Intelligent psychological safety early warning system oriented to education field

Abstract

The invention discloses an intelligent psychological safety early warning system oriented to the education field, and relates to the technical field of psychological safety early warning and edge computing. The system comprises a data acquisition and edge preprocessing module, a six-dimensional personality modeling module, an interpersonal interaction relationship evaluation module, an emotion detection and psychological stress evaluation module and a comprehensive psychological state evaluation and early warning module. The technical key points are as follows: multi-dimensional data fusion: comprehensively evaluating a psychological state from three levels of emotion, behavior and environment; setting dynamic period evaluation: considering short-term fluctuation and long-term trend; a personalized early warning mechanism is displayed; differential management is realized through early warning grade division; edge and cloud collaborative architecture support: ensuring real-time performance, privacy protection and system expansibility; by adopting the mechanism, intelligent psychological safety early warning system operation is realized, and the system is not only suitable for the education field, but also has a wide application prospect in other related fields such as enterprise employee psychological health monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the fields of psychological safety early warning and edge computing technology, and specifically to an intelligent psychological safety early warning system for the education field. Background Art

[0002] Psychological safety early warning refers to the identification and prediction of possible psychological abnormalities by collecting multi-dimensional data such as individual behavior, emotions, and physiology, combined with artificial intelligence and big data analysis technology, and completing early warnings according to the situation to achieve early intervention; existing systems mostly rely on centralized data processing methods, and there are problems such as response delays and privacy leakage risks. Edge computing is usually used in psychological safety assessments. It is a computing architecture that sinks data processing tasks from the cloud to edge devices close to the data source (such as smart cameras and terminal servers); in current psychological safety early warning systems, emotion recognition mostly uses cloud-based model reasoning, which results in a long time for data upload and download; at the same time, the centralized storage of large amounts of sensitive behavioral data also brings security risks. Therefore, more and more studies have begun to explore the introduction of edge computing into psychological safety early warning systems to improve their response speed and deployment security.

[0003] Existing mental health monitoring technologies generally have problems related to single data dimensions, insufficient analysis accuracy and delayed response. For example, traditional methods mostly rely on questionnaires or manual observations, which make it difficult to capture students' emotional fluctuations and behavioral changes in real time, and easily lead to assessment bias. Although some systems have introduced physiological signal analysis, they lack comprehensive consideration of environmental factors and are prone to misjudging short-term emotional reactions as serious psychological problems, resulting in a high false alarm rate. At the same time, when considering physiological signals, it is impossible to achieve non-sensitive safety warning operations. In addition, most systems do not adopt edge computing architecture, and data upload to the cloud for processing brings delays and privacy risks, affecting real-time and scalability. A campus psychological monitoring platform only judges stress status through heart rate and facial expression recognition. Because it does not combine dynamic evaluation of interpersonal interaction matching and historical similarity, false alarms frequently occur, which to a certain extent also interferes with normal teaching order. Therefore, there is an urgent need for an intelligent early warning system that integrates multi-dimensional data, has dynamic periodic evaluation capabilities and supports edge deployment to solve the many problems existing in traditional technologies. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: An intelligent psychological safety early warning system for the education sector, including: The data collection and edge pre-processing module collects behavioral data of individuals in the target environment and performs preliminary processing actions; The six-dimensional personality modeling module builds the SPA model based on the Jungian personality model and the two-dimensional emotional dynamics model. It is used to divide personality into two categories: emotional dynamics and information coping, and each category contains several personality dimensions. The interpersonal interaction assessment module integrates the individual's process based on the individual's personality dimensions to evaluate the individual's interpersonal relationship level in a set scenario. Based on the interpersonal relationship level index, it outputs the integrated interpersonal interaction matching degree. The process includes at least: feelings, views, ideas, practices, and execution; The emotion detection and psychological stress assessment module uses deep learning algorithms to model behavioral data, extract emotional characteristics and perform comparative actions to derive a set of individual psychological state indicators. After normalizing the set of psychological state indicators, it executes a periodic judgment mechanism to output whether to issue a warning instruction; The comprehensive psychological state assessment and early warning module, upon receiving the early warning instruction, fuses the individual's corresponding psychological state indicator set and interpersonal interaction matching degree, outputs a comprehensive psychological state score, and triggers the corresponding level of early warning signal according to the interval threshold.

[0005] Furthermore, behavioral data includes at least: facial expressions, body postures, and voice intonation; The preliminary processing actions are performed on the edge device, and the preliminary processing actions at least include: image cropping and data compression.

[0006] Furthermore, the personality dimensions of emotional dynamics include the opposing directness and flexibility, calmness and passion, and extroversion and introversion; the personality dimensions of information coping include the opposing details and concepts, emotion and mathematics, and deferred judgment and judgment; Among them, the output of the SPA model is: each personality dimension and its corresponding percentage.

[0007] Furthermore, the set scenarios include at least: family scenarios and campus scenarios; the interpersonal relationship indicators correspond to: family relationship compatibility and mentor system matching; When the scenario is set as a family scenario, the process of evaluating family relationship compatibility is as follows: Based on the pre-built rule engine, the status indicators of the five processes of feelings, views, ideas, practices and execution between individuals and parents are judged. The status indicators include: blind spots, misunderstandings and consensus areas, and the output family relationship compatibility is calculated based on the basic scores corresponding to the status indicators; among them, the basic scores corresponding to blind spots, misunderstandings and consensus areas are marked as 0, 0.5 and 1 respectively; when the setting scenario is a campus scenario, the evaluation of the tutor system matching degree is the same as the evaluation of family relationship compatibility.

[0008] Furthermore, the psychological state indicator set includes at least: an aggression value, a stress value, a tension value, and a historical similarity; When using deep learning algorithms to model behavioral data, model structure optimization is also included: introducing an attention mechanism to enable the model to focus on key facial micro-expressions or body postures; adding a time window sliding mechanism to output the attack value, stress value, and tension value of the current emotional state every Q seconds, where the value of Q ranges from 0 to 100.

[0009] Furthermore, each mental state indicator in the mental state indicator set is generated as follows: The aggression value is generated by using a ResNet + LSTM hybrid model to identify anger and hostility in emotional features. Video data of several angry and aggressive behaviors is used as training data. The probability of aggression tendency is output for each frame and the average is taken to obtain the aggression value. The stress value is generated by using body movement sequences, sitting posture changes, and hand movement frequency as emotional features, combined with a 3D-CNN video action recognition model to output the stress value. The tension value is generated by using the OpenPose model to extract body posture, combined with facial muscle tension analysis as emotional features, and support vector machine classification to output the tension value. The historical similarity is generated by using a Siamese network for comparative learning. The corresponding comparative actions are input with the current behavioral characteristics and the behavioral characteristics of individuals with historical psychological problems, and the historical similarity is output.

[0010] Furthermore, the content of the executed periodic judgment mechanism is: set the monitoring period T, record the current individual's daily psychological state indicator set, and issue a warning instruction when the historical similarity exceeds the standard and the frequency of occurrence exceeds the set value; otherwise, no response action will be taken for the time being.

[0011] Furthermore, the data fusion process is as follows: extract the maximum values ​​of the daily attack value, stress value, and tension value within the monitoring period T, combine them with the current historical similarity, and use the pre-derived interpersonal interaction matching degree to make corrections, output the corresponding comprehensive psychological state assessment index for each day once, and then output the average value of each comprehensive psychological state assessment index within the monitoring period T twice to generate a comprehensive psychological state score.

[0012] Furthermore, when the corresponding level of warning signal is triggered according to the interval threshold, the range of the interval threshold is [L, Z*L]; among them, the values ​​of L and Z are both greater than 0; when the comprehensive psychological state score is less than L, a yellow warning is issued; when the comprehensive psychological state score is greater than or equal to L and less than Z*L, an orange warning is issued; when the comprehensive psychological score is greater than or equal to Z*L, a red warning is issued, and the urgency represented by the yellow warning, orange warning and red warning increases in sequence.

[0013] The present invention provides an intelligent psychological safety early warning system for the education field, which has the following beneficial effects: This solution achieves comprehensive monitoring and early warning of students' mental health status through multi-dimensional and multi-scenario data collection and analysis, solving the problems of incomplete data collection and inaccurate analysis in traditional mental health monitoring methods. The system uses edge computing technology to improve the efficiency of data processing and analysis, and reduce system latency and energy consumption. In addition, the system is real-time and non-invasive, and can provide students with comprehensive mental health services without affecting their normal learning and life. The historical similarity adopted in this solution is used in the system. On the one hand, it can serve as the core judgment basis for the generation of early warning signals to ensure that the system can issue early warnings; on the other hand, it can serve as an important parameter for psychological state assessment, helping the system to more accurately reflect the individual's psychological state, improve the scientificity and robustness of the assessment, and enable the system to effectively reduce the false alarm rate, improve the accuracy of early warnings, and enhance personalized service capabilities while ensuring response speed, providing solid technical support for mental health monitoring. This solution uses the interpersonal interaction matching degree assessment module to determine interpersonal interaction matching. This can serve as an early warning indicator for mental health risks and a correction factor to adjust the individual's overall mental state for the day, ensuring the accuracy of the feedback and reducing the false alarm rate to a certain extent. This solves the problem that traditional models judge psychological problems based on emotional fluctuations or historical similarities, which easily misjudge certain short-term emotional reactions as serious psychological problems. By incorporating interpersonal interaction matching, correction processing can be completed, avoiding false alarms caused by abnormal single indicators and achieving more comprehensive risk coverage. This solution design enhances the intelligence level of the system. To sum up, this plan proposes a highly executable mental health assessment and early warning mechanism, the core of which lies in: multi-dimensional data fusion: comprehensive assessment of mental state from three levels: emotion, behavior, and environment; setting up dynamic periodic assessment: taking into account both short-term fluctuations and long-term trends; displaying a personalized early warning mechanism: achieving differentiated management through early warning level division; edge + cloud collaborative architecture support: ensuring real-time, privacy protection and system scalability; by adopting the above mechanism, the operation of an intelligent psychological safety early warning system can be realized, which is not only applicable to the field of education, but also has broad application prospects in other related fields such as corporate employee mental health monitoring, community psychological services, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Schematic diagram of the SPA model in the present invention in actual display state; Figure 2 A schematic diagram of the corresponding scores of the tutor system matching degree in the teacher-student matching relationship of the present invention; Figure 3 This is a schematic diagram of the actual operation state of the comprehensive psychological state assessment and early warning module in the present invention. DETAILED DESCRIPTION

[0015] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] Example: See also Figures 1 to 3 This embodiment provides an intelligent psychological safety early warning system for the education field. This system adopts a hybrid architecture of edge computing + cloud collaboration, and deploys edge devices with AI reasoning capabilities (such as embedded AI boxes, smart cameras) at the front end to collect and process students' facial expressions, voice intonation, body posture and other behavioral data in real time (this is just a general explanation, and the details can be referred to in subsequent embodiments); after preliminary modeling on the edge, key emotional indicators (attack value, stress value, tension value, etc.) are extracted and encrypted and uploaded to the central server, and the cloud completes long-term trend analysis, matching evaluation and comprehensive early warning judgment.

[0017] The psychological safety early warning system consists of several functional modules that run in sequence, including: 1. Data acquisition and edge preprocessing module: Collect behavioral data of individuals in the target environment and perform preliminary processing actions; The target environment refers to classrooms, campuses, or psychological resource rooms that are common in the education field. Smart cameras deployed in the target environment can collect behavioral data in real time. Behavioral data include: facial expressions, body posture (or language), and voice intonation; Facial expressions: A camera captures an individual's facial image in real time, extracting key features (such as eyebrows, eyes, and mouth). Body posture: This records the individual's posture, movements, and gestures. Additionally, traditional solutions collect individual physiological data, including heart rate, blood pressure, and skin conductance, as needed. These data are obtained through wearable devices or sensors and can also reflect the user's emotional state. This solution avoids the use of these data, which require sensory interpretation, ensuring seamless operation of the overall system. The initial processing actions (performed on local edge devices) include: image cropping and data compression; data can be processed locally to avoid the privacy risks of uploading to the cloud, and encrypted upload can also be selected; the local edge device used can be equipped with a lightweight deep learning model for face detection and key point extraction.

[0018] Six-dimensional personality modeling module (SPA modeling): Based on the Jungian personality model and the two-dimensional emotional dynamics model, a six-dimensional personality analysis tool, the SPA model, was constructed. This tool categorizes personality into two main categories: emotional dynamics and information response. Each category contains three dimensions, each with yin and yang tendencies. The former focuses on interpersonal interaction, while the latter focuses on the reception and response to information and objects. Each dimension has an English letter code. By dividing each individual's personality, a "genetic modeling" of human personality is achieved. Reference Figure 1 As shown, we can see that the dimension structure is: Emotional dynamics: Directness (B) ↔ Tact (I), Calmness (C) ↔ Passion (A), and Extraversion (E) ↔ Introversion (I); Information response: Details (S) ↔ Concepts (N), Feelings (F) ↔ Mathematics (T), and Deferral (P) ↔ Judgment (J); The percentage data corresponding to each dimension is obtained by obtaining initial personality data through questionnaires and behavioral observations, and classifying the behavioral data using machine learning algorithms; the machine learning algorithms are SVM or XGBoost; and probabilistic models, such as Bayesian networks, are used to calculate the tendency probability of each dimension and output the percentage. Figure 1 The percentages given are examples; The output content is: each dimension and its corresponding percentage; for example, a student actively speaks in class and is good at expressing his or her opinions. The questionnaire results show that his or her score is 70% E and 65% A, indicating that his or her personality tends to be extroverted and proactive.

[0019] 3. Interpersonal interaction relationship assessment module: Based on the personality dimensions of individuals in the SPA model, the system evaluates the degree of interpersonal relationships in a given scenario by integrating the individual's feelings, views, thoughts, practices, and execution. Based on the interpersonal relationship degree indicators, the integrated interpersonal interaction matching degree is output; The setting scenarios include: family scenarios and campus scenarios, and may also include social scenarios as needed. However, this embodiment targets the field of student education, so the benchmark setting scenarios use family scenarios and campus scenarios; the interpersonal relationship degree indicators correspond to: family relationship compatibility and mentor system matching degree; In the family scenario, a pre-built rule engine is used to determine the status indicators of the relationship between individuals and parents across five stages: feelings, views, ideas, practices, and execution. These indicators include blind spots, misunderstandings, and areas of consensus. These indicators can be derived through questionnaires. The family relationship compatibility (a score between 0 and 1) is calculated based on the corresponding basic scores of the status indicators. Blind spots indicate that there are differences between the individual and the parent in the corresponding process (for example, the individual is introverted in the feeling process, and the parent is extroverted); misconceptions indicate that there are ambiguous differences between the individual and the parent in the corresponding process (for example, the individual is introverted in the feeling process, and the parent does not judge it); consensus areas indicate that there is consensus between the individual and the parent in the corresponding process (for example, the individual is introverted in the feeling process, and the parent is also introverted). Views correspond to directness and flexibility; thoughts correspond to emotions and mathematics; practices correspond to judgment and deferred judgment; At the same time, a basic score is set according to each status indicator: Consensus area (green dot): There is consensus, and the basic score is marked as 1 (can be adjusted as needed, ranging from 1 to 2); Misunderstanding area (yellow dot): There is fuzzy difference, and the basic score is marked as 0.5 (can be adjusted as needed, ranging from 0 to 0.5); Blind area (red dot): There is a difference, and the basic score is marked as 0; The calculation process based on the basic score corresponding to the status indicator is as follows: For the corresponding individual, the number of parents can be more than 1 or several. Corresponding weights are assigned to different processes, and then the weighted calculation is performed to obtain the corresponding family relationship compatibility between the individual and any parent; if the number of parents is 1, the family relationship compatibility is used as the final family relationship compatibility output; if the number of parents exceeds 1, the average of the corresponding family relationship compatibilities is used as the final family relationship compatibility output.

[0020] Effect description: By adopting the above analysis method, it is possible to comprehensively and objectively evaluate the matching degree of family members in various dimensions, and formulate corresponding communication and interaction strategies accordingly, effectively solving the subjective and one-sided problems existing in traditional family relationship assessments, and improving the accuracy and scientific nature of family relationship assessments; at the same time, by real-time monitoring and analysis of the emotional state and behavioral performance of family members, on the one hand, it is possible to timely discover potential family contradictions and conflicts, take effective intervention measures, and promote the harmony and stability of family relationships; on the other hand, it can also provide an effective basis or reference for the risk of individual mental health problems.

[0021] In a campus setting, a preset rule engine is used to determine the status indicators between individuals and teachers in the five processes of feelings, views, ideas, practices, and execution. These indicators include blind spots, misunderstandings, and areas of consensus. These indicators can be obtained through questionnaires. The tutoring system matching degree (a score between 0 and 1) is calculated based on the basic scores corresponding to the status indicators. Similarly, the basic score is set according to each status indicator (refer to Figure 2 shown); Similarly, for a corresponding individual, the number of teachers can be more than one or more. Corresponding weights are assigned to different processes, and then the weighted calculation is used to obtain the corresponding tutor matching degree between the individual and any teacher. If the number of teachers is 1, the tutor matching degree is used as the final tutor matching degree output; if the number of teachers is more than 1, the average of the corresponding tutor matching degrees is used as the final tutor matching degree output. It should be noted that, usually in different scenarios, there is only one parent or teacher. That is, parents usually choose the one who spends more time with the individual as the representative, and teachers usually choose the class teacher as the representative. Based on the final interpersonal relationship degree index, an interpersonal relationship map can be constructed. Refer to the attached Figure 2 , an example is given below: Based on the basic score of each status indicator, weights are assigned to the feeling process, view process, thought process, practice process, and execution process. The value range of each weight is set to 0 to 1, and the result of weighted calculation is: family relationship compatibility = 0.98; it should be noted that during the calculation period, an additional scoring mechanism can be introduced as needed to reward excellent performance in certain processes, which will not be elaborated here; the weight coefficient is determined using the coefficient of variation method, which is a method of assigning weights to each indicator based on the degree of variation between the current value and the target value of each evaluation indicator; if the numerical difference of an indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and therefore the indicator should be given a larger weight; conversely, if the numerical difference of the evaluated objects on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak, and therefore the indicator should be given a smaller weight; this method directly uses the information contained in each indicator to obtain the weight of the indicator through calculation, and therefore is objective.

[0022] The process of synthesizing interpersonal relationship indicators to obtain interpersonal interaction matching is as follows: The family relationship compatibility and mentorship matching are weighted, and then the comprehensive interpersonal interaction matching score is output. This comprehensive matching score can be used as an early warning indicator of mental health risks. Through long-term observation, if the interpersonal interaction matching is always below a certain threshold (such as 0.6), it indicates that the individual may face greater interpersonal relationship pressure, thereby increasing the risk of mental health problems, and timely intervention treatment or prevention is needed.

[0023] Effect description: This system solves the problems of incomplete data collection and inaccurate analysis in traditional mental health monitoring methods. Through multi-dimensional and multi-scenario data collection and analysis, it realizes comprehensive monitoring and early warning of students' mental health status; the system adopts edge computing technology to improve the efficiency of data processing and analysis, and reduce system latency and energy consumption; in addition, the system also has the characteristics of real-time and non-sensitive, and can provide students with comprehensive mental health services without affecting their normal study and life.

[0024] 4. Emotion detection and psychological stress assessment module: A deep learning algorithm is used to model behavioral data, extract emotional features and perform comparative actions, and then derive a set of individual psychological state indicators. After normalizing the set of psychological state indicators, a periodic judgment mechanism is implemented. The psychological state indicator set includes: aggression value, stress value, tension value, and historical similarity. The specific steps of modeling are: modeling based on facial micro-expressions, body movements, etc. in the video stream corresponding to the behavioral data; Model structure optimization: An attention mechanism can be introduced as needed to enable the model to focus on key facial micro-expressions or body movements. At the same time, a time window sliding mechanism is added to output the attack value, stress value, and tension value of the current emotional state every Q seconds. The value of Q ranges from 0 to 100. In this embodiment, the value of Q is specifically 30. Among them, the calculation method of each psychological state indicator in the psychological state indicator set is as follows: Aggression value: The model used is: ResNet + LSTM, which is used to identify emotions such as anger and hostility (as emotional features). The training data includes a large amount or a number of video data of angry and aggressive behaviors. The output is: the probability of attack tendency is output for each frame, and the attack value is averaged. The higher the attack tendency, the stronger the attack tendency; Stress value: Using body movement sequences, sitting posture changes, and hand movement frequency as emotional features, combined with a 3D-CNN video action recognition model, the output is: stress value. Higher stress values ​​indicate greater psychological load; Tension value: The model used is: OpenPose to extract body posture, combined with facial muscle tension analysis (as an emotional feature), and support vector machine classification. The output is: tension value. Higher tension values ​​indicate greater tension; Historical similarity (i.e., similarity with historical psychological problems): The model used is: using Siamese network for comparative learning, corresponding to the comparative actions performed; input current behavioral characteristics and behavioral characteristics of individuals with historical psychological problems, and output historical similarity. The higher the historical similarity, the closer the current state is to individuals with historical psychological crises.

[0025] The purpose of normalization is to map the attack value, pressure value, and tension value to the interval [0, 100]; The execution periodic judgment mechanism is as follows: Set a monitoring period T (where T is greater than 1, and in this embodiment, it is set to 7 days, or one week), and record the current individual's daily psychological state indicator set. When the historical similarity exceeds the standard (for example, exceeds the preset standard threshold of 0.6) and the frequency of occurrence exceeds the set value (for example, 3 times / week), an early warning instruction is issued; otherwise, it is considered a "suspected false alarm" and no response action is taken for the time being; The conditions for issuing an early warning order are as follows: ; In the formula, Warning_conditions represents the warning conditions, sim t represents the historical similarity of day t, t=1, 2, ..., T, θ s represents the preset standard threshold, which is recommended to be set to 0.6; I() represents an exponential function, which returns 1 if the condition is met, otherwise returns 0; N represents the set value, which is recommended to be set to 3; the output based on the above formula is: whether to issue a warning command (True / False); Note: The above formula combines the two dimensions of historical similarity exceeding the standard and frequency of occurrence to avoid false positives caused by a single high value. Only when the historical similarity not only exceeds the preset standard threshold but also occurs a certain frequency within the period is it considered that there is a high psychological risk, thereby ensuring the accuracy and effectiveness of the output results.

[0026] Effect description: This system realizes the intelligent, personalized and real-time mental health management in the field of education by adopting the technical means of edge computing + SPA personality modeling + vibration image emotion detection. It not only solves the subjectivity and lag problems of traditional psychological assessment, but also provides schools and families with scientific interpersonal relationship management tools. It has broad application prospects and social value.

[0027] 5. Comprehensive psychological state assessment and early warning module: Upon receiving the warning instruction, the data of the corresponding psychological state indicator set and interpersonal interaction matching degree of the individual is fused from the historical data to output the comprehensive psychological state score, and the warning signals of different levels are triggered according to the interval threshold. The data fusion process is as follows: Extract the maximum values ​​of the daily aggression, stress, and tension values ​​within the monitoring period T (as representative values), combine them with the current historical similarity, and use the pre-derived interpersonal interaction matching degree to make corrections. Output the corresponding comprehensive psychological state assessment index for each day. Then, output the average value of each comprehensive psychological state assessment index within the monitoring period T twice to generate a comprehensive psychological state score, which is used to judge the severity of psychological problems. Specifically, the maximum attack value at t on that day: Ax t ; The maximum pressure value under t on the day: Px t ; The maximum value of the tension value at t on the day: Tx t ; Among them, Ax t 、Px t 、Tx t ∈[0, 100], indicating the maximum intensity of emotional fluctuations on that day; Historical similarity: sim t ∈[0, 1], which means the degree of similarity between the individual's behavior pattern on day t and that of people with known psychological problems. The closer it is to 1, the more likely it is to be in a state of psychological risk, and judgment and attention will be made based on the above preset standard threshold; Introducing interpersonal interaction matching Sr as a correction factor to construct a daily comprehensive psychological state assessment index: ; Where, Pgwell t It represents the comprehensive psychological state evaluation index on day t. w1 to w5 are weight coefficients, and the value range is [0, 1]. The result is the result of one output: Pgwell t The calculation basis of comprehensive psychological state score is: Epg_avg represents the average value within the monitoring period T, that is, the comprehensive psychological state score, which is the secondary output result: Epg_avg; Key Points: The above approach uses multi-dimensional data fusion, including emotional indicators (aggression, stress, tension), historical similarity, and interpersonal interaction matching. Each dimension reflects the psychological state from a different perspective, improving the comprehensiveness of the assessment. Daily dynamic assessment + periodic summary is used, and the final Epg_avg is used to obtain the overall psychological trend within the cycle, avoiding interference from occasional events. Long-term interpersonal relationship adaptation is also taken into consideration to reflect the impact of "environmental factors" on mental health. The overall design is applicable to different populations and scenarios, and ensures the rationality and effectiveness of the results. Effect description: The historical similarity used in the above scheme has a dual role in the intelligent psychological safety early warning system. On the one hand, it serves as the core judgment basis for the generation of early warning signals, ensuring that the system only issues an early warning when the individual behavior pattern deviates significantly from the normal group; on the other hand, it serves as an important parameter for psychological state assessment, helping the system to more accurately reflect the individual's psychological state and improve the scientificity and robustness of the assessment. Through the design of this dual composite mechanism, the system can effectively reduce the false alarm rate, improve the accuracy of early warning, and enhance personalized service capabilities while ensuring response speed, providing solid technical support for mental health monitoring.

[0028] The warning level classification standards are shown in the following table: Table 1 (Example): Reflects the meaning of the corresponding warning levels when the interval threshold is triggered:

[0029] From the example in the table above, it can be seen that the range of the interval threshold in this embodiment is [L, Z*L]; where the values ​​of L and Z are both greater than 0 and can be set according to actual needs or historical data. No further details will be given here. In this embodiment, the value of L is 30 and Z is 2, so the specific range is 30 to 60, and the corresponding endpoint values ​​are 30 and 60; The above clarifies the three-level warning standards to facilitate intervention decisions and resource allocation; Effect Description: The interpersonal interaction matching degree obtained in the interpersonal interaction matching degree assessment module can be used as an early warning indicator of mental health risks. On the other hand, it can also be used as a correction factor to feedback and adjust the individual's comprehensive mental state on the day, ensuring the accuracy of the result feedback and reducing the false alarm rate to a certain extent. The specific performance is as follows: Relying solely on emotional data and historical similarity to judge psychological states is susceptible to occasional mood swings or short-term environmental changes. Long-term interpersonal compatibility reflects the quality of an individual's social support system. If interpersonal compatibility is poor (i.e., low Sr), even if current emotions are stable, there may be potential psychological vulnerability due to lack of emotional support. Introducing interpersonal interaction matching as a correction factor can help improve the stability and predictive power of assessments. Traditional models that judge psychological issues solely based on mood swings or historical similarities can easily misjudge brief emotional reactions as serious psychological issues. By incorporating interpersonal interaction matching, these "isolated events" can be corrected, avoiding false positives caused by abnormal single indicators. It can also identify "silent" risk groups with low emotional indicators but poor interpersonal adaptation, achieving more comprehensive risk coverage. In summary, by incorporating interpersonal interaction matching into the comprehensive psychological state assessment system, this solution can significantly improve the scientificity, accuracy, and stability of the assessment while ensuring real-time responsiveness. This design not only enhances the intelligence level of the system, but also provides reliable data support for subsequent personalized intervention strategies.

[0030] This solution proposes a highly executable mental health assessment and early warning mechanism, the core of which lies in: multi-dimensional data fusion: comprehensive assessment of mental state from three levels: emotion, behavior, and environment; setting up dynamic periodic assessment: taking into account both short-term fluctuations and long-term trends; displaying a personalized early warning mechanism: achieving differentiated management through early warning level division; edge + cloud collaborative architecture support: ensuring real-time, privacy protection and system scalability; by adopting the above mechanism, an intelligent psychological safety early warning system can be realized, which is not only suitable for the field of education (for example: primary and secondary schools), but also has broad application prospects in the fields of corporate employee mental health monitoring, community psychological services, etc.

[0031] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0032] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0033] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. An intelligent psychological safety early warning system for the education sector, including: The data collection and edge pre-processing module collects behavioral data of individuals in the target environment and performs preliminary processing actions; The six-dimensional personality modeling module builds the SPA model based on the Jungian personality model and the two-dimensional emotional dynamics model. It is used to divide personality into two categories: emotional dynamics and information coping, and each category contains several personality dimensions. Its characteristics are: It also includes an interpersonal interaction relationship assessment module, which integrates the individual's process based on the individual's personality dimensions to evaluate the individual's interpersonal relationship level indicators in a set scenario, and outputs the integrated interpersonal interaction matching degree based on the interpersonal relationship level indicators; the process includes at least: feelings, views, ideas, practices and execution; The emotion detection and psychological stress assessment module uses deep learning algorithms to model behavioral data, extract emotional characteristics and perform comparative actions to derive a set of individual psychological state indicators. After normalizing the set of psychological state indicators, it executes a periodic judgment mechanism to output whether to issue a warning instruction; The comprehensive psychological state assessment and early warning module, upon receiving the early warning instruction, fuses the individual's corresponding psychological state indicator set and interpersonal interaction matching degree, outputs a comprehensive psychological state score, and triggers the corresponding level of early warning signal according to the interval threshold.

2. The intelligent psychological safety early warning system for the education field according to claim 1 is characterized by: Behavioral data includes at least: facial expressions, body posture, and voice intonation; The preliminary processing actions are performed on the edge device, and the preliminary processing actions at least include: image cropping and data compression.

3. The intelligent psychological safety early warning system for the education field according to claim 1 is characterized by: The personality dimensions of emotional dynamics are: directness and flexibility, calmness and passion, and extroversion and introversion. The personality dimensions of information coping are: details and concepts, emotion and mathematics, and deferred judgment and judgement. Among them, the output of the SPA model is: each personality dimension and its corresponding percentage.

4. The intelligent psychological safety early warning system for the education field according to claim 1 is characterized by: The setting scenarios include at least: family scenarios and campus scenarios; the interpersonal relationship indicators correspond to: family relationship compatibility and mentor system matching; When the scenario is set as a family scenario, the process of evaluating family relationship compatibility is as follows: Based on the pre-built rule engine, the status indicators of the five processes of feelings, views, ideas, practices and execution between individuals and parents are judged. The status indicators include: blind spots, misunderstandings and consensus areas, and the output family relationship compatibility is calculated based on the basic scores corresponding to the status indicators; among them, the basic scores corresponding to blind spots, misunderstandings and consensus areas are marked as 0, 0.5 and 1 respectively; when the setting scenario is a campus scenario, the evaluation of the tutor system matching degree is the same as the evaluation of family relationship compatibility.

5. The intelligent psychological safety early warning system for the education field according to claim 2 is characterized by: The psychological state indicator set includes at least: aggression value, stress value, tension value and historical similarity; When using deep learning algorithms to model behavioral data, model structure optimization is also included: introducing an attention mechanism to enable the model to focus on key facial micro-expressions or body postures; adding a time window sliding mechanism to output the attack value, stress value, and tension value of the current emotional state every Q seconds, where the value of Q ranges from 0 to 100.

6. The intelligent psychological safety early warning system for the education field according to claim 5 is characterized by: The generation method of each psychological state indicator in the psychological state indicator set is as follows: The aggression value is generated by using a ResNet + LSTM hybrid model to identify anger and hostility in emotional features. Video data of several angry and aggressive behaviors is used as training data. The probability of aggression tendency is output for each frame and the average is taken to obtain the aggression value. The stress value is generated by using body movement sequences, sitting posture changes, and hand movement frequency as emotional features, combined with a 3D-CNN video action recognition model to output the stress value. The tension value is generated by using the OpenPose model to extract body posture, combined with facial muscle tension analysis as emotional features, and support vector machine classification to output the tension value. The historical similarity is generated by using a Siamese network for comparative learning. The corresponding comparative actions are input with the current behavioral characteristics and the behavioral characteristics of individuals with historical psychological problems, and the historical similarity is output.

7. The intelligent psychological safety early warning system for education according to claim 1 is characterized by: The content of the executed periodic judgment mechanism is: set the monitoring period T, record the current individual's daily psychological state indicator set, and issue a warning instruction when the historical similarity exceeds the standard and the frequency of occurrence exceeds the set value; otherwise, no response action will be taken for the time being.

8. The intelligent psychological safety early warning system for the education field according to claim 6 is characterized by: The data fusion process is as follows: extract the maximum values ​​of the daily attack value, stress value, and tension value within the monitoring period T, combine them with the current historical similarity, and use the pre-derived interpersonal interaction matching degree to make corrections, output the corresponding comprehensive psychological state assessment index for each day, and then output the average value of each comprehensive psychological state assessment index within the monitoring period T twice to generate a comprehensive psychological state score.

9. The intelligent psychological safety early warning system for the education field according to claim 1 is characterized by: When the corresponding level of warning signal is triggered according to the interval threshold, the range of the interval threshold is [L, Z*L]; among them, the values ​​of L and Z are both greater than 0; when the comprehensive psychological state score is less than L, a yellow warning is issued; when the comprehensive psychological state score is greater than or equal to L and less than Z*L, an orange warning is issued; when the comprehensive psychological score is greater than or equal to Z*L, a red warning is issued, and the urgency represented by the yellow warning, orange warning and red warning increases in sequence.

Images