CN-121709274-B - Psychological early warning method, device, storage medium and equipment for education-oriented non-semantic handwriting

CN121709274BCN 121709274 BCN121709274 BCN 121709274BCN-121709274-B

Abstract

The invention provides a psychological early warning method, device, storage medium and equipment for non-semantic handwriting oriented to education, which comprise the steps of obtaining a paper handwriting image related to the education task, inputting the paper handwriting image into a handwriting feature extraction model to obtain handwriting feature vectors of a subject, comparing the handwriting feature vectors of the subject with handwriting feature datum lines of the education task of the same type of the subject to obtain handwriting feature offset, inputting academic performance fluctuation, attendance record and campus behaviors of the subject into a multi-layer perceptron to obtain behavior embedded vectors with consistent dimensions, converting the aligned handwriting feature offset and behavior embedded vectors into standard structured inference tokens, respectively transmitting the standard structured inference tokens to a handwriting analysis agency and a campus behavior perception agency to obtain physiological pressure and a social function state of the subject, and transmitting the physiological pressure and the social function state to a multi-mode coordination agency to generate suggestions of a quantized risk report and an education scene.

Inventors

ZHANG QIANMENG
Zhu Yichong
Yang Shirao
WANG YUYANG

Assignees

香港科技大学(广州)

Dates

Publication Date: 20260512
Application Date: 20260224

Claims (8)

1. The psychological early warning method for the education-oriented non-semantic handwriting is characterized by comprising the following steps of: acquiring a paper handwriting image related to an educational task; inputting the paper handwriting image into a handwriting feature extraction model to obtain a handwriting feature vector of a subject, wherein the handwriting feature vector comprises the following specific steps: the CNN layer captures the second-order gradient of the handwriting edge by using the local receptive field to obtain the micro handwriting characteristics; the local self-attention layer calculates the spatial correlation between the image blocks by adopting a moving window mechanism to obtain mesoscopic handwriting characteristics; The global feature pyramid adopts an FPN structure to conduct downsampling aggregation on the bottom features to obtain macroscopic handwriting features; integrating the micro handwriting features, the mesoscopic handwriting features and the macro handwriting features to obtain handwriting feature vectors; Comparing the handwriting feature vector of the subject with a handwriting feature reference line of the same type of educational task of the subject to obtain a handwriting feature offset; Inputting academic performance fluctuation, attendance records and campus behaviors of the subjects to a multi-layer perceptron to obtain behavior embedding vectors consistent with handwriting feature vector dimensions; Converting the aligned handwriting characteristic offset and the behavior embedding vector into a standard structured reasoning token; The standard structured reasoning tokens are respectively sent to a handwriting analysis agent and a campus behavior perception agent to obtain physiological pressure and social function states of the subjects; the physiological pressure and the social function state of the subject are sent to a multi-modal coordination agent to generate suggestions for quantifying risk reports and educational scenes, specifically: inputting physiological stress and social function states into psychological analysis results after multi-modal coordination agency processing, wherein the psychological analysis results comprise stress risks, depression risks, anxiety risks and overall risks of a subject; Pushing positive guidance feedback to the subject and sending advice of routine observations to the monitoring end when the risk in all dimensions is below the lowest intervention threshold; When the risk of any dimension exceeds the minimum intervention threshold and is lower than the high risk threshold, highlighting the abnormal deviation point of the handwriting characteristic by combining the handwriting characteristic thermodynamic diagram of the subject, and sending a suggestion for one-to-one care conversation to the teacher end; when the risk of any dimension exceeds the high risk threshold, a specific psychological health early warning report is generated by combining the psychological analysis file of the subject, and a transfer professional psychological institution is sent to the monitoring end to carry out continuous psychological state tracking suggestion.
2. The method of claim 1, wherein the paper handwriting image includes a subject's paper job, classroom handwriting, and test answer sheet.
3. A psychological pre-warning method for education-oriented non-semantic handwriting according to claim 1 and also comprising, prior to inputting a paper-based handwriting image into said handwriting feature extraction model: Dividing the paper handwriting image into a plurality of image blocks with fixed sizes; each image block is converted into a one-dimensional vector with position codes through linear mapping; and inputting the one-dimensional vector to a handwriting feature extraction model.
4. A psychological pre-warning method for education-oriented non-semantic handwriting according to claim 3 and also comprising, after dividing said paper-based handwriting image into a plurality of image blocks of fixed size: obtaining the directional gradient of each handwriting pixel point in each image block to obtain a directional gradient histogram; If the directional gradients of the handwriting pixel points in the image block are uniformly distributed in all directions, setting an attention mask of a first weight for the handwriting pixel points; If the directional gradient of each handwriting pixel point in the image block has consistent directional guidance and motion track, setting an attention mask of a second weight for the handwriting pixel point; Extracting and corresponding attention masks according to the direction of the pixel points to obtain contribution values of the pixel points; and if the contribution value of the pixel point is smaller than a preset threshold value, setting the contribution value of the pixel point to zero.
5. The psychological pre-warning method for education-oriented non-semantic handwriting according to claim 1, wherein the psychological pre-warning method for education-oriented non-semantic handwriting is deployed in a cloud, the cloud is in communication connection with each edge, and after each edge collects a paper surface handwriting image, the psychological pre-warning method further comprises: the edge end inputs the collected paper handwriting image of the subject to a salinized encryption model to obtain preliminary encryption data; the edge end inputs the primary encrypted data into a hash encryption model to obtain an encrypted hash value; And the edge end sends the encrypted hash value to the cloud end.
6. A psychological early warning device for education-oriented non-semantic handwriting, comprising: The image acquisition module is used for acquiring a paper handwriting image related to an educational task; the handwriting feature extraction module is used for inputting the paper surface handwriting image into a handwriting feature extraction model to obtain a handwriting feature vector of a subject, and specifically comprises the following steps: the CNN layer captures the second-order gradient of the handwriting edge by using the local receptive field to obtain the micro handwriting characteristics; the local self-attention layer calculates the spatial correlation between the image blocks by adopting a moving window mechanism to obtain mesoscopic handwriting characteristics; The global feature pyramid adopts an FPN structure to conduct downsampling aggregation on the bottom features to obtain macroscopic handwriting features; integrating the micro handwriting features, the mesoscopic handwriting features and the macro handwriting features to obtain handwriting feature vectors; The handwriting deviation module is used for comparing the handwriting feature vector of the subject with the handwriting feature reference line of the educational task of the same type of subject to obtain handwriting feature deviation; The objective behavior acquisition module is used for inputting the academic performance fluctuation, the attendance record and the campus behavior of the subject to the multi-layer perceptron to obtain a behavior embedding vector consistent with the handwriting feature vector dimension; the feature standardization module is used for converting the aligned handwriting feature offset and the behavior embedding vector into a standard structured reasoning token; the agent distribution module is used for respectively sending the standard structured reasoning tokens to the handwriting analysis agent and the campus behavior perception agent to obtain the physiological pressure and the social function state of the subject; The risk early warning module is used for sending the physiological pressure and the social function state of the subject to the multi-mode coordination agent to generate suggestions of quantized risk reports and education scenes, and specifically comprises the following steps: inputting physiological stress and social function states into psychological analysis results after multi-modal coordination agency processing, wherein the psychological analysis results comprise stress risks, depression risks, anxiety risks and overall risks of a subject; Pushing positive guidance feedback to the subject and sending advice of routine observations to the monitoring end when the risk in all dimensions is below the lowest intervention threshold; When the risk of any dimension exceeds the minimum intervention threshold and is lower than the high risk threshold, highlighting the abnormal deviation point of the handwriting characteristic by combining the handwriting characteristic thermodynamic diagram of the subject, and sending a suggestion for one-to-one care conversation to the teacher end; when the risk of any dimension exceeds the high risk threshold, a specific psychological health early warning report is generated by combining the psychological analysis file of the subject, and a transfer professional psychological institution is sent to the monitoring end to carry out continuous psychological state tracking suggestion.
7. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor performs the steps of the educational non-semantic handwriting oriented psychological pre-warning method according to any one of claims 1 to 5.
8. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor, when executing the computer program, performs the psychological pre-warning method of education-oriented non-semantic handwriting of any one of claims 1-5.

Description

Psychological early warning method, device, storage medium and equipment for education-oriented non-semantic handwriting Technical Field The invention relates to the field of handwriting analysis, in particular to a psychological early warning method, device, storage medium and equipment for education-oriented non-semantic handwriting. Background In the existing mental health screening system, self-reporting tools such as standardized questionnaires, structured interviews and the like still take absolute predominance, and the inherent defect of the mode has become a key bottleneck for limiting screening quality and coverage. From the subjective aspect, the results of the tool completely depend on subjective expression and cognitive judgment of a subject, different individuals have obvious differences in perception thresholds of emotion and pressure, the same psychological state can present different test results due to different personal tolerability, and in addition, partial subjects have concerns about psychological health problems, negative emotion such as depression and anxiety is deliberately hidden, so that the accuracy of screening results is seriously influenced. With the continuous development of image recognition technology, research has been attempted to analyze mental health by using subject handwriting, but the conventional text handwriting analysis must rely on OCR to recognize text content, if the subject writing content is irrelevant to emotion, such as a mechanical transcription formula, the mental health cannot be fed back through user text handwriting, and the conventional recognition mode depends on special hardware such as an electronic handwriting board to acquire user writing pressure and speed information, so that the conventional text handwriting analysis cannot be compatible with widely existing paper surface operation scenes. In addition, the existing mental health screening is mostly displayed in a score evaluation mode, and a mental health screening-management closed loop cannot be formed due to the lack of mental health problem grading intervention guidance aiming at specific scenes. Disclosure of Invention Based on the method, the device, the storage medium and the equipment for psychological pre-warning of the non-semantic handwriting for education are provided, constraint of semantic recognition of handwriting content in the existing psychological health screening process is eliminated, no matter whether a formula and a symbol written by a subject are set, nonsensical graffiti can be used for assessing psychological health, interference of the writing content of the subject is reduced, the assessment of the handwriting characteristics of the subject replaces uniform standards of groups in the prior art through personal benchmark deviation analysis corresponding to the attribute of an educational task, misjudgment rate can be greatly reduced, closed-loop management of educational scenes is achieved, specific physiological pressure and social function states are coupled through a multi-mode coordination agent, and a quantized risk report and decision advice related to the teaching scenes are generated. In a first aspect, the present invention provides a psychological pre-warning method for education-oriented non-semantic handwriting, including: acquiring a paper handwriting image related to an educational task; inputting the paper handwriting image into a handwriting feature extraction model to obtain a handwriting feature vector of a subject; Comparing the handwriting feature vector of the subject with a handwriting feature reference line of the same type of educational task of the subject to obtain a handwriting feature offset; Inputting academic performance fluctuation, attendance records and campus behaviors of the subjects to a multi-layer perceptron to obtain behavior embedding vectors consistent with handwriting feature vector dimensions; Converting the aligned handwriting characteristic offset and the behavior embedding vector into a standard structured reasoning token; The standardized structured reasoning tokens are respectively sent to a handwriting analysis agent and a campus behavior perception agent to obtain physiological pressure and social function states of the subjects; And sending the physiological pressure and the social function state of the subject to a multi-modal coordination agent to generate suggestions for quantifying risk reports and educational scenes. Further, the paper handwriting image comprises paper work, classroom handwriting and test answer sheets of the testee. Further, the step of inputting the paper handwriting image to a handwriting feature extraction model to obtain a handwriting feature vector of the subject, specifically comprises the following steps: the CNN layer captures the second-order gradient of the handwriting edge by using the local receptive field to obtain the micro handwriting characteristics; the local self-attention lay