded intervention measures.

[0198] As mentioned above, the system's pre-set thresholds are determined through statistical modeling and machine learning analysis of a large amount of historical behavioral data. This historical data includes typical student behavior across various scenarios, such as grade fluctuations, attendance frequency distribution, spending trends, and temporal patterns of entry and exit records.

[0199] Based on this data, the system builds a baseline model of normal behavior and sets a critical value (i.e., a threshold) for risk scoring. This threshold serves as a criterion for determining whether the target's behavior is within the normal range. If the behavior score is lower than or equal to the threshold, it is considered normal behavior; if the score is higher than the threshold, it may indicate that the behavior has deviated from the normal pattern and is potentially abnormal.

[0200] The behavior prediction result is a risk score for one or more dimensions, including academic performance risk score, attendance risk score, consumer behavior risk score, and entry and exit record risk score. The score for each dimension is compared with the corresponding preset threshold:

[0201] The output of the behavior prediction model is not a single numerical value, but a risk score covering multiple dimensions, each of which corresponds to a specific type of student behavior, including:

[0202] Academic performance risk score: reflects the possibility of a student's academic performance declining or continuing to be unsatisfactory;

[0203] Attendance risk score: assesses the probability that a student will frequently arrive late, leave early, or be absent from class;

[0204] Consumption behavior risk score: measures the extent to which students' campus card spending patterns have changed significantly;

[0205] Access record risk score: Identifies the frequency with which students enter and exit the campus or specific areas at unusual times.

[0206] The system sets a preset threshold for each risk score and compares each one to determine if there are any abnormal trends in any particular area. For example, even if a student's attendance is good, if their consumption behavior score far exceeds the threshold, relevant management procedures must still be initiated.

[0207] If the behavior prediction result does not exceed the preset threshold, the behavior is marked as normal behavior and the current monitoring status is maintained; if it exceeds the preset threshold, the behavior is marked as abnormal behavior and the abnormality management process is triggered:

[0208] After completing the multi-dimensional score and threshold comparison, the system will make the following processing:

[0209] Normal behavior handling: If the scores of all dimensions do not exceed their respective thresholds, the target object's current behavior is considered to be within the normal range, and the system will continue to monitor it normally without taking further intervention measures.

[0210] Abnormal behavior handling method: If the score of any dimension exceeds the threshold, the behavior will be marked as "abnormal behavior" and the subsequent abnormality management process will be automatically triggered, entering the in-depth analysis and strategy formulation stage.

[0211] Determine the anomaly type based on the behavior prediction results that exceed the threshold:

[0212] After marking as abnormal behavior, the system will determine the type of abnormality based on which type of risk score exceeds the standard. For example:

[0213] If the academic performance risk score exceeds the standard, it will be judged as "abnormal academic performance";

[0214] If the attendance risk score exceeds the standard, it will be judged as "abnormal attendance";

[0215] If the consumption behavior score exceeds the standard, it will be judged as "abnormal consumption";

[0216] If the entry and exit record score exceeds the standard, it will be judged as "abnormal entry and exit record".

[0217] This classification helps to carry out subsequent management processes more accurately and avoid a one-size-fits-all approach.

[0218] Extract background information related to the anomaly type, including student profiles, historical behavioral data, and social relationship data:

[0219] To more fully understand the causes of abnormal behavior, the system extracts contextual information related to the abnormality type from multiple sources, including:

[0220] Student personal files: such as family background, health status, reward and punishment records, etc.;

[0221] Historical behavioral data: such as changes in grades, attendance records, spending habits, etc. over the past few years;

[0222] Social relationship data: such as the frequency of interactions with classmates and roommates, participation in group activities, etc.

[0223] This background information can help the system determine the possible causes behind abnormal behavior, such as academic pressure, mental health issues, financial difficulties or interpersonal conflicts.

[0224] Based on the abnormality type and the corresponding background information, the abnormal behavior management analysis results are formed, which include personalized management and graded intervention measures:

[0225] After integrating the anomaly type and background information, the system will generate the final abnormal behavior management analysis results, including but not limited to the following two aspects:

[0226] Personalized management strategy:

[0227] For abnormal academic performance: suggest increasing tutoring time, adjusting course schedules, and recommending learning resources;

[0228] For abnormal attendance: provide psychological counseling services and adjust attendance rules;

[0229] Regarding abnormal consumption: investigating financial difficulties and providing financial consultation;

[0230] Regarding abnormal entry and exit records: strengthen access and exit authority management and carry out safety education.

[0231] Tiered interventions:

[0232] Risk level 1: Light reminders are sent, and the class teacher or counselor regularly follows up on the student's status;

[0233] Second risk level: A priority alert is triggered, and a psychological counselor is involved in the assessment and ongoing monitoring of behavioral changes;

[0234] Third risk level: An emergency alarm is triggered, and the school sets up a special team to investigate and establish a long-term tracking mechanism.

[0235] Through the above process, the system can realize a complete closed-loop process from behavior prediction to anomaly identification to management intervention, forming a scientific, efficient and personalized campus abnormal behavior management system.

[0236] According to one embodiment of the present application, the personalized management is specifically:

[0237] Provide different management methods according to different exception types;

[0238] Dynamically adjust management methods based on the severity of abnormal behavior and the student's background information.

[0239] As mentioned above, student behavior abnormalities are diverse and complex during campus behavior management. Different types of abnormal behavior have different causes and require different interventions. Therefore, the system automatically matches the corresponding management approach based on the specific abnormality type identified, achieving a precise and categorized management response.

[0240] Common exception types include but are not limited to the following:

[0241] Abnormal academic performance: such as a sudden drop in grades, a decrease in homework submission rate, etc.;

[0242] Abnormal attendance: such as frequent lateness, early departure or unexplained absences;

[0243] Abnormal consumption behavior: such as a sudden drop or surge in campus card spending;

[0244] Abnormal entry and exit records: such as frequent entry and exit of dormitory areas or teaching buildings at abnormal times.

[0245] For each of the above exception types, the system presets corresponding management strategies:

[0246] Abnormal academic performance: notify the instructor or counselor and suggest increasing tutoring time, optimizing learning resource recommendations, adjusting course schedules, etc.

[0247] Abnormal attendance: The class teacher or counselor will investigate the situation, handle it flexibly based on the attendance rules, and provide psychological counseling services as appropriate;

[0248] Abnormal spending: Notify the student aid department to investigate financial difficulties, provide financial counseling support, and monitor for abnormal spending behavior (such as stolen credit cards);

[0249] Abnormal access records: Notify the security department to strengthen access authority management, conduct safety education if necessary, and initiate personnel verification process.

[0250] This type-based management approach ensures that intervention measures match actual problems, improving management efficiency and targeting.

[0251] Dynamically adjust management methods based on the severity of abnormal behavior and student background information:

[0252] In addition to distinguishing between abnormal types, the system further considers the severity of the abnormal behavior itself and the individual background information of the target object, thereby achieving dynamic adjustment of management methods and improving management flexibility and adaptability.

[0253] Adjustments are made based on the severity of the abnormal behavior:

[0254] The system's risk assessment module will quantify and score each type of abnormal behavior and set multiple risk levels (such as low risk, medium risk, and high risk). Different levels correspond to different management intensities:

[0255] Low risk: Only reminders are sent, and relevant managers pay attention to them regularly;

[0256] Medium risk: Initiate special intervention processes, such as arranging psychological counseling, temporarily adjusting course schedules, etc.;

[0257] High risk: The emergency response mechanism is triggered, and the school sets up a special team to investigate and establish a long-term tracking mechanism.

[0258] Adjustments based on student's personal background information:

[0259] Individual differences among students dictate that a unified management approach cannot be adopted. When formulating management strategies, the system will also comprehensively analyze the following background information of students:

[0260] Family status (e.g., whether from a family with financial difficulties, whether there have been any family changes);

[0261] Health records (such as whether there are any chronic diseases, mental health records);

[0262] Social relationships (e.g., lack of interpersonal communication, whether influenced by groups);

[0263] Past behavior records (such as whether there is similar historical abnormal behavior).

[0264] For example:

[0265] For abnormal attendance caused by sudden family changes, the system can suggest a certain attendance allowance and arrange psychological counseling;

[0266] For abnormal consumption caused by financial difficulties, the system will give priority to recommending scholarship applications or temporary subsidies;

[0267] For students with multiple abnormal entry and exit records, the system will upgrade the management level, strengthen access and exit permission control, and include them in the key monitoring list.

[0268] Dynamic update management policy:

[0269] The system supports a dynamic update mechanism for management strategies. With the continuous input of new data (such as behavioral changes after intervention, feedback evaluation, etc.), the system can evaluate the effectiveness of current management measures in real time and adjust strategies based on actual conditions.

[0270] For example:

[0271] If an intervention measure does not achieve the expected effect, the system will automatically recommend a higher-level intervention plan;

[0272] If the student's behavior gradually returns to normal, the system can gradually reduce the management intensity and return to normal monitoring status.

[0273] According to one embodiment of the present application, the graded intervention measures are specifically:

[0274] Conduct time series analysis on identified abnormal behaviors to obtain development trends of abnormal behaviors;

[0275] Risk levels are divided according to the severity of abnormal behavior and its development trend;

[0276] Trigger different intervention measures based on different risk levels;

[0277] If risky behavior is not intervened in time and the risk escalates, the intervention measures will be automatically upgraded from automatic alarm to manual verification.

[0278] As mentioned above, after the system identifies a target user's unusual behavior, it conducts time series analysis on their historical data to further assess its potential impact and development trends. This analysis model identifies changing trends by modeling behavioral records over a period of time (such as changes in grades, attendance frequency, and spending amounts).

[0279] For example, if a student has been late for three consecutive weeks, the system will extract their daily attendance records and build a time series model to determine whether the behavior is occasional fluctuation or continuous deterioration;

[0280] For example, if a student's campus card spending amount has been gradually decreasing in the past two weeks, the system can determine whether there is a possibility of financial difficulties or psychological problems through trend fitting.

[0281] Through time series analysis, the system can more accurately determine whether abnormal behavior has a tendency to continue and worsen, providing a basis for subsequent risk level classification.

[0282] Risk levels are divided according to the severity of abnormal behavior and its development trend:

[0283] After obtaining the type of abnormal behavior, current score, and development trend, the system will classify the risk level based on two key factors:

[0284] Severity of abnormal behavior:

[0285] That is, the degree to which the current behavior deviates from the normal pattern, which is usually reflected by the risk score output by the behavior prediction model.

[0286] For example, a 50% drop in academic performance is more serious than a 10% drop;

[0287] A sudden drop in consumption to zero is more alarming than a small fluctuation.

[0288] Development trend of abnormal behavior:

[0289] If abnormal behavior shows a trend of continued deterioration (e.g., grades decline month by month, attendance rate continues to decrease), it indicates that the risk is increasing;

[0290] If the abnormal behavior shows short-term fluctuations and has stabilized, it can be judged as low risk.

[0291] Based on the above two dimensions, the system divides risks into multiple levels, such as:

[0292] Level 1 risk (low risk): The abnormality is mild, with no obvious development trend, and no immediate intervention is required;

[0293] Second risk level (medium risk): The abnormality is more obvious and there is a certain development trend, and preliminary intervention measures need to be initiated;

[0294] The third risk level (high risk): extremely serious and showing a trend of continuous deterioration, requiring immediate and intensive management measures.

[0295] Different intervention measures are triggered according to different risk levels:

[0296] The system automatically matches and triggers corresponding intervention measures based on the risk levels after classification, forming a hierarchical response mechanism to ensure rational allocation of resources and accurate and effective intervention measures.

[0297] First risk level (low risk):

[0298] Trigger lightweight reminders, such as sending notification messages to class teachers or counselors;

[0299] The system maintains a routine monitoring state, regularly updates the behavior score, and observes whether there are further changes.

[0300] Second risk level (medium risk):

[0301] Initiate a priority alert and notify relevant management personnel (such as psychological counselors and funding specialists) to intervene;

[0302] Arrange a dedicated follow-up plan, such as generating weekly behavioral change reports to evaluate intervention effectiveness;

[0303] Provide personalized support based on the type of abnormality, such as arranging academic tutoring, psychological counseling or financial counseling.

[0304] Third risk level (high risk):

[0305] Trigger an emergency alarm to notify the security department, student affairs office, and school management;

[0306] Establish a special investigation team to conduct in-depth situation verification and comprehensive intervention;

[0307] Establish a long-term tracking mechanism, formulate phased intervention plans, and dynamically adjust management strategies.

[0308] If risky behavior is not intervened in time and the risk escalates, the intervention measure will be automatically upgraded from automatic alarm to manual verification:

[0309] In order to prevent abnormal behavior from further deteriorating due to untimely intervention, the system has set up a risk escalation detection mechanism.

[0310] When the system detects the following situations, it will automatically increase the intervention level:

[0311] A behavior originally marked as low-risk does not improve within the set timeframe but instead continues to worsen;

[0312] A certain type of abnormal behavior exceeds the standard in multiple dimensions simultaneously, showing complex risk characteristics;

[0313] The original intervention measures failed to effectively control the development trend of behavior, and the system determined that human intervention needed to be strengthened.

[0314] At this point, the system upgrades the original automatic warning mechanism to a manual verification process, which includes:

[0315] Automatically push detailed behavior analysis reports to relevant personnel (such as counselors, psychologists, and security personnel);

[0316] Launch a manual interview mechanism, with dedicated personnel conducting face-to-face communication and assessment of students;

[0317] The manual verification results are fed back to the system to update the behavior prediction model and risk assessment system.

[0318] Through this mechanism, the system achieves seamless connection from automated monitoring to deep manual intervention, improves the ability to deal with complex abnormal behaviors, and ensures the timeliness and effectiveness of campus management.

[0319] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk, or an optical disk.

[0320] On the other hand, the present invention further provides a computer program product, comprising a computer program. The computer program may be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the method provided by each of the above methods, including:

[0321] Analyze the acquired campus scene data to identify special events;

[0322] Determine target objects associated with special events based on special events;

[0323] The target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the target object's behavior prediction results;

[0324] The behavior prediction results are compared with the preset threshold, and the behavior prediction results that exceed the preset threshold are analyzed through abnormal management to obtain abnormal behavior management analysis results.

[0325] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for cigarette box image recognition provided by the above methods is implemented, and the method includes:

[0326] Analyze the acquired campus scene data to identify special events;

[0327] Determine target objects associated with special events based on special events;

[0328] The target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the target object's behavior prediction results;

[0329] The behavior prediction results are compared with the preset threshold, and the behavior prediction results that exceed the preset threshold are analyzed through abnormal management to obtain abnormal behavior management analysis results.

[0330] Anything not described in this application can be achieved by adopting or drawing on existing technologies.

[0331] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0332] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A data analysis and behavior prediction management method based on campus security network, characterized in that: include: Analyze the acquired campus scene data to identify special events; Determine target objects associated with special events based on special events; The target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the target object's behavior prediction results; The behavior prediction results are compared with the preset threshold, and the behavior prediction results that exceed the preset threshold are analyzed through abnormal management to obtain abnormal behavior management analysis results.

2. The method according to claim 1, characterized in that The acquired campus scene data is analyzed to identify special events, specifically: Through the cameras, sensors and access control system equipment installed in the smart campus, scene data of multiple areas are collected in real time. The scene data includes image data, sensor data and positioning data, and the scene data is denoised, format converted and cleaned; Analyze video data through image recognition algorithms to identify people's activity trajectories and gathering behavior patterns; Perform time series analysis on sensor data to identify abnormal data points; Use clustering algorithms to analyze abnormal clustering patterns within the region; Based on the analysis results, the identified events are classified as special events.

3. The method according to claim 1, characterized in that The step of determining the target object associated with the special event according to the special event is as follows: Filter data related to special events from campus scene data; If the relevant data is image data, identity information is extracted through a face recognition algorithm and compared with the campus database; If the relevant data is access records or check-in information, identity information is extracted from it and matched against the campus database to determine the target object.

4. The method according to claim 1, wherein The preset behavior prediction model is specifically: Load the basic behavior prediction model; Using historical event record data similar to the special event to train the initial model to generate a first intermediate model; Use the first intermediate model to predict the data of associated objects and associated events to generate pseudo labels; Generate enhanced data versions by time shifting, time scaling, or adding noise; Use the augmented data version to perform unsupervised training on the model to generate a second intermediate model; The feature representation extracted by the second intermediate model is used as input, and supervised and unsupervised training are performed alternately until the model performance reaches the preset conditions.

5. The method according to claim 1, characterized in that The target object is combined with campus scene data and analyzed through a preset behavior prediction model to identify the behavior prediction result of the target object, specifically: Integrate the target object's learning data and life data, and combine it with campus scene data; The behavior of the target object is analyzed using a preset behavior prediction model, and potential abnormal behaviors that do not conform to the normal pattern are identified as the behavior prediction results of the target object.

6. The method according to claim 1, characterized in that The behavior prediction results are compared with the preset threshold, and the behavior prediction results exceeding the preset threshold are analyzed through abnormal management to obtain abnormal behavior management analysis results, specifically: The preset threshold is a risk score critical value derived from statistical analysis of historical behavior data, and is used to determine whether the target object's behavior deviates from the normal behavior pattern; The behavior prediction result is a risk score for one or more dimensions, including academic performance risk score, attendance risk score, consumer behavior risk score, and entry and exit record risk score. The score of each dimension is compared with the corresponding preset threshold. If the behavior prediction result does not exceed the preset threshold, the behavior is marked as normal behavior and the current monitoring status is maintained; If the preset threshold is exceeded, the behavior will be marked as abnormal and the abnormal management process will be triggered; Determine the abnormality type based on the behavior prediction result exceeding the threshold; Extracting contextual information related to the anomaly type, including student profiles, historical behavioral data, and social relationship data; According to the abnormality type and the corresponding background information, an abnormal behavior management analysis result is formed, and the abnormal behavior management analysis result includes personalized management and graded intervention measures.

7. The method according to claim 6, characterized in that The personalized management is specifically: Provide different management methods according to different exception types; Dynamically adjust management methods based on the severity of abnormal behavior and the student's background information.

8. The method according to claim 7, characterized in that The graded intervention measures are specifically: Conduct time series analysis on identified abnormal behaviors to obtain development trends of abnormal behaviors; Risk levels are divided according to the severity of abnormal behavior and its development trend; Trigger different intervention measures based on different risk levels; If risky behavior is not intervened in time and the risk escalates, the intervention measures will be automatically upgraded from automatic alarm to manual verification.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method according to any one of claims 1 to 8 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the method according to any one of claims 1 to 8 are implemented.

Images