Education-oriented non-semantic handwriting psychological warning method and device, storage medium and equipment

Abstract

The application provides an education-oriented non-semantic handwriting psychological warning method and device, a storage medium and equipment, which comprises the following steps: obtaining a paper handwriting image related to an education 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 baseline of the subject in the same type of education task to obtain a handwriting feature offset; inputting academic record fluctuations, attendance records and campus behaviors of the subject into a multilayer perception machine to obtain a dimension-consistent behavior embedding vector; converting the aligned handwriting feature offset and the behavior embedding vector into standard structured reasoning tokens, and respectively sending them to a handwriting analysis agent and a campus behavior perception agent to obtain physiological stress and social function status of the subject; and sending the physiological stress and the social function status to a multi-modal coordination agent to generate a quantitative risk report and suggestions for an education scene.

Description

Technical Field

[0001] This invention relates to the field of handwriting analysis, and more specifically to a psychological early warning method, device, storage medium, and equipment for non-semantic handwriting analysis aimed at education. Background Technology

[0002] In existing mental health screening systems, self-report tools, such as standardized questionnaires and structured interviews, still dominate. The inherent flaws of this model have become a key bottleneck restricting the quality and coverage of screening. From a subjective perspective, the results of these tools rely entirely on the subject's subjective expression and cognitive judgment. Different individuals have significantly different thresholds for perceiving emotions and stress; the same mental state may produce different test results due to varying individual tolerance. Furthermore, some subjects, concerned about mental health issues, deliberately conceal negative emotions such as depression and anxiety, severely affecting the accuracy of screening results.

[0003] With the continuous development of image recognition technology, some studies have attempted to analyze mental health using participants' handwriting. However, current handwriting analysis relies on OCR to recognize text content. If the content written by the participant is unrelated to their emotions, such as mechanically copying formulas, it is impossible to obtain feedback on mental health through the user's handwriting. Furthermore, existing recognition methods mostly rely on special hardware such as electronic writing tablets to obtain information on the user's writing pressure and speed, which is incompatible with the widespread paper-based work scenarios. In addition, current mental health screenings mostly present assessment scores, lacking tiered intervention guidance for specific mental health problems in specific scenarios, and failing to form a closed loop of mental health screening and management. Summary of the Invention

[0004] Based on this, the present invention provides a non-semantic handwriting psychological early warning method, device, storage medium, and equipment for education, which breaks away from the constraints of semantic recognition of handwriting content in existing mental health screening processes. Whether it is formulas, symbols, or meaningless scribbles written by the subject, they can all be used for mental health assessment, reducing interference from the subject's writing content. The assessment of the subject's handwriting characteristics also replaces the group-wide standard in existing technologies with individual benchmark offset analysis corresponding to the attributes of the educational task, which can greatly reduce the misjudgment rate and achieve closed-loop management of the educational scenario. A multimodal coordination agent is used to couple specific physiological stress and social functional states to generate quantitative risk reports and decision-making suggestions related to the teaching scenario, which has higher reference value than existing technologies that only provide psychological analysis conclusions.

[0005] In a first aspect, the present invention provides a non-semantic handwriting psychological early warning method for education, comprising:

[0006] Acquire images of handwriting on paper related to educational tasks;

[0007] The paper handwriting image is input into the handwriting feature extraction model to obtain the subject's handwriting feature vector;

[0008] The handwriting feature vector of the subject is compared with the handwriting feature baseline of the subject in the same type of educational task to obtain the handwriting feature offset.

[0009] The subjects' academic performance fluctuations, attendance records, and campus behavior are input into a multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector.

[0010] The aligned handwriting feature offsets and behavior embedding vectors are converted into standard structured inference tokens;

[0011] The standardized structured reasoning tokens are sent to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social functioning status.

[0012] The subjects' physiological stress and social functioning status are sent to a multimodal coordination agent to generate a quantitative risk report and recommendations for educational scenarios.

[0013] Furthermore, the written handwriting images include the subject's written assignments, classroom handwriting, and test answers.

[0014] Furthermore, the step of inputting the paper handwriting image into the handwriting feature extraction model to obtain the subject's handwriting feature vector specifically involves:

[0015] The CNN layer uses local receptive fields to capture the second-order gradient of the handwriting edges, thus obtaining micro-handwriting features;

[0016] The local self-attention layer uses a moving window mechanism to calculate the spatial correlation between image patches and obtain mesoscopic handwriting features;

[0017] The global feature pyramid uses an FPN structure to downsample and aggregate the low-level features to obtain macro-handwriting features;

[0018] The microscopic, mesoscopic, and macroscopic handwriting features are integrated to obtain a handwriting feature vector.

[0019] Furthermore, before inputting the paper handwriting image into the aforementioned handwriting feature extraction model, the following steps are also included:

[0020] The paper handwriting image is divided into several image blocks of fixed size;

[0021] Each image patch is converted into a one-dimensional vector with positional encoding through a linear mapping;

[0022] The one-dimensional vector is input into the handwriting feature extraction model.

[0023] Furthermore, after dividing the paper handwriting image into several fixed-size image blocks, the method further includes:

[0024] Obtain the directional gradient of each handwriting pixel in each image block to obtain the directional gradient histogram;

[0025] If the directional gradient of each handwriting pixel in the image block is uniformly distributed in all directions, an attention mask with a first weight is set for the handwriting pixel.

[0026] If the directional gradients of each handwriting pixel in the image block have consistent directional guidance and motion trajectory, then a second-weighted attention mask is set for the handwriting pixel.

[0027] The contribution value of the pixel is obtained by extracting the direction of the pixel and the corresponding attention mask;

[0028] If the contribution value of the pixel is less than a preset threshold, the contribution value of the pixel is set to zero.

[0029] Furthermore, the step of sending the subject's physiological stress and social functioning status to a multimodal coordination agent to generate a quantitative risk report and suggestions for educational scenarios specifically involves:

[0030] Physiological stress and social functioning status are input into the psychoanalysis results after multimodal coordinated agent processing, including the subject's stress risk, depression risk, anxiety risk, and overall risk;

[0031] When the risk in all dimensions is below the minimum intervention threshold, positive guidance feedback is sent to the subjects, and suggestions for routine observation are sent to the monitoring end.

[0032] When the risk in any dimension exceeds the minimum intervention threshold but is below the high-risk threshold, the abnormal deviation points of handwriting features are highlighted in the heat map of the subject's handwriting features, and suggestions for one-on-one care conversations are sent to the teacher.

[0033] When the risk in any dimension exceeds the high-risk threshold, a specific mental health early warning report is generated by combining the subject's psychological analysis file, and a recommendation to refer the subject to a professional psychological institution for continuous psychological status monitoring is sent to the monitoring end.

[0034] Furthermore, the educational non-semantic handwriting psychological early warning method is deployed in the cloud, and the cloud is communicatively connected to various edge devices. After each edge device acquires a paper handwriting image, it further includes:

[0035] The edge inputs the collected paper handwriting images of the subjects into the salted encryption model to obtain preliminary encrypted data;

[0036] The edge device inputs the preliminary encrypted data into the hash encryption model to obtain the encrypted hash value;

[0037] The edge device sends the encrypted hash value to the cloud.

[0038] Secondly, the present invention also provides a non-semantic handwriting psychological early warning device for education, comprising:

[0039] The image acquisition module is used to acquire images of handwriting on paper related to educational tasks.

[0040] The handwriting feature extraction module is used to input the paper handwriting image into the handwriting feature extraction model to obtain the subject's handwriting feature vector;

[0041] The handwriting offset module is used to compare the subject's handwriting feature vector with the subject's handwriting feature baseline in the same type of educational task to obtain the handwriting feature offset amount.

[0042] The objective behavior acquisition module is used to input the subject's academic performance fluctuations, attendance records, and campus behavior into the multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector.

[0043] The feature standardization module is used to convert aligned handwriting feature offsets and behavior embedding vectors into standard structured inference tokens;

[0044] The agent distribution module is used to send the standardized structured reasoning tokens to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social function status.

[0045] The risk warning module is used to send the subject's physiological stress and social functioning status to the multimodal coordination agent to generate a quantitative risk report and suggestions for educational scenarios.

[0046] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the educational non-semantic handwriting psychological warning method described in any of the first aspects.

[0047] Fourthly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the psychological early warning method for non-semantic handwriting for education as described in any of the first aspects.

[0048] The beneficial effects of adopting the above technical solution are as follows: The non-semantic handwriting psychological early warning method for education in this embodiment breaks away from the constraints of semantic recognition of handwriting content in the existing mental health screening process. Whether it is the formulas or meaningless doodles with symbols written by the subject, they can all be used for mental health assessment, reducing the interference of the subject's writing content; the assessment of the subject's handwriting characteristics also replaces the group uniform standard in the existing technology with personal benchmark offset analysis corresponding to the attributes of the educational task, which can greatly reduce the misjudgment rate and realize closed-loop management in the educational scenario; the use of a multimodal coordination agent to couple specific physiological stress and social functional state generates quantitative risk reports and decision suggestions related to the teaching scenario, which has higher reference significance than the existing technology that only provides psychological analysis conclusions. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0050] Figure 1 This is a schematic diagram of a non-semantic handwriting psychological early warning method for education, based on an embodiment of this application.

[0051] Figure 2 This is a schematic diagram of a non-semantic handwriting psychological early warning device for education, according to one embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. To describe the present invention in more detail, the non-semantic handwriting psychological early warning method, apparatus, storage medium, and device for education provided by the present invention will be specifically described below with reference to the accompanying drawings.

[0053] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an," "a," or "the" do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. The terms "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0054] In existing mental health screening systems, self-report tools, such as standardized questionnaires and structured interviews, still dominate. The inherent flaws of this model have become a key bottleneck restricting the quality and coverage of screening. From a subjective perspective, the results of these tools rely entirely on the subject's subjective expression and cognitive judgment. Different individuals have significantly different thresholds for perceiving emotions and stress; the same mental state may produce different test results due to varying individual tolerance. Furthermore, some subjects, concerned about mental health issues, deliberately conceal negative emotions such as depression and anxiety, severely affecting the accuracy of screening results.

[0055] With the continuous development of image recognition technology, some studies have attempted to analyze mental health using participants' handwriting. However, current handwriting analysis relies on OCR to recognize text content. If the content written by the participant is unrelated to their emotions, such as mechanically copying formulas, it is impossible to obtain feedback on mental health through the user's handwriting. Furthermore, existing recognition methods mostly rely on special hardware such as electronic writing tablets to obtain information on the user's writing pressure and speed, which is incompatible with the widespread paper-based work scenarios. In addition, current mental health screenings mostly present assessment scores, lacking tiered intervention guidance for specific mental health problems in specific scenarios, and failing to form a closed loop of mental health screening and management.

[0056] Based on this, this invention proposes a non-semantic handwriting psychological early warning method for education. This method is applied to a terminal device, which includes, but is not limited to, smartphones and computer devices. The computer device can be at least one of a desktop computer, portable computer, laptop computer, mainframe computer, tablet computer, etc. The device control terminal assesses the subject's anxiety state based on the received handwriting image, combined with fluctuations in the subject's academic performance, attendance records, and campus behavior. (See attached...) Figure 1 The diagram shown illustrates a psychological early warning method for non-semantic handwriting in education, and the specific implementation steps of an embodiment of this method are explained.

[0057] Step S100: Obtain paper handwriting images related to the educational task.

[0058] Unlike existing technologies that require additional electronic writing tablets or dedicated handwriting input hardware, this embodiment only requires conventional scanning or photography to obtain images of the subject's handwriting on paper, greatly reducing the application threshold and deployment cost of this embodiment.

[0059] Considering that the psychological early warning scheme in this embodiment is mainly aimed at educational scenarios, in order to avoid causing the subjects to feel disgusted or resistant during the sample collection process, this embodiment uses paper handwriting images related to educational tasks as input samples, specifically including the subjects' paper assignments, classroom handwriting, test answers, etc.

[0060] Based on this, considering that the analysis of the subjects' anxiety states involves the subjects' privacy, in order to protect the subjects' privacy information, this embodiment, after acquiring the paper handwriting images related to the subjects' educational tasks, identifies whether the paper handwriting images contain the subjects' personal identity information. If the paper handwriting images contain the subjects' personal identity information, such as the subjects' names, student IDs, etc., the subjects' personal identity information is cropped out to achieve desensitization of the collected paper handwriting images.

[0061] Step S200: Input the paper handwriting image into the handwriting feature extraction model to obtain the subject's handwriting feature vector.

[0062] Handwriting features can reflect various handwriting attributes that reflect a subject's anxiety state, such as writing habits, physiological characteristics, and writing state. These handwriting features range from microscopic handwriting edge jitter to macroscopic neatness, reflecting the subject's anxiety state from different angles. To improve the comprehensiveness of the anxiety state analysis results corresponding to the subject's paper handwriting images, this embodiment employs a cascaded architecture of CNN and Transformer to extract handwriting feature vectors at different levels from paper handwriting images. The CNN model can efficiently capture low-level visual features such as edges and textures of images through sliding sampling of convolutional kernels, and possesses translation invariance and parameter sharing characteristics, resulting in high computational efficiency and superior local perception capabilities. The Transformer model can divide the image into several image patch sequences and capture global contextual information by calculating the correlation between any two image patches. This embodiment combines the CNN and Transformer models, retaining the CNN model's sensitivity to detail while also possessing the Transformer model's global understanding.

[0063] Since the handwriting feature extraction model in this embodiment, which includes a CNN-Transformer model architecture, needs to extract handwriting feature vectors of different dimensions, to solve the problem of multi-dimensional handwriting feature extraction, this embodiment uses a CNN network architecture as the backbone network and introduces a Transformer architecture at feature layers of different scales. Simultaneously, it combines a Feature Pyramid Network (FPN) to downsample and aggregate the extracted microscopic and mesoscopic image features, thus fusing the extracted multi-dimensional features. The aforementioned CNN-Transformer model architecture with the fused Feature Pyramid Network performs the feature extraction process of the paper handwriting image to obtain handwriting feature vectors of different levels, as follows:

[0064] (1) Extraction of microscopic handwriting features

[0065] The CNN layer uses local receptive fields to capture the second-order gradient of the handwriting edge, and obtains micro-handwriting features that simulate writing pressure, pen speed variation, and end-point burr rate. These micro-handwriting features can reflect the physiological tremor of the subject during the writing process.

[0066] (2) Extraction of mesoscopic handwriting features

[0067] The local self-attention layer uses a moving window mechanism to calculate the spatial correlation between image blocks, and obtains meso-level handwriting features such as word spacing variation coefficient and line alignment consistency. These meso-level handwriting features can reflect the logical stability of the subject's thinking.

[0068] (3) Extraction of macroscopic handwriting features

[0069] The global feature pyramid uses an FPN structure to downsample and aggregate the bottom-level features to obtain macro-handwriting features such as center of gravity deviation, white space rhythm, and layout balance. These macro-handwriting features can reflect the overall psychological defense state of the subject.

[0070] The aforementioned micro, meso, and macro handwriting features are integrated into a complete handwriting feature set. This complete handwriting feature set includes not only micro-level aspects such as pen speed, solid-light transitions, and burr rate, but also macro-level aspects such as center deviation and white space rhythm in the handwriting image on paper. This allows the feature set to reflect the subject's anxiety state from different dimensions.

[0071] Furthermore, before inputting the paper handwriting image into the aforementioned handwriting feature extraction model, a preprocessing step is included. This involves dividing the paper handwriting image into several fixed-size image patches, each of which is linearly mapped into a one-dimensional vector with positional encoding. By converting the paper handwriting image into sequential data, the handwriting recognition model is ensured from the ground up to analyze only the handwriting texture and pixel distribution, reducing the impact of the text content of the paper handwriting image on the handwriting feature extraction results.

[0072] In addition, to avoid the influence of ink droplets or other stains on the handwriting image on the handwriting feature extraction results, the feature extraction layer of the above handwriting feature extraction model introduces an oriented gradient histogram and an attention mask to denoise ink droplets or other stains. Specifically:

[0073] Obtain the directional gradient of each handwriting pixel in each image block to obtain the directional gradient histogram;

[0074] If the directional gradient of each handwriting pixel in the image block is uniformly distributed in all directions, an attention mask with a first weight is set for the handwriting pixel.

[0075] If the directional gradients of each handwriting pixel in the image block have consistent directional guidance and motion trajectory, then a second-weighted attention mask is set for the handwriting pixel.

[0076] The contribution value of the pixel is obtained by extracting the direction of the pixel and the corresponding attention mask;

[0077] If the contribution value of the pixel is less than a preset threshold, the contribution value of the pixel is set to zero.

[0078] The first weight is lower than the second weight.

[0079] By introducing histogram of oriented gradients (HARQ) analysis into the feature extraction layer, we can distinguish between real handwriting and stains. Real handwriting exhibits significant linear topological features at the microscopic level, such as clear directional guidance and movement trajectories, while stains appear as isotropic random spots. Through direction vector recognition, regions lacking dynamic patterns can be ignored. After the HARQ distinguishes between real handwriting and stains, the image regions are scored for saliency during the self-attention mechanism calculation: real handwriting portions that conform to writing kinematics (starting stroke, turning point, ending stroke) are assigned high attention weights; noisy portions lacking regularity and writing logic (such as paper creases, ink smudges, etc.) have lower confidence scores. In this case, combined with an attention mask, the contribution of noise features is set to zero before entering the final risk assessment, thus ensuring that the evaluation result is determined only by the effective handwriting features in the paper handwriting image.

[0080] Step S300: Compare the subject's handwriting feature vector with the subject's handwriting feature baseline for the same type of educational task to obtain the handwriting feature offset.

[0081] Specifically, considering the significant emotional fluctuations or changes in psychological state of subjects under different educational tasks, to improve the accuracy of the assessment of subjects' mental health in this embodiment, different handwriting feature baselines are set for different attributes of educational tasks during the handwriting feature vector comparison process. The attributes of the educational tasks include, but are not limited to, high-stress timed tests and low-stress in-class note-taking, each corresponding to different handwriting feature baselines. When the educational task is a high-stress timed test, such as a final timed exam, the sensitivity thresholds of the handwriting feature baselines for handwriting tremor and compactness (stress indicators) can be relaxed to eliminate interference from physiological tension caused by the exam environment, ensuring that the assessment results reflect the subject's underlying psychological state and avoiding misjudging exam anxiety as a long-term risk of anxiety disorder.

[0082] Step S400: Input the subject's academic performance, attendance records, and campus behavior into the multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector.

[0083] Specifically, to improve the comprehensiveness of the analysis results on the anxiety state of the subjects in this embodiment, this embodiment also introduces objective behavioral data such as academic performance (e.g., fluctuations in grades, sudden changes in rankings), attendance records (e.g., frequency of lateness, absenteeism, and leave requests), and campus behavior. By coupling the subjects' multi-source objective behavioral data with handwriting feature vectors, and supplementing the handwriting feature vectors with the subjects' objective behavioral data, the psychological analysis results are made more complete.

[0084] Step S500: The aligned handwriting feature offset and behavior embedding vector are converted into standard structured inference tokens.

[0085] In step S600, the standardized structured reasoning token is sent to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social functioning status.

[0086] This embodiment employs a multi-agent collaborative decoupled CNN-Transformer model to analyze aligned handwriting vector features and behavioral embedding vectors within the same data space, achieving distributed and efficient execution of complex models. In the cognitive collaboration stage, the aligned handwriting offsets and behavioral embedding vectors are converted into standard structured reasoning tokens, which are then assigned to corresponding specialized agents (such as handwriting analysis agents and behavioral data agents). Each agent performs detailed analysis on specific feature vectors, obtaining the physiological stress reflected by the handwriting offsets and the social functional state reflected by the objective behavioral embedding vectors.

[0087] In step S700, the physiological stress and social functioning status of the subject are sent to the multimodal coordination agent to generate a quantitative risk report and suggestions for educational scenarios.

[0088] Specifically, this embodiment adopts a multi-agent collaborative architecture. After the handwriting analysis agent and the campus behavior perception agent obtain specific results from the analysis of the corresponding feature vectors, the multimodal coordination agent calls the chain thinking mechanism and the consensus-based reasoning protocol to jointly infer the coupling degree of the subject's physiological stress and social functional state. This enables the exchange of feature vectorized encoded information among multiple agents, generates a quantitative risk report containing a logical evidence chain and scenario-based decision-making suggestions, and obtains a closed-loop early warning system from "feature perception" to "reasoning decision".

[0089] Specifically, based on the input of physiological stress and social functioning status into the psychological analysis results after multimodal coordinated agent processing, this embodiment can output the subject's mental health status from four dimensions: stress risk, depression risk, anxiety risk, and overall risk. By comparing the risk in each of the four dimensions with the minimum intervention threshold, different intervention paths are adopted according to the specific risk, as follows:

[0090] When the risk in all dimensions is below the minimum intervention threshold, positive guidance feedback is sent to the subjects, and suggestions for routine observation are sent to the monitoring end.

[0091] When the risk in any dimension exceeds the minimum intervention threshold but is below the high-risk threshold, the abnormal deviation points of handwriting features are highlighted in the heat map of the subject's handwriting features, and suggestions for one-on-one care conversations are sent to the teacher.

[0092] When the risk in any dimension exceeds the high-risk threshold, a specific mental health early warning report is generated by combining the subject's psychological analysis file, and a recommendation to refer the subject to a professional psychological institution for continuous psychological status monitoring is sent to the monitoring end.

[0093] Furthermore, considering that the psychological analysis data of the subjects may be distributed across different systems within the school (such as grade management systems, attendance management systems, etc.), and even that the data may be distributed across different campus data systems due to the subjects' advancement to higher education or transfer, in order to improve the completeness of the subjects' psychological analysis results, the non-semantic handwriting psychological early warning method for education in the above embodiment can be deployed in the cloud. Each school system is equipped with a corresponding edge terminal. The cloud and the edge terminals use a federated learning strategy for data processing. Each edge terminal uploads the collected paper handwriting images to the cloud. The cloud processes the received paper handwriting images according to the steps S100-S700 described above. At this time, each edge terminal also sets up privacy processing steps for the collected paper handwriting images as follows:

[0094] Step S801: Input the collected paper handwriting image of the subject into the salted encryption model to obtain preliminary encrypted data.

[0095] Step S802: Input the preliminary encrypted data into the hash encryption model to obtain the encrypted hash value.

[0096] Step S803: Send the encrypted hash value to the cloud.

[0097] Through the above-mentioned salted encryption model and hash encryption model, the original paper handwriting images collected by each edge terminal are uploaded to the cloud after the operation of "inserting random values ​​and converting to hash codes". This enables the transmission of subject psychological analysis-related data from multiple terminals without exchanging the original handwriting data, improving data processing efficiency while also enhancing the privacy of subject psychological analysis-related data.

[0098] Furthermore, regarding the integration of the subject's physiological stress and social functioning status by the multimodal coordination agent in step S700 above, this embodiment can also adjust the weights of relevant parameters according to the subject's psychoanalysis stage, specifically as follows:

[0099] Step S701: Retrieve the subject's psychological analysis file based on the subject's identity ID.

[0100] Step S702: Determine the psychoanalysis stage of the subject based on the subject's psychoanalysis profile.

[0101] Step S703: When the subject is in the early crisis stage of psychoanalysis, the first weight corresponding to physiological stress is increased by a compensation weight.

[0102] Step S704: When the subject is in the late crisis stage of psychoanalysis, the second weight corresponding to the social functioning state is increased with a compensation weight.

[0103] Step S705: Based on the adjusted first or second weight, physiological stress, and social function status, input to the multimodal coordination agent to obtain a quantitative risk report and scenario-based decision-making suggestions.

[0104] The non-semantic handwriting psychological early warning method for education provided in this embodiment can break free from the constraints of semantic recognition of handwriting content in existing mental health screening processes. Whether it's formulas, symbols, or meaningless scribbles written by the subject, they can all be used for mental health assessment, reducing interference from the subject's writing content. Furthermore, the assessment of the subject's handwriting characteristics replaces the group-wide standard in existing technologies with individual baseline offset analysis corresponding to the attributes of the educational task, significantly reducing the misjudgment rate and achieving closed-loop management in educational scenarios. This embodiment employs a multimodal coordination agent to couple specific physiological stress and social functional states, generating quantitative risk reports and decision-making suggestions related to the teaching scenario, which has higher reference value compared to existing technologies that only provide psychological analysis conclusions.

[0105] To enhance understanding of the educational-oriented non-semantic handwriting psychological early warning method in this application embodiment, the following examples are used for illustration:

[0106] Example 1:

[0107] In a typical teaching setting at a middle school, a unique psychological analysis file was created for each participant. Weekly handwriting images from participants' Chinese homework and math quizzes were collected. Personal identifiers such as name and student ID were removed from the handwriting images, and a salted encryption algorithm was used to generate random numbers. These random numbers were then combined with the handwriting images to convert them into hash values, which were then uploaded to the cloud.

[0108] For subject A, a baseline for high-pressure handwriting characteristics is generated based on the handwriting images of math quizzes uploaded by subject A in the previous four weeks, and a baseline for low-pressure handwriting characteristics is generated based on the handwriting images of Chinese homework uploaded by subject A in the previous four weeks.

[0109] By comparing the handwriting feature vector of the math quiz paper image uploaded by Subject A in week 5 with the high-pressure handwriting feature baseline, it was detected that Subject A's stroke curvature smoothness decreased in the micro-handwriting feature vector dimension, the line spacing abnormally contracted in the meso-handwriting feature vector dimension, and the overall layout tilted to the left in the macro-handwriting feature vector dimension, with the above handwriting feature offsets exceeding 40%. Combined with Subject A's absence from class once in week 5 and the decline in Subject A's academic performance, the multimodal coordination agent output "Subject A's overall risk 75%" and pushed a high-risk warning report to the homeroom teacher, Duan, recommending timely referral for psychological counseling.

[0110] Example 2:

[0111] In large-scale unified examination scenarios, a non-semantic handwriting psychological early warning method for education is deployed on the local area network server of the education bureau to process student exam answer sheets uploaded by multiple schools in the jurisdiction. Each school deploys an edge processor, while the education bureau's local area network server deploys a cloud processor. After removing personal privacy information from the student exam answer sheets, the edge processors in each school introduce random numbers through a salted encryption algorithm, convert them into encrypted hash values, and then send them to the cloud processor.

[0112] After receiving Subject B's exam answer sheet, the cloud processor calls up the handwriting feature vectors of Subject B's previous major exams to construct the high-pressure handwriting feature baseline of Subject B.

[0113] The handwriting feature vector of Subject B's current exam answer sheet is compared with the baseline of high-stress handwriting features. The judgment threshold of the psychological stress dimension is appropriately increased to eliminate the physiological handwriting fluctuations caused by Subject B's tension during the exam. In addition, objective behavioral data such as Subject B's borrowing frequency and attendance records in the library, fed back by the edge processor of Subject B's school, are combined with Subject B's handwriting feature vector to form a complementary chain of evidence.

[0114] When Subject B's anxiety risk exceeds the minimum intervention threshold (40%) but is below the high-risk threshold (80%), the abnormal deviation points of Subject B's handwriting characteristics are highlighted in the handwriting characteristic heatmap and sent to the teacher's end. This allows the teacher to see which part of the handwriting indicates an increased anxiety risk, and at the same time, suggestions for one-on-one care conversations are sent to the teacher's end.

[0115] It should be understood that, although attached Figure 1 The steps in the flowchart are shown sequentially according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders. Furthermore, [the following is a list of steps]. Figure 1At least some of the steps in the process may include multiple sub-steps or sub-stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0116] The embodiments disclosed above describe in detail a non-semantic handwriting psychological early warning method for education. Since the above-disclosed method can be implemented using various types of devices, this invention also discloses a non-semantic handwriting psychological early warning device for education, in conjunction with the appendix. Figure 2 The following are specific embodiments for detailed explanation.

[0117] Image acquisition module 901 is used to acquire images of handwriting on paper related to educational tasks;

[0118] The handwriting feature extraction module 902 is used to input the paper handwriting image into the handwriting feature extraction model to obtain the subject's handwriting feature vector;

[0119] The handwriting offset module 903 is used to compare the subject's handwriting feature vector with the subject's handwriting feature baseline in the same type of educational task to obtain the handwriting feature offset amount.

[0120] The objective behavior acquisition module 904 is used to input the subject's academic performance fluctuations, attendance records, and campus behavior into the multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector.

[0121] Feature standardization module 905 is used to convert aligned handwriting feature offsets and behavior embedding vectors into standard structured inference tokens;

[0122] The agent distribution module 906 is used to send the standardized structured reasoning token to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social function status.

[0123] The risk warning module 907 is used to send the subject's physiological stress and social functioning status to the multimodal coordination agent to generate a quantitative risk report and suggestions for educational scenarios.

[0124] For information on non-semantic handwriting-based psychological early warning devices for education, please refer to the methodological limitations outlined above; they will not be repeated here. Each module in the aforementioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the terminal device in hardware form or independently of it, or stored in the memory of the terminal device in software form, so that the processor can call and execute the corresponding operations of each module.

[0125] In one embodiment, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described psychological warning method for non-semantic handwriting in education.

[0126] The computer-readable storage medium may be an electronic storage device such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), hard disk, or ROM. Optionally, the computer-readable storage medium includes a non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program code that performs any of the method steps described above. This program code can be read from or written to one or more computer program products, and the program code may be compressed in an appropriate form.

[0127] In one embodiment, the present invention provides a computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the above-described educational non-semantic handwriting psychological early warning method.

[0128] The computer device includes a memory, a processor, and one or more computer programs, wherein the one or more computer programs may be stored in the memory and configured to be executed by one or more processors, and the one or more application programs are configured to perform the above-described educational non-semantic handwriting psychological warning method.

[0129] A processor may include one or more processing cores. The processor connects to various parts of the computer device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory, and by calling data stored in memory. Optionally, the processor may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also be implemented separately as a communication chip, without being integrated into the processor.

[0130] The memory may include random access memory (RAM) or read-only memory (ROM). The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), and instructions for implementing the various method embodiments described above. The data storage area may also store data created by the terminal device during use.

[0131] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A non-semantic handwriting psychological early warning method for education, characterized in that, include: Acquire images of handwriting on paper related to educational tasks; The handwriting image on the paper is input into the handwriting feature extraction model to obtain the subject's handwriting feature vector, specifically: The CNN layer uses local receptive fields to capture the second-order gradient of the handwriting edges, thus obtaining micro-handwriting features; The local self-attention layer uses a moving window mechanism to calculate the spatial correlation between image patches and obtain mesoscopic handwriting features; The global feature pyramid uses an FPN structure to downsample and aggregate the low-level features to obtain macro-handwriting features; The microscopic, mesoscopic, and macroscopic handwriting features are integrated to obtain a handwriting feature vector. The handwriting feature vector of the subject is compared with the handwriting feature baseline of the subject in the same type of educational task to obtain the handwriting feature offset. The subjects' academic performance fluctuations, attendance records, and campus behavior are input into a multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector. The aligned handwriting feature offsets and behavior embedding vectors are converted into standard structured inference tokens; The standard structured reasoning tokens are sent to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social functioning status; The physiological stress and social functioning status of the subjects are sent to a multimodal coordination agent to generate a quantitative risk report and suggestions for educational scenarios, specifically: Physiological stress and social functioning status are input into the psychoanalysis results after multimodal coordinated agent processing, including the subject's stress risk, depression risk, anxiety risk, and overall risk; When the risk in all dimensions is below the minimum intervention threshold, positive guidance feedback is sent to the subjects, and suggestions for routine observation are sent to the monitoring end. When the risk in any dimension exceeds the minimum intervention threshold but is below the high-risk threshold, the abnormal deviation points of handwriting features are highlighted in the heat map of the subject's handwriting features, and suggestions for one-on-one care conversations are sent to the teacher. When the risk in any dimension exceeds the high-risk threshold, a specific mental health early warning report is generated by combining the subject's psychological analysis file, and a recommendation to refer the subject to a professional psychological institution for continuous psychological status monitoring is sent to the monitoring end.

2. The non-semantic handwriting psychological early warning method for education as described in claim 1, characterized in that, The handwriting images include subjects' written assignments, classroom handwriting, and test answers.

3. The non-semantic handwriting psychological early warning method for education as described in claim 1, characterized in that, Before inputting the paper handwriting image into the aforementioned handwriting feature extraction model, the following steps are also included: The paper handwriting image is divided into several image blocks of fixed size; Each image patch is converted into a one-dimensional vector with positional encoding through a linear mapping; The one-dimensional vector is input into the handwriting feature extraction model.

4. The non-semantic handwriting psychological early warning method for education as described in claim 3, characterized in that, After dividing the paper handwriting image into several fixed-size image blocks, the process further includes: Obtain the directional gradient of each handwriting pixel in each image block to obtain the directional gradient histogram; If the directional gradient of each handwriting pixel in the image block is uniformly distributed in all directions, an attention mask with a first weight is set for the handwriting pixel. If the directional gradients of each handwriting pixel in the image block have consistent directional guidance and motion trajectory, then a second-weighted attention mask is set for the handwriting pixel. The contribution value of the pixel is obtained by extracting the direction of the pixel and the corresponding attention mask; If the contribution value of a pixel is less than a preset threshold, the contribution value of the pixel is set to zero.

5. The non-semantic handwriting psychological early warning method for education as described in claim 1, characterized in that, The educational non-semantic handwriting psychological early warning method is deployed in the cloud, which is communicatively connected to various edge devices. After each edge device acquires a paper handwriting image, it further includes: The edge inputs the collected paper handwriting images of the subjects into the salted encryption model to obtain preliminary encrypted data; The edge device inputs the preliminary encrypted data into the hash encryption model to obtain the encrypted hash value; The edge device sends the encrypted hash value to the cloud.

6. A psychological early warning device for non-semantic handwriting analysis aimed at education, characterized in that, include: The image acquisition module is used to acquire images of handwriting on paper related to educational tasks. The handwriting feature extraction module is used to input the paper handwriting image into the handwriting feature extraction model to obtain the subject's handwriting feature vector, specifically: The CNN layer uses local receptive fields to capture the second-order gradient of the handwriting edges, thus obtaining micro-handwriting features; The local self-attention layer uses a moving window mechanism to calculate the spatial correlation between image patches and obtain mesoscopic handwriting features; The global feature pyramid uses an FPN structure to downsample and aggregate the low-level features to obtain macro-handwriting features; The microscopic, mesoscopic, and macroscopic handwriting features are integrated to obtain a handwriting feature vector. The handwriting offset module is used to compare the subject's handwriting feature vector with the subject's handwriting feature baseline in the same type of educational task to obtain the handwriting feature offset amount. The objective behavior acquisition module is used to input the subject's academic performance fluctuations, attendance records, and campus behavior into the multilayer perceptron to obtain a behavior embedding vector with the same dimension as the handwriting feature vector. The feature standardization module is used to convert aligned handwriting feature offsets and behavior embedding vectors into standard structured inference tokens; The agent distribution module is used to send the standard structured reasoning token to the handwriting analysis agent and the campus behavior perception agent respectively to obtain the subject's physiological stress and social function status. The risk warning module is used to send the subject's physiological stress and social functioning status to the multimodal coordination agent, generating a quantitative risk report and suggestions for the educational scenario, specifically: Physiological stress and social functioning status are input into the psychoanalysis results after multimodal coordinated agent processing, including the subject's stress risk, depression risk, anxiety risk, and overall risk; When the risk in all dimensions is below the minimum intervention threshold, positive guidance feedback is sent to the subjects, and suggestions for routine observation are sent to the monitoring end. When the risk in any dimension exceeds the minimum intervention threshold but is below the high-risk threshold, the abnormal deviation points of handwriting features are highlighted in the heat map of the subject's handwriting features, and suggestions for one-on-one care conversations are sent to the teacher. When the risk in any dimension exceeds the high-risk threshold, a specific mental health early warning report is generated by combining the subject's psychological analysis file, and a recommendation to refer the subject to a professional psychological institution for continuous psychological status monitoring is sent to the monitoring end.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the psychological early warning method for non-semantic handwriting in education as described in any one of claims 1-5.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it performs the psychological early warning method for non-semantic handwriting for education as described in any one of claims 1-5.

Images