A campus anti-bullying system based on large models

Abstract

The present invention relates to the technical field of security event monitoring. Specifically, it relates to a campus anti-bullying system based on a large model, which includes: a multi-modal signal perception unit that uses a sensor network to collect multi-modal bullying signals in real time; a large model intelligent analysis unit that extracts spatio-temporal features of video signals and audio signals based on the multi-modal bullying signals, constructs an audio-visual fluid momentum analysis equation to detect physical violence actions and semantic threats, and then constructs a physiological response perception analysis equation to analyze individual physiological responses, and comprehensively obtains the analysis results of bullying events; a dynamic early warning and response unit that uses a risk grading strategy to early warn and respond to bullying events in real time; and an event retention record unit that stores multi-modal bullying signals and bullying event analysis result data. The campus anti-bullying system based on the large model combines the audio-visual fluid momentum analysis equation and the physiological response perception analysis equation to detect physical violence and semantic threats, and early warns and responds to bullying events.

Description

Technical Field

[0001] The present invention relates to the technical field of security event monitoring, and more specifically, to a campus anti-bullying system based on a large model. Background Art

[0002] The campus anti-bullying system based on a large model aims to accurately detect and respond to bullying incidents on campus in real time and analyze the interaction relationship between students' physiological responses and environmental location signals. Through the deep fusion of multi-modal signals and the innovative application of hydrodynamics and reaction-diffusion equations, it controls the early warning and dynamic response of bullying behaviors, and realizes the monitoring, evaluation, and response to bullying incidents.

[0003] Existing campus anti-bullying systems usually have difficulty effectively integrating video, audio, location signals, and physiological response signals, and cannot accurately capture multi-modal bullying signals in complex scenarios. Moreover, since bullying incidents usually occur relatively secretly, a single video signal cannot be detected in time and accurately, which will lead to a slow response to low-intensity bullying incidents and miss the best opportunity for timely intervention. For example, the hidden threats or subtle emotional changes cannot be recognized in time, and ultimately there are large blind spots for school administrators in timely identification and taking effective measures. Therefore, a campus anti-bullying system based on a large model is designed. Summary of the Invention

[0004] The purpose of the present invention is to provide a campus anti-bullying system based on a large model to solve the problem proposed in the above background art that due to the relatively secret occurrence of bullying incidents, a single video signal cannot be detected in time and accurately, which will lead to a slow response to low-intensity bullying incidents and miss the best opportunity for timely intervention.

[0005] To achieve the above purpose, the present invention aims to provide a campus anti-bullying system based on a large model, including:

[0006] A multi-modal signal perception unit, which uses a sensor network to collect multi-modal bullying signals in real time and preprocesses the multi-modal bullying signals;

[0007] It further includes a large model intelligent analysis unit, which extracts the spatio-temporal features of video signals and audio signals based on the multi-modal bullying signals, constructs an audio-visual fluid momentum analysis equation to detect physical violence actions and semantic threats, and then constructs a physiological response perception analysis equation to analyze individual physiological responses, and comprehensively obtains the analysis result of bullying incidents;

[0008] It further includes a dynamic early warning and response unit, which based on the analysis result of the bullying situation, uses a risk grading strategy to give an early warning and respond to bullying incidents in real time;

[0009] It further includes an event retention record unit, which is used to store multi-modal bullying signals and bullying event analysis result data.

[0010] As a further improvement of this technical solution, the multi-modal signal perception unit includes a signal acquisition module and a signal processing module;

[0011] Among them, the spatio-temporal synchronization signal acquisition module uses a sensor network to collect multi-modal bullying signals in real time;

[0012] The multi-modal bullying signals include video signals, audio signals, position signals, and physiological signals;

[0013] The signal processing module is used to preprocess the multi-modal bullying signals collected by the signal acquisition module.

[0014] As a further improvement of this technical solution, the spatio-temporal synchronization signal acquisition module uses a sensor network to collect multi-modal bullying signals in real time. The specific method is as follows:

[0015] S1.1.1. Obtain video signals using a multi-spectral camera network ;

[0016] S1.1.2. Obtain audio signals using a microphone array ;

[0017] S1.1.3. Locate the student's position based on the multi-spectral camera to obtain the position signal ;

[0018] S1.1.4. Obtain physiological signals based on the biosensors worn by the students ;

[0019] Among them, is the heart rate; is the skin conductance response.

[0020] As a further improvement of this technical solution, the large model intelligent analysis unit includes a spatio-temporal behavior perception module, a physiological position analysis module, and a result judgment module;

[0021] Among them, the spatio-temporal behavior perception module is used to extract the spatio-temporal features of video signals and audio signals, and construct an audio-visual fluid momentum analysis equation to analyze limb violent actions and threatening semantic intensity;

[0022] The physiological position analysis module constructs a physiological response perception analysis equation based on the interaction between physiological signals and position signals to analyze individual physiological responses;

[0023] The result judgment module calculates the Lyapunov exponent based on the threat semantic intensity, and comprehensively obtains the analysis result of the bullying event by combining two situations of comparing the physiological response signal concentration with the physiological response signal concentration threshold.

[0024] As a further improvement of this technical solution, the spatio-temporal behavior perception module is used to extract the spatio-temporal features of video signals and audio signals, and construct an audio-visual fluid momentum analysis equation to analyze limb violent actions and threat semantic intensity. The specific method steps are as follows:

[0025] S2.1.1. Based on the video frame sequence , calculate the optical flow field and vorticity field, and combine them into video spatio-temporal features, including the optical flow field and the vorticity field ;

[0026] S2.1.2. Based on the sound information , calculate the sound pressure field and sound pressure density, and combine them into audio spatio-temporal features, including the sound pressure field and the sound pressure density ;

[0027] S2.1.3. Based on the optical flow field and the sound pressure field, construct an audio-visual fluid momentum analysis equation to analyze limb violent actions and threat semantic intensity. The audio-visual fluid momentum analysis equation is as follows:

[0028] ;

[0029] Among them, is the behavior density; is the video pressure field; is the social viscosity coefficient; is the limb conflict momentum transfer term; is the audio modulation force;

[0030] S2.1.4. Based on the vorticity field and the sound energy density, perform semantic threat analysis to calculate the threat semantic intensity .

[0031] As a further improvement of this technical solution, the physiological position analysis module constructs a physiological response perception analysis equation to analyze the individual physiological response based on the interaction between physiological signals and position signals. The specific method steps are as follows:

[0032] S2.2.1. Obtain the position signal and the physiological signal ;

[0033] S2.2.2. Analyze the individual's physiological response using the physiological response perception analysis equation based on the interaction between the physiological signal and the position signal, and calculate the physiological response signal concentration through the physiological response perception analysis equation;

[0034] S2.2.3. Solve the physiological response perception analysis equation to obtain the physiological response signal concentration at the position at time , and compare it with the physiological response signal concentration threshold:

[0035] If , the individual's physiological response is in danger and there is a bullying incident;

[0036] If , the individual's physiological response is relatively safe, and it is necessary to further determine whether there is a bullying incident;

[0037] Among them, is the physiological response signal concentration threshold.

[0038] As a further improvement of this technical solution, the result judgment module calculates the Lyapunov exponent based on the threat semantic intensity, and combines the two situations of comparing the physiological response signal concentration with the physiological response signal concentration threshold to comprehensively obtain the bullying incident analysis result, specifically as follows:

[0039] S2.3.1. Define the state vector, construct the state equation, and calculate the eigenvalues of the Jacobian matrix and the Lyapunov exponent based on the physiological response signal concentration and the threat semantic intensity:

[0040] The state vector is: ;

[0041] The state equation:

[0042] ;

[0043] The Jacobian matrix :

[0044] ;

[0045] The Lyapunov exponent:

[0046] ;

[0047] Among them, is the state vector; is the small change of the state vector;

[0048] S2.3.2. If , and the Lyapunov exponent , there is a bullying incident;

[0049] If , and the Lyapunov exponent , there is no bullying incident.

[0050] As a further improvement of this technical solution, the dynamic warning and response unit includes a risk grading module and an event response module.

[0051] As a further improvement of this technical solution, the risk grading module grades the risk of bullying incidents based on the concentration of physiological response signals and the intensity of threatening semantics, and generates a graded warning strategy;

[0052] The event response module executes the graded warning strategy in real time according to the risk grading result.

[0053] As a further improvement of this technical solution, the event retention record unit uses a distributed edge cloud data lake to store multi-modal bullying signals and bullying event analysis result data;

[0054] The multi-modal bullying signals include video signals, audio signals, location signals and physiological signals;

[0055] The bullying event analysis result data includes the concentration of physiological response signals, the intensity of threatening semantics, the data of the fluid momentum analysis equation for audio and video, and the data of the physiological response perception analysis equation.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] 1. In this campus anti-bullying system based on a large model, based on the fluid momentum analysis equation for audio and video, limb violent actions and threatening voices are identified, and through the spatio-temporal feature coupling of video signals and audio signals, dynamic analysis of physical conflicts and voice threats is carried out, so as to timely detect hidden bullying behaviors in complex campus scenarios, such as violent behaviors such as pushing and kicking and the emergence of threatening language.

[0058] 2. In this campus anti-bullying system based on a large model, through the physiological response perception analysis equation, in-depth coupling analysis of physiological signals and environmental location signals is realized. By monitoring the concentration of individual physiological responses, such as heart rate and skin conductance response, combined with the influence of the surrounding environmental location and audio signals, the physiological stress state of individuals can be accurately quantified, and by comparing with the physiological response signal threshold, it can be judged in real time whether an individual is at risk of bullying. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is the overall flow block diagram of the present invention;

[0060] The meanings of the various labels in the figure are as follows:

[0061] 1. Multimodal Signal Sensing Unit; 2. Large Model Intelligent Analysis Unit; 3. Dynamic Early Warning and Response Unit; 4. Event Retention and Recording Unit; 11. Signal Acquisition Module; 12. Signal Processing Module; 21. Spatiotemporal Behavior Sensing Module; 22. Physiological Location Analysis Module; 23. Result Judgment Module; 31. Risk Grading Module; 32. Event Response Module. Detailed Implementation Manner

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] Please refer to Figure 1 as shown, a campus anti-bullying system based on a large model is provided, including:

[0064] A multimodal signal sensing unit 1, which uses a sensor network to collect multimodal bullying signals in real time and preprocesses the multimodal bullying signals;

[0065] The multimodal signal sensing unit 1 includes a signal acquisition module 11 and a signal processing module 12;

[0066] Among them, the spatiotemporal synchronization signal acquisition module 11 uses a sensor network to collect multimodal bullying signals in real time;

[0067] The multimodal bullying signals include video signals, audio signals, location signals, and physiological signals;

[0068] The signal processing module 12 is used to preprocess the multimodal bullying signals collected by the signal acquisition module 11.

[0069] In this embodiment, the spatiotemporal synchronization signal acquisition module 11 uses a sensor network to collect multimodal bullying signals in real time. The specific method is as follows:

[0070] S1.1.1. Use a multispectral camera network to obtain video signals ;

[0071] S1.1.2. Use a microphone array to obtain audio signals ;

[0072] S1.1.3. Locate the student's position based on the multispectral camera to obtain the location signal ;

[0073] S1.1.4. Obtain physiological signals based on the biosensors worn by students ;

[0074] Wherein, is the heart rate; is the galvanic skin response.

[0075] It also includes a large model intelligent analysis unit 2. The large model intelligent analysis unit 2 extracts the spatio-temporal features of video signals and audio signals based on multi-modal bullying signals, constructs an audio-visual fluid momentum analysis equation to detect limb violent actions and semantic threats, and then constructs a physiological response perception analysis equation to analyze the individual's physiological response, and comprehensively obtains the analysis result of the bullying event;

[0076] The large model intelligent analysis unit 2 includes a spatio-temporal behavior perception module 21, a physiological position analysis module 22 and a result judgment module 23;

[0077] Wherein, the spatio-temporal behavior perception module 21 is used to extract the spatio-temporal features of video signals and audio signals, and construct an audio-visual fluid momentum analysis equation to analyze limb violent actions and the intensity of threatening semantics;

[0078] The physiological position analysis module 22 constructs a physiological response perception analysis equation to analyze the individual's physiological response based on the interaction between physiological signals and position signals;

[0079] The result judgment module 23 calculates the Lyapunov exponent based on the intensity of threatening semantics, and comprehensively obtains the analysis result of the bullying event by combining two situations of comparing the concentration of physiological response signals with the threshold of physiological response signal concentration.

[0080] In this embodiment, the audio-visual fluid momentum analysis equation is constructed based on the Navier-Stokes equation; the Navier-Stokes equation is the basic equation describing the motion of viscous fluids. The actions of people in the video can be analogized to the motion of fluid particles, the optical flow field is similar to the fluid velocity field, and the modulation effect of the sound pressure field of the audio signal on the optical flow field of the video signal is similar to the influence of the pressure field on the velocity field in the fluid;

[0081] The actions of people in the video are continuous in time and space, similar to the motion characteristics of fluids; the smoothness of the optical flow field can be described by the viscous term, reflecting the viscosity of limb conflicts; the propagation of the sound pressure field can be described by the pressure term, reflecting the modulation effect of sound on limb actions;

[0082] Identify high-risk actions such as pushing and kicking through the optical flow field and vorticity field;

[0083] Identify threatening voices through the sound pressure field and sound energy density;

[0084] Construct an audio-visual fluid momentum analysis equation to achieve deep coupling of video and audio signals, and improve the accuracy of bullying event detection;

[0085] In this embodiment, the physiological response perception analysis equation is constructed based on the reaction-diffusion equation; the reaction-diffusion equation is used to describe the diffusion and reaction process of the substance concentration in space; the physiological signal can be analogized to the reactant concentration, and the environmental position signal can be analogized to the reaction condition; the interaction between the physiological signal and the environmental position signal can be described by the reaction term and the coupling term;

[0086] The propagation of the physiological signal in space can be described by the diffusion term, reflecting the propagation speed of the individual physiological response in space;

[0087] The influence of the environmental position signal on the physiological signal can be reflected by the coupling term, reflecting the influence of the environmental position on the individual physiological response;

[0088] The triggering effect of the audio signal on the physiological signal can be described by the triggering term, reflecting the influence of the sound on the individual physiological response;

[0089] Quantify the individual's physiological stress state through the physiological response signal concentration; judge whether there is a bullying event through the physiological response signal concentration and the threshold; realize the deep coupling of the physiological signal and the environmental position signal through the physiological response perception analysis equation, and improve the accuracy of bullying event detection.

[0090] In this embodiment, the spatio-temporal behavior perception module 21 is used to extract the spatio-temporal features of the video signal and the audio signal, and construct an audio-visual fluid momentum analysis equation to analyze the limb violence actions and the threat semantic intensity. The specific method steps are as follows:

[0091] S2.1.1. Based on the video frame sequence , calculate the optical flow field and the vorticity field, and combine them into the video spatio-temporal features;

[0092] The video spatio-temporal features include the optical flow field and the vorticity field , specifically as follows:

[0093] ;

[0094] ;

[0095] Among them, is the two-dimensional pixel position coordinate; is the time; is the optical flow field at the position at the time ; is the video frame sequence; is the derivative of the video frame sequence with respect to time; is the spatial gradient of the video frame; is the optical flow vector to be solved; is the regularization coefficient; is the L2 norm; is the vorticity field; is the curl operation of the optical flow field;

[0096] S2.1.2. Based on the sound information , calculate the sound pressure field and the sound pressure density, and combine them into audio spatio-temporal features:

[0097] The audio spatio-temporal features include the sound pressure field and the sound pressure density , specifically as follows:

[0098] ;

[0099] ;

[0100] Among them, is the three-dimensional spatial position coordinate; is the time; is the sound pressure field at the position at time ; is the integration variable; is the Dirac delta function; is the distance between the sound source and the receiving point; is the speed of sound; is the sound pressure attenuation factor; is the sound energy density; is the spatial integration region; is the spatial integration variable;

[0101] S2.1.3. Based on the optical flow field and the sound pressure field, construct an audio-visual fluid momentum analysis equation to analyze limb violent actions and threatening semantic intensities. The audio-visual fluid momentum analysis equation is specifically as follows:

[0102] ;

[0103] ;

[0104] ;

[0105] Among them, is the behavior density; is the video pressure field; is the social viscosity coefficient; is the limb conflict momentum transfer term; is the violence intensity coefficient; is the th student label; is the Student label; For the th student and the th student; Is the audio modulation force; Is the audio modulation coefficient;

[0106] S2.1.4. Based on the vorticity field and the acoustic energy density, perform semantic threat analysis to calculate the threat semantic intensity , specifically as follows:

[0107] ;

[0108] Among them, Is the threat semantic intensity at time ; Is the video vorticity weight coefficient; Is the audio acoustic energy weight coefficient.

[0109] In this embodiment, the physiological position analysis module 22 constructs a physiological response perception analysis equation to analyze the individual's physiological response based on the interaction between the physiological signal and the position signal. The specific method steps are as follows:

[0110] S2.2.1. Obtain the position signal And the physiological signal ;

[0111] S2.2.2. Based on the interaction between the physiological signal and the position signal, use the physiological response perception analysis equation to analyze the individual's physiological response, and calculate the physiological response signal concentration through the physiological response perception analysis equation. Specifically as follows:

[0112] Physiological response perception analysis equation:

[0113] ;

[0114] Among them, Is the physiological response signal concentration at position At time , representing the individual's physiological stress response; Is the physiological signal diffusion coefficient, indicating the diffusion degree of the physiological signal in space and reflecting the propagation speed of the individual's physiological response in space; Is the Laplace operator of the physiological signal concentration, indicating the change of the physiological response in space; Is the enhancement coefficient of the physiological signal, indicating the self-enhancing effect of the physiological response; Is the physiological signal saturation value, representing the maximum value of the individual's physiological response, usually the maximum response level under physiological limit or extreme stress state; is the coupling strength between the physiological signal and the position signal, representing the degree of influence of the position on the individual's physiological response; is the concentration of the environmental position signal; is the audio trigger coefficient, representing the degree of influence of the audio signal on the physiological response; is the threshold of the audio signal; is the Heaviside step function; is the coupling coefficient between the physiological signal and the position signal; is the gradient of the physiological signal concentration; is the influence coefficient of heart rate on the physiological response; is the influence coefficient of skin conductance response on the physiological response; is the bias coefficient; is the change rate of the physiological response signal concentration;

[0115] S2.2.3. Solve the physiological response perception analysis equation to obtain the physiological response signal concentration at position at time and compare it with the physiological response signal concentration threshold:

[0116] If , the individual's physiological response is in danger and there is a bullying incident;

[0117] If , the individual's physiological response is relatively safe and it is necessary to further determine whether there is a bullying incident;

[0118] Among them, is the physiological response signal concentration threshold.

[0119] In this embodiment, the physiological response signal concentration threshold is obtained by experimentally measuring the physiological responses of a large number of individuals in different situations; it may be adjusted according to factors such as the age, gender, and health status of different individuals and is set in the biosensor worn by students.

[0120] In this embodiment, the result judgment module 23 calculates the Lyapunov exponent based on the threat semantic intensity, and combines the two situations of comparing the physiological response signal concentration with the physiological response signal concentration threshold to comprehensively obtain the bullying incident analysis result, specifically as follows:

[0121] S2.3.1. Based on the physiological response signal concentration and the threat semantic intensity, define the state vector, construct the state equation, and calculate the eigenvalues of the Jacobian matrix and the Lyapunov exponent:

[0122] The state vector is: ;

[0123] State equation:

[0124] ;

[0125] Jacobian matrix :

[0126] ;

[0127] Lyapunov exponent:

[0128] ;

[0129] Among them, is the state vector; is the small change of the state vector;

[0130] S2.3.2. If , and the Lyapunov exponent , there is a bullying incident;

[0131] If , and the Lyapunov exponent , there is no bullying incident.

[0132] It also includes a dynamic early warning and response unit 3. The dynamic early warning and response unit 3 uses a risk grading strategy based on the bullying situation analysis result to give real-time early warning and response to bullying incidents;

[0133] In this embodiment, the dynamic early warning and response unit 3 includes a risk grading module 31 and an event response module 32.

[0134] The risk grading module 31 grades the risk of bullying incidents based on the physiological response signal concentration and the threat semantic intensity to generate a grading early warning strategy;

[0135] The event response module 32 executes the grading early warning strategy in real time according to the risk grading result.

[0136] In this embodiment, the grading early warning strategy is as follows:

[0137] Low-risk event: and , push a lightweight early warning message to the head teacher and security personnel;

[0138] Low-risk event: or , send an acoustic deterrence signal to the target area to warn the participants in the bullying incident;

[0139] Low-risk event: and , send an acoustic deterrence signal to the target area and notify the head teacher and security personnel to go to the event address immediately;

[0140] It further includes an event retention record unit 4, and the event retention record unit 4 is used to store multi-modal bullying signals and bullying event analysis result data;

[0141] In this embodiment, the event retention record unit 4 uses a distributed edge cloud data lake to store multi-modal bullying signals and bullying event analysis result data;

[0142] The multi-modal bullying signals include video signals, audio signals, location signals, and physiological signals;

[0143] The bullying event analysis result data includes physiological response signal concentration, threatening semantic intensity, audio-video fluid momentum analysis equation data, and physiological response perception analysis equation data.

[0144] The data lake adopts a distributed edge architecture. The data is not only stored in the cloud but also stored on edge devices such as cameras, which can reduce latency. And on the edge devices, preliminary processing and analysis can be performed first, and then uploaded to the cloud for further analysis and storage;

[0145] The multi-modal bullying signal data is stored in the data lake and maintains a time series relationship, which is conducive to analysis and processing at any time;

[0146] The bullying event analysis results are stored in the data lake in a structured format, which is convenient for subsequent querying and updating;

[0147] The data lake manages the relationships between different types of data. The video signals, audio signals, physiological signals, and location signals are stored and marked as relevant data of the same event, ensuring the consistency and relevance of these multi-modal signals in time and space.

[0148] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A campus anti-bullying system based on a large model, characterized in that: include: A multimodal signal sensing unit (1), wherein the multimodal signal sensing unit (1) uses a sensor network to collect multimodal bullying signals in real time and pre-processes the multimodal bullying signals; A large model intelligent analysis unit (2), wherein the large model intelligent analysis unit (2) extracts the spatiotemporal features of the video signal and the audio signal based on the multimodal bullying signal, constructs an audio-video fluid momentum analysis equation to detect physical violence and semantic threat, and then constructs a physiological response perception analysis equation to analyze individual physiological responses, thereby comprehensively obtaining a bullying event analysis result; A dynamic warning and response unit (3), wherein the dynamic warning and response unit (3) uses a risk classification strategy based on the bullying situation analysis result to provide a real-time warning and response to bullying incidents; An event storage and recording unit (4), the event storage and recording unit (4) being used to store multimodal bullying signals and bullying event analysis result data; The multimodal signal sensing unit (1) comprises a signal acquisition module (11) and a signal processing module (12); Wherein, the spatiotemporal synchronous signal acquisition module (11) uses a sensor network to collect multimodal bullying signals in real time; Multimodal bullying signals include video signals, audio signals, location signals, and physiological signals; The signal processing module (12) is used to pre-process the multimodal bullying signal collected by the signal collection module (11); The large model intelligent analysis unit (2) comprises a spatiotemporal behavior perception module (21), a physiological position analysis module (22) and a result judgment module (23); The spatiotemporal behavior perception module (21) is used to extract the spatiotemporal features of the video signal and the audio signal, and construct an audio-video fluid momentum analysis equation to analyze the physical violence action and the threatening semantic intensity; The physiological position analysis module (22) constructs a physiological response perception analysis equation to analyze individual physiological responses based on the interaction between the physiological signal and the position signal, and calculates the physiological response signal concentration through the physiological response perception analysis equation; The result judgment module (23) calculates the Lyapunov index based on the threatening semantic intensity, combines the two situations of the comparison of the physiological reaction signal concentration and the physiological reaction signal concentration threshold, and comprehensively obtains the bullying event analysis result; The method of calculating the Lyapunov index based on the threatening semantic intensity includes: defining a state vector based on the threatening semantic intensity and the physiological response signal concentration and calculating the Lyapunov index.

2. The campus anti-bullying system based on a large model according to claim 1 is characterized in that: The spatiotemporal synchronous signal acquisition module (11) uses a sensor network to collect multimodal bullying signals in real time, and the specific method is as follows: S1.1.

1. Acquiring video signals using a multispectral camera network ; S1.1.

2. Using microphone array to obtain audio signals ; S1.1.

3. Locate the student's position based on the multi-spectral camera and obtain the position signal ; S1.1.

4. Obtaining physiological signals based on biosensors worn by students ; in, is the heart rate; It is the electrical response of the skin.

3. The campus anti-bullying system based on a large model according to claim 2 is characterized in that: The spatiotemporal behavior perception module (21) is used to extract the spatiotemporal features of the video signal and the audio signal, and construct an audio-video fluid momentum analysis equation to analyze the physical violence action and the threatening semantic intensity. The specific method steps are as follows: S2.1.

1. Based on video signal , calculate the optical flow field and vorticity field, and combine them into video spatiotemporal features. Video spatiotemporal features include optical flow field and vorticity field ; S2.1.

2. Based on audio signal , calculate the sound pressure field and sound pressure density, and combine them into audio time-space features. Audio time-space features include sound pressure field and sound pressure density ; S2.1.

3. Based on the optical flow field and the sound pressure field, an audio-visual fluid momentum analysis equation is constructed to analyze the physical violence action and the threatening semantic intensity. The audio-visual fluid momentum analysis equation is as follows: ； ； ； in, is the behavior density; is the video pressure field; is the social viscosity coefficient; is the momentum transfer term for physical conflict; is the violence intensity coefficient; For the Student ID; For the Student ID; For the Students and The distance between students; For audio modulation power; is the audio modulation coefficient; S2.1.

4. Based on the vorticity field and acoustic pressure density, perform semantic threat analysis and calculation to obtain the threat semantic intensity .

4. The campus anti-bullying system based on a large model according to claim 3 is characterized in that: The physiological position analysis module (22) constructs a physiological response perception analysis equation to analyze individual physiological responses based on the interaction between the physiological signal and the position signal. The specific method steps are as follows: S2.2.

1. Obtaining position signal and physiological signals ; S2.2.

2. Based on the interaction between physiological signals and position signals, the physiological response perception analysis equation is used to analyze the individual physiological response, and the concentration of the physiological response signal is calculated by the physiological response perception analysis equation; S2.2.

3. Solve the physiological response perception analysis equation to obtain Upper position Physiological response signal concentration , compared with the physiological response signal concentration threshold: like , the individual's physiological response is at risk, and there is bullying; like , the individual's physiological response is relatively safe, and further judgment is needed on whether there is a bullying incident; in, is the concentration threshold of the physiological response signal.

5. The campus anti-bullying system based on a large model according to claim 4 is characterized in that: The result judgment module (23) calculates the Lyapunov index based on the threatening semantic intensity, and combines the two situations of comparing the physiological reaction signal concentration with the physiological reaction signal concentration threshold to comprehensively obtain the bullying event analysis results, which are as follows: S2.3.

1. Based on the concentration of physiological response signals and the intensity of threatening semantics, define the state vector, construct the state equation, and calculate the eigenvalues ​​of the Jacobian matrix and the Lyapunov index: The state vector is: ; Equation of state: ； Jacobian matrix : ； Lyapunov exponent : ； in, is the state vector; is a small change in the state vector; S2.3.2 If , and the Lyapunov exponent , there are incidents of bullying; like , and the Lyapunov exponent , there is no bullying incident.

6. The campus anti-bullying system based on a large model according to claim 5 is characterized in that: The dynamic early warning and response unit (3) comprises a risk grading module (31) and an event response module (32).

7. The campus anti-bullying system based on a large model according to claim 6 is characterized in that: The risk grading module (31) performs risk grading on bullying incidents based on the concentration of physiological response signals and the intensity of threatening semantics to generate a graded warning strategy; The event response module (32) executes a graded warning strategy in real time according to the risk grading result.

8. The campus anti-bullying system based on a large model according to claim 7 is characterized in that: The event retention recording unit (4) uses a distributed edge cloud data lake to store multimodal bullying signals and bullying event analysis result data; Multimodal bullying signals include video signals, audio signals, location signals, and physiological signals; The bullying incident analysis results include physiological response signal concentration, threatening semantic intensity, audio and video fluid momentum analysis equation data, and physiological response perception analysis equation data.

Images