An interactive classroom student dynamic image generation method, medium and system

Abstract

The application provides an interactive classroom student dynamic portrait generation method, medium and system, belongs to the technical field of student dynamic portrait, and first collects a student face image sequence, extracts face feature points, constructs an expression feature matrix, combines eye movement tracking technology to obtain visual focus data, calculates eye light concentration degree indexes, establishes an individual correlation model, identifies specific expressions, and analyzes attention fluctuation characteristics. The expression similarity vector, eye light concentration degree index and other multi-dimensional characteristics are fused into a dynamic attention state score. Based on a multi-layer bidirectional transformer network, a learning state evaluation is realized, wherein the multi-head attention mechanism parameters are determined by the expression change frequency, the eye light concentration degree threshold and the attention fluctuation rate. Finally, a student dynamic portrait feature map is generated, a classroom interaction effect evaluation index is formed through cluster analysis, and the technical problem that a single feature cannot accurately evaluate the dynamic change of the student classroom attention state is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of student dynamic portrait, and specifically relates to an interactive classroom student dynamic portrait generation method, medium and system. BACKGROUND

[0002] In the process of classroom teaching, accurately evaluating the attention state of students is crucial for optimizing teaching strategies. Traditional classroom attention evaluation mainly relies on teacher subjective observation or questionnaire survey, which is difficult to realize objective quantitative evaluation. With the development of computer vision technology, attention evaluation methods based on facial expression recognition have gradually been applied in the field of education. Through camera equipment to collect student facial data and conduct emotional and concentration analysis, teachers can be provided with classroom feedback.

[0003] However, the existing technology has obvious limitations. First, most methods are based on single feature (such as facial expression or head posture) for evaluation, without considering multi-dimensional physiological behavior parameters such as gaze tracking; second, existing systems mostly use static evaluation models, which are difficult to capture the timing dynamic changes of student attention; finally, there is a lack of individual difference modeling mechanism, which is difficult to adapt to the expression and attention correlation mode of different students.

[0004] This leads to the fact that the existing technology is difficult to accurately evaluate the dynamic changes of student attention state in complex teaching scenarios, and cannot provide real-time and accurate classroom interaction optimization suggestions for teachers, which seriously restricts the application effect of intelligent education technology. That is to say, there is a technical problem of being difficult to accurately evaluate the dynamic changes of student attention state in the classroom based on single feature in the existing technology. SUMMARY

[0005] Therefore, the present application provides an interactive classroom student dynamic portrait generation method, medium and system, which can solve the technical problem of being difficult to accurately evaluate the dynamic changes of student attention state in the classroom based on single feature in the existing technology.

[0006] The application is implemented in the following manner: a first aspect of the application provides a student dynamic portrait generation method for interactive classroom, comprising: collecting a sequence of student facial images in the classroom, extracting facial key point data to construct an expression feature matrix; calculating the similarity of the expression feature matrix and a standard expression library to generate an expression similarity vector, recording the expression change frequency, and establishing an expression dynamic change curve; collecting a sequence of visual focus point coordinates, calculating the eye focus area and residence time, generating a gaze heat map, and quantifying the eye focus degree index; combining the expression feature matrix and the eye focus degree index to establish an individual correlation model; identifying specific expressions and marking learning state abnormal points; using a sliding time window to analyze the relevance of attention concentration degree and classroom content nodes, constructing an attention fluctuation curve; applying a state transition function to model the attention fluctuation, generating an attention state score; based on a learning state evaluation network model, fusing multi-dimensional features to generate a dynamic portrait feature map as a student dynamic portrait; further comprising applying a clustering algorithm to analyze multiple student dynamic portrait feature maps to form a classroom interaction effect evaluation index.

[0007] The expression change frequency is the number of significant changes in the expression similarity vector of the student within a unit of time, which is obtained by calculating the difference between adjacent time points, and reflects the activity level of the student's emotional state change.

[0008] The specific expression is an expression change of the student in the classroom that has a significant statistical difference from the individual expression baseline or the average expression state of the class, which is identified by the Mahalanobis distance or Z-score method, and reflects the student's unconventional reaction to the teaching content.

[0009] The eye focus degree index is the proportion of the residence time of the student's visual line in the specified area to the total observation time, which is obtained by calculating the spatial aggregation degree and time density of the eye movement trajectory points, and reflects the stability of the student's visual attention.

[0010] The eye focus degree threshold is a critical value for determining whether the student's visual line is in a concentrated state, which is determined by analyzing historical data, and is used to convert continuous eye focus degree index into discrete attention state judgment.

[0011] The eye focus area is the spatial range where the student's visual line frequently stays, which is obtained by analyzing the sequence of visual focus point coordinates using a density clustering algorithm, and is represented as a set of hot area coordinates and weight distribution on a two-dimensional plane.

[0012] The attention concentration degree index is a quantitative index of the student's cognitive input level calculated by integrating the expression feature stability and the eye focus degree index, which is obtained by weighting and fusing multiple physiological behavior feature parameters, and reflects the student's attention level and cognitive processing depth to the classroom content.

[0013] The individual correlation model is a mapping function between the facial feature matrix and the attention concentration index for each student, is trained by a deep learning algorithm, and is used to adjust the evaluation parameter weight according to the individual difference of the student, and improve the attention state judgment accuracy.

[0014] The state transition function is used to predict the attention state transition trend according to the multi-dimensional feature information and the historical state of the student at the current moment, the input includes the expression similarity vector, the gaze concentration index, the specific expression frequency, the attention historical value and the classroom content difficulty coefficient, and the output is the attention state score representing the current attention input degree of the student and the change trend prediction, and the student attention state is continuously tracked and predicted through recursive calculation.

[0015] The learning state evaluation network model structure is a multi-layer bidirectional transformer network architecture, includes an encoder layer and a decoder layer, the encoder layer is responsible for extracting the student facial feature matrix and the gaze concentration index time sequence feature, the decoder layer is responsible for fusing the time sequence feature and outputting the student attention state score, the model core is a multi-head attention mechanism, the number of heads is determined by the expression change frequency, the dimension of each attention head is determined by the gaze concentration threshold, the attention layer depth is calculated by the attention fluctuation rate, the model adopts a residual connection structure to reduce the gradient disappearance problem, a layer normalization operation is added after each layer to improve the training stability, and the high-dimensional feature is mapped to the student attention state score space through a full connection layer.

[0016] The second aspect of the present application provides a computer readable storage medium, the computer readable storage medium stores program instructions, the program instructions are used to execute the above-mentioned interactive classroom student dynamic portrait generation method when running in the computer.

[0017] The third aspect of the present application provides an interactive classroom student dynamic portrait generation system, including the above-mentioned computer readable storage medium, the system is any one of a computer, a server and a single-chip microcomputer, the computer readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0018] Compared with the prior art, the present application provides an interactive classroom student dynamic portrait generation method, medium and system, the present application provides an interactive classroom student dynamic portrait generation method based on multi-modal feature fusion, the facial expression sequence and the eye movement track data of the student are collected and analyzed at the same time, the expression feature matrix and the gaze concentration index are constructed, and the individual correlation model is established by combining the deep learning technology, so that the attention state of the student is accurately evaluated and predicted.

[0019] The method effectively solves the limitations of the prior art. Through the multi-feature fusion mechanism, the one-sidedness of single feature evaluation is overcome, and the evaluation accuracy is improved; the sliding time window technology and the state transition function are used to realize continuous tracking of the dynamic changes of the attention state; the individual correlation model is introduced to solve the problem of one-size-fits-all evaluation criteria and realize personalized evaluation for different students.

[0020] Therefore, the present application successfully solves the technical problem of accurately evaluating the dynamic changes of students' classroom attention states, provides real-time and accurate classroom interaction optimization suggestions for teachers, and significantly improves the application value and teaching effect of intelligent education technology. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the method of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application.

[0023] As Figure 1 shown is a flowchart of a dynamic portrait generation method for students in an interactive classroom provided by the first aspect of the present application, and the method includes the following steps:

[0024] S01, a plurality of student classroom face image sequences are collected, student face key point data is extracted through face feature point positioning technology, an expression feature matrix is constructed, and the trajectory of each student face feature point changing with time is marked;

[0025] S02, the similarity of each student expression feature matrix and the standard expression library is calculated, an expression similarity vector is generated, the expression change frequency is recorded, and a student classroom expression dynamic change curve is established;

[0026] S03, a student gaze focus coordinate sequence is collected through eye tracking technology, an eye concentration area and a residence time are calculated, a gaze heat map is generated, and an eye concentration degree index is quantified;

[0027] S04, an individual correlation model is established by combining the student expression feature matrix and the eye concentration degree index, and an attention concentration degree evaluation function of the student individual is trained through a deep learning algorithm;

[0028] S05, the student expression feature matrix in different time periods of the classroom is statistically analyzed, and a specific expression deviating greatly from the average expression state of the classroom is identified and marked as a learning state abnormal point;

[0029] S06, analyze the time correlation between the student attention concentration index and the classroom content node using the sliding time window technique, and construct a dynamic attention fluctuation curve;

[0030] S07, apply a state transition function to model the student attention fluctuation, input parameters including expression similarity vector, eye concentration index, specific expression frequency, attention history value, and classroom content difficulty coefficient, to generate the student's current attention state score;

[0031] S08, based on the learning state evaluation network model, fuse multi-dimensional features, wherein the multi-head attention mechanism parameters are determined by expression change frequency, eye concentration threshold, and attention fluctuation rate, to generate a dynamic portrait feature map of each student as a student dynamic portrait;

[0032] S09, optionally, apply clustering algorithm to analyze multiple student dynamic portrait feature maps, identify class learning state distribution, form classroom interaction effect evaluation index, and provide real-time classroom interaction optimization suggestions for teachers;

[0033] Among them, the expression feature matrix specifically refers to the data set of the spatial position relationship of multiple key points of the student's face and its change over time, including the position, motion speed and relative position change rate of key points such as eyebrows, eyes and corners of the mouth. Multi-dimensional feature data.

[0034] Among them, the expression similarity vector specifically refers to the Euclidean distance or cosine similarity calculation result between the student's current expression feature matrix and each type of expression template in the pre-defined standard expression library, which is used to quantify the closeness of the student's expression to the standard learning state of concentration, confusion, understanding, etc.

[0035] Among them, the expression change frequency specifically refers to the number of times the student's expression similarity vector changes significantly within a unit of time, which is obtained by calculating the difference between adjacent time points of the expression similarity vector, reflecting the activity level of the student's emotional state change.

[0036] Among them, the specific expression specifically refers to the expression change of the student in the classroom that has a significant statistical difference with the student's personal expression baseline or the average expression state of the class, which is identified by Mahalanobis distance or Z-score method, reflecting the student's unconventional reaction to the teaching content.

[0037] Among them, the specific expression frequency specifically refers to the number of times the student produces specific expressions within a unit of time, which is obtained by counting statistics, as a quantitative indicator of the student's cognitive conflict or emotional fluctuation.

[0038] The gaze concentration degree index is specifically a proportion of a time for which a student's line of sight stays in a specified area in total observation time, is obtained by calculating a spatial aggregation degree and a time density of eye movement track points, and reflects stability of the student's visual attention.

[0039] The gaze concentration degree threshold is specifically a critical value for judging whether the student's line of sight is in a concentrated state, is determined by analyzing historical data, and is used to convert continuous gaze concentration degree indexes into discrete attention state judgments.

[0040] The gaze concentration area is specifically a spatial range in which the student's line of sight frequently stays, is obtained by analyzing a line-of-sight focus coordinate sequence by a density clustering algorithm, and is represented as a hotspot area coordinate set and a weight distribution on a two-dimensional plane.

[0041] The attention concentration degree index is specifically a student's cognitive input degree quantitative index calculated by integrating expression feature stability and the gaze concentration degree index, reflects a degree of attention and a depth of cognitive processing of the student on a classroom content by weightedly fusing multiple physiological behavior feature parameters.

[0042] The attention history value is specifically a sequence of the student's attention concentration degree indexes in a past time window, records a time evolution of the student's attention state, and is used for time sequence modeling of a state transition function.

[0043] The attention fluctuation rate is specifically a variation amplitude of the student's attention concentration degree index in a unit time, is obtained by calculating a first derivative or a difference of the attention concentration degree index, and reflects stability of the student's attention.

[0044] The classroom content difficulty coefficient is specifically a quantitative evaluation of a cognitive complexity of a current teaching content, is pre-set by a teacher or automatically calculated from student feedback data, and is used as a background parameter for adjusting an attention state evaluation standard.

[0045] The attention state score is specifically a numerical value quantitatively representing a current cognitive input degree of the student, is calculated by a state transition function, and is used as a core dimension of a dynamic portrait feature map.

[0046] The dynamic portrait feature map is specifically a multi-dimensional feature set and a time evolution pattern describing a student's learning state, includes time sequence data of multiple dimensions such as expression similarity vectors, gaze concentration degree indexes, and attention state scores, and presents a student's cognitive and emotional state in a visualized form.

[0047] The classroom interaction effect evaluation index is specifically a numerical index quantifying effectiveness of classroom teaching activities, is obtained by analyzing multiple student dynamic portrait feature maps, and includes multiple dimensions such as a class average attention state score, attention synchronization, and emotion distribution.

[0048] The individual correlation model is specifically a mapping function between the facial feature matrix and the attention concentration index established for each student, is trained by a deep learning algorithm, can adjust the evaluation parameter weight according to the individual difference of the student, and improves the accuracy of the attention state judgment.

[0049] The state transition function is used to predict the transition trend of the attention state of the student according to the multi-dimensional feature information and the historical state of the student at the current moment, the input includes the expression similarity vector, the gaze concentration index, the specific expression frequency, the attention history value and the classroom content difficulty coefficient, and the output is the attention state score representing the current attention input degree of the student and the change trend prediction, and the continuous tracking and prediction of the attention state of the student are realized through recursive calculation.

[0050] The specific structure of the learning state evaluation network model is a multi-layer bidirectional transformer network architecture, including an encoder layer and a decoder layer, wherein the encoder layer is responsible for extracting the time sequence features of the facial feature matrix and the gaze concentration index of the student, the decoder layer is responsible for fusing the time sequence features and outputting the attention state score of the student, the core of the model is a multi-head attention mechanism, the number of heads is determined by the expression change frequency, the dimension of each attention head is determined by the gaze concentration threshold, and the depth of the attention layer is calculated by the attention fluctuation rate, the model adopts a residual connection structure to alleviate the gradient disappearance problem, and a layer normalization operation is added after each layer to improve the training stability, and finally the high-dimensional features are mapped to the student attention state score space through a fully connected layer.

[0051] The steps of establishing the training data set in the training process of the learning state evaluation network model specifically include that firstly, facial expression sequences and eye movement data of students in different subject classrooms are collected from multiple classes of multiple schools, and the student attention state labels and objective test scores rated by teachers are recorded as supervision signals, then data cleaning and preprocessing are performed, including abnormal value detection and elimination, data standardization and enhancement, then the processed data is divided according to the time window to build sequence samples, each sample contains a fixed length of expression sequence and eye movement sequence, and is accompanied by the corresponding attention state label, finally the data set is divided into training set, validation set and test set according to the ratio of 7:2:1, and the data of different classes and subjects is uniformly distributed in the three data sets, so as to ensure the generalization ability of the model.

[0052] The step of training the learning state evaluation network model specifically comprises firstly model initialization, the weight adopts the truncated normal distribution initialization method, then adopts the two-stage training strategy, the first stage uses the self-supervised learning method to train, learns the internal representation of the expression sequence and the eye movement sequence through the design of the timing mask prediction task, the second stage uses the supervised learning method to fine-tune, and the student attention state label is used as the supervision signal to optimize the model parameter, the cosine annealing learning rate scheduling strategy is used in the training process, the initial learning rate is set to 0.001, the performance of the validation set is gradually reduced during the training process, the gradient clipping technology is used to prevent gradient explosion, and the early stopping strategy is used to avoid overfitting, the performance index on the validation set is stopped when there is no improvement for 5 consecutive epochs, and the whole training process is carried out in parallel on four GPUs, the batch size of each GPU is set to 32, and a total of 100 epochs are trained.

[0053] The specific implementation of the above steps is described in detail below.

[0054] The specific implementation of step S01 is to detect and track the student facial key points by using a deep convolutional neural network model. First, a high-definition camera is used to collect videos of students in the classroom, and the frame rate is set to 30 frames per second and the resolution is 1080P to ensure clear and identifiable facial features. After video collection, a face detection algorithm is used to locate the face area in each frame of image, and an improved MTCNN (Multi-Task Convolutional Neural Network) algorithm is used to detect the face area and the positions of eyes, nose, mouth and other main organs at the same time, with a detection accuracy of more than 98%. The improved 68-point facial feature point positioning algorithm in the Dlib library is applied to the detected face area to extract a complete set of facial feature points including eyebrows (6 points on the left and right), eyes (6 points on the left and right), nose (9 points), mouth (20 points) and contour (17 points). After feature point extraction, the coordinates of each feature point are normalized to eliminate the scale difference caused by the distance between the student and the camera, and an expression feature matrix F is constructed, with a dimension of n x 68 x 2, where n represents the number of video frames, 68 represents the number of feature points, and 2 represents the x and y coordinates of each feature point. The Kalman filter algorithm is used to smooth the feature point trajectory to eliminate tracking noise caused by changes in light or small head movements, ensuring the continuity and stability of the feature point trajectory. Finally, the displacement vector of the feature points between adjacent frames is calculated, the motion trajectory of each feature point over time is recorded, and a complete expression feature matrix time series dataset is constructed. The computational complexity of the whole process is O(n), which can realize real-time processing.

[0055] The specific implementation of step S02 is to construct an expression similarity calculation and analysis framework. First, a standard expression library is established, which includes standard expression templates corresponding to 10 typical learning states such as concentration, confusion, understanding, boredom, fatigue, etc. Each expression template is annotated by a professional expression recognition expert and verified multiple times. Principal component analysis (PCA) is used to reduce the dimensionality of the facial feature point data, extract the main mode of expression change, and retain the principal components that explain 95% of the variance, usually 15-20 principal components. For the expression feature matrix extracted from each frame of student facial image, the cosine similarity with each template in the standard expression library is calculated to form a 10-dimensional expression similarity vector S. The improved cosine similarity formula is used for cosine similarity calculation, and the feature point weight coefficient is introduced to give higher weight to the key areas of emotional expression such as eyes and mouth, and to enhance the sensitivity of expression recognition. The similarity threshold is set to 0.75, and if it is higher than this threshold, it is determined as the corresponding expression category. Based on the sliding time window technology (window size is 3 seconds, step is 1 second), the expression change frequency is counted, and by calculating the Euclidean distance of the expression similarity vectors in adjacent time windows, when the distance exceeds the threshold value 0.3, it is recorded as an effective expression change. The number of expression changes in a unit of time (such as 1 minute) is accumulated to form the expression change frequency indicator F. Combining the expression similarity vector S and the expression change frequency F, the student's classroom expression dynamic change curve C is constructed, which reflects the trend of the student's emotional state changing with the progress of the classroom. The curve uses the spline interpolation method to ensure smooth continuity, which is convenient for subsequent analysis and visualization.

[0056] Alternatively, the standard expression templates corresponding to 10 typical learning states are as follows: Concentration: slightly furrowed eyebrows, focused and determined gaze, slightly closed lips; Confusion: furrowed brows, obvious forehead wrinkles, slightly tilted corners of the mouth, and a bewildered look; Understanding: relaxed eyebrows, bright eyes, and slightly raised corners of the mouth; Boredom: distracted eyes, half-lowered eyelids, expressionless face, and slightly open mouth; Fatigue: half-closed eyes, drooping eyelids, possible yawn, and relaxed face; Distraction: wandering eyes, unfocused gaze, and absent-minded expression; Enthusiasm: wide-open eyes, raised eyebrows, smiling face, and focused expression; Daydreaming: dull eyes, empty gaze, and blank expression; Anxiety: slightly furrowed brows, tightly closed lips, and uneasy eyes; Eureka: suddenly wide-open eyes, raised eyebrows, slightly open mouth, and a bright expression.

[0057] The specific implementation of step S03 is to realize an eye-tracking and gaze analysis system. First, based on the principles of infrared illumination and corneal reflection, a high-precision eye-tracking device (sampling rate 120Hz) is used to collect students' eye movement data. Eye-tracking data preprocessing includes steps such as removing blink interference, filtering noise points, and correcting drift. A second-order Butterworth low-pass filter (cutoff frequency 20Hz) is used to smooth the raw eye-tracking data to ensure signal quality. The I-VT (velocity threshold recognition) algorithm is used to identify fixation points and saccades in the gaze trajectory. The velocity threshold is set to 30° / second; eye movements below this threshold are classified as fixations, and those above are classified as saccades. For each identified fixation point, its spatial coordinates (x, y) and duration t are recorded, forming a gaze focus coordinate sequence {(x1, y1, t1), (x2, y2, t2), ..., (x... n y n , t n The DBSCAN (Density-Based Spatial Clustering) algorithm was used to cluster fixation points and identify areas of concentrated gaze. The algorithm parameters were set as follows: ε (neighborhood radius) = 50 pixels, MinPts (minimum number of points) = 5. Weights were calculated based on the number and duration of fixation points within each cluster, generating a two-dimensional gaze heatmap H. The heatmap was smoothed using a Gaussian kernel function, with the standard deviation of the kernel function set to the pixel value corresponding to a 1.5° viewing angle. Based on the heatmap, the area ratio and fixation time ratio of concentrated gaze areas were statistically analyzed to calculate the gaze concentration index E. This index ranges from [0, 1], with a larger value indicating more concentrated gaze. In practical applications, the gaze concentration threshold E0 was set to 0.6; values ​​higher than this were considered a state of focused attention.

[0058] The specific implementation of step S04 involves constructing an individual association model and an attention assessment system. For each student, an individual association model M is established, consisting of an facial expression feature matrix and an eye focus index. A Long Short-Term Memory (LSTM) network structure is used to capture the temporal dependency between facial expression changes and attentional states. The model input is the temporal data of the facial expression feature matrix F (truncated from a historical window of length T, T = 10 seconds), and the output is the predicted attention focus index. The LSTM network comprises an input layer, a bidirectional LSTM layer (number of hidden units = 128), a full connection layer, and an output layer, wherein the bidirectional LSTM layer is used to capture the forward and backward temporal dependencies of the expression changes. The model training uses the teacher-labeled attention state as a supervision signal, uses mean square error (MSE) as a loss function, and optimizes the parameters by the stochastic gradient descent method. In order to process the individual differences of students, an adaptive layer is introduced in the model to dynamically adjust the weight distribution according to the historical data of the students. The adaptive layer uses a self-attention mechanism to adaptively adjust the importance weight of different expression features. The model performance is evaluated by the cross-validation method, and the prediction accuracy on the validation set reaches more than 87%. The attention concentration index output by the model is combined with the actual eye light concentration index to form a comprehensive attention concentration evaluation index A through weighted average (weight ratio is 0.6:0.4). The index reflects the overall cognitive input of students to the classroom content, and the value range is [0, 1].

[0059] The specific implementation of step S05 is to perform specific expression recognition and learning state anomaly detection. Based on the statistical characteristics of the expression feature matrix, a student individual expression baseline and a class average expression state reference model are established. The individual expression baseline is obtained by calculating the mean and covariance matrix of the expression feature matrix of the student in the recent classroom (such as the previous three classes), and the class average expression state is calculated in real time based on the expression feature matrix of all students in the current classroom. The Mahalanobis distance is used to measure the deviation of the current expression of the student from the baseline or the class average state. The Mahalanobis distance calculation formula takes into account the correlation between features, and can more accurately reflect the abnormal state in the multi-dimensional feature space. The Mahalanobis distance threshold is set to 2.5. When the Mahalanobis distance between the expression feature matrix of the student and the baseline exceeds this threshold, it is marked as individual-specific expression; when the Mahalanobis distance with the class average state exceeds the threshold, it is marked as group-specific expression. The specific expression is classified and analyzed, and according to the expression similarity vector, it is judged whether it belongs to positive reaction (such as surprise, enlightenment, etc.) or negative reaction (such as confusion, boredom, etc.). A sliding window (window size is 5 minutes, step is 1 minute) is used to count the frequency of specific expression, and when the frequency exceeds the threshold (3 times / minute), a learning state anomaly warning is triggered. The time correspondence between specific expression and classroom content is established to identify the teaching content or activity that causes the abnormal reaction of students, and to provide a basis for teaching adjustment. The anomaly detection results are added to the student dynamic portrait in the form of time stamp annotations to form a set of learning state anomaly point markers.

[0060] The specific implementation of step S06 is to perform dynamic attention analysis and content association modeling. The sliding time window technique is used to analyze the time variation characteristics of the student attention concentration index, and the window size is set to 3 minutes and the sliding step is 30 seconds, ensuring the continuity and sensitivity of the analysis. The mean, standard deviation, and change rate of the attention concentration index are calculated for each time window to form an attention state description vector. Combined with the teaching content progress information, the classroom content is divided into multiple content nodes according to knowledge points or teaching activity forms, and each node is marked with a start and end timestamp. The Pearson correlation coefficient is used to calculate the time correlation between the attention concentration index and the content node, and the correlation coefficient takes a value in the range [-1, 1], a positive value indicating a positive correlation (attention improvement) and a negative value indicating a negative correlation (attention decline). Based on the attention state description vector and the content node timestamp, a dynamic attention fluctuation curve is constructed, and the curve uses a cubic spline interpolation method to ensure smoothness. The attention fluctuation rate is calculated, defined as the maximum change amplitude of the attention concentration index per unit time (e.g., 1 minute), and the fluctuation rate threshold is set to 0.2. Fluctuations exceeding this threshold are recorded as significant fluctuation points. The time correspondence between attention significant fluctuation points and teaching content nodes is analyzed to identify key content factors that affect student attention concentration. An attention fluctuation pattern library is established, including typical patterns such as rising, falling, fluctuating, and stable, and the student attention fluctuation curve is matched and classified to provide a basis for teaching strategy adjustment.

[0061] The specific implementation of step S07 is to implement a state transition function and an attention state evaluation system. An attention state transition function based on a Hidden Markov Model (HMM) is constructed, and the attention state space S = {S1, S2, S3, S4, S5} is defined, corresponding to five attention levels: very low, low, medium, high, and very high. The state transition matrix P is initialized based on expert experience and historical data analysis, and the matrix elements P ij represent the probability of transitioning from state S i to state S j . The observation probability matrix B is set, and the matrix elements B ij represent the probability of observing the attention concentration index in state S iThe probability of observing the eigenvector j is observed. The state transition function input parameters include: expression similarity vector (dimension 10), eye concentration index (scalar value, range [0, 1]), specific expression frequency (scalar value, unit: times / minute), attention history value (attention state sequence in the past 5 minutes), and course content difficulty coefficient (scalar value, range [0, 1]). The most likely hidden state sequence is solved using the Viterbi algorithm, and the most likely attention state at the current time is obtained. The state is mapped to an attention state score A, and the score range is [0, 100], wherein state S1 corresponds to 0-20 points, S2 corresponds to 21-40 points, S3 corresponds to 41-60 points, S4 corresponds to 61-80 points, and S5 corresponds to 81-100 points. For different course content difficulties, a dynamic threshold adjustment mechanism is set, and when the content difficulty coefficient increases, the attention state judgment threshold is appropriately reduced, and vice versa. The model parameters are optimized based on historical data by the expectation maximization (EM) algorithm, and the model parameters are updated once a semester to adapt to the changes in student cognitive development.

[0062] The specific implementation of step S08 is to construct an efficient learning state evaluation network and generate a dynamic portrait atlas as a student dynamic portrait. A multi-layer bidirectional transformer network architecture is used to implement the learning state evaluation network model, which includes 6 encoder layers and 2 decoder layers, each of which uses a multi-head attention mechanism. The encoder layer is responsible for extracting the time sequence features of the student expression feature matrix and the eye concentration index, and the input sequence length is 128 (corresponding to about 4 minutes of classroom data), and the feature dimension is 134 (including the x and y coordinates of 68 facial feature points and multi-dimensional eye movement indicators). The encoder uses position encoding technology to process the time sequence information of the sequence data, and the position encoding uses sine and cosine functions to generate. The number of heads of the multi-head attention mechanism is determined by the expression change frequency, and the calculation formula is where F is the expression change frequency, and the number of heads is usually between 4 and 8. The dimension d of each attention head is determined by the eye concentration threshold, and the calculation formula is where E0 is the eye concentration threshold, and the dimension is usually between 38 and 45. The depth L of the attention layer is calculated from the attention fluctuation rate R, and the calculation formula is The depth is usually between 6 and 8. The model uses a residual connection structure to alleviate the gradient disappearance problem, and adds a layer normalization operation after each layer to improve the training stability, and the normalization parameter ε is set to 10 -6 The decoder layer is responsible for fusing the time sequence features extracted by the encoder, and predicting the future state through the mask self-attention mechanism. Finally, the high-dimensional features are mapped to the student attention state score space through a fully connected layer, and the output dimension is 1, representing the attention state score at the current time.

[0063] The specific steps for establishing the learning state evaluation network model training data set include first collecting facial expression sequences and eye movement data of students in different subjects in the classroom from multiple classes in multiple schools. The collection equipment includes a high-definition camera (resolution 1080P) and a professional eye tracking instrument (sampling rate 120Hz). At the same time, the teacher's rating of the student's attention state label (using a 5-level rating: 1 - completely inattentive, 2 - slightly attentive, 3 - moderately attentive, 4 - relatively high attention, 5 - high attention) and objective test scores (including classroom quizzes and homework scores) are recorded as supervision signals. The duration of the collection is not less than 10 teaching periods for each class to ensure that the data covers different teaching content and learning stages. The collected raw data is cleaned and preprocessed, including outlier detection and rejection (using the 3σ principle to identify abnormal data points), data standardization (normalizing each feature to the [0, 1] interval), and data augmentation (expanding the training samples through time series disturbance, noise injection, and feature replacement techniques). Then the processed data is divided into sequence samples according to the time window, with a window length of 128 frames (about 4 minutes) and a sliding step of 32 frames (about 1 minute). Each sample contains a fixed-length expression sequence and eye movement sequence, along with the attention state label for the corresponding time period. To balance the sample distribution of different attention states, a stratified sampling strategy is used to ensure that the sample proportion of each attention state is close to 1:1:1:1:1. Finally, the data set is divided into training set, validation set and test set according to the ratio of 7:2:1. During the division process, stratified random sampling method is used to ensure that the data of different classes and subjects is evenly distributed in the three data sets to ensure the generalization ability of the model.

[0064] The specific steps for training the learning state evaluation network model include first initializing the model. The weights are initialized using the truncated normal distribution initialization method with a truncation range of [-0.02, 0.02], and the bias term is initialized to 0. Then a two-stage training strategy is used. In the first stage, the model is trained using a self-supervised learning method to learn the internal representation of the expression sequence and eye movement sequence by designing a time mask prediction task. The mask proportion is set to 15%, i.e., randomly masking 15% of the time steps in the input sequence, and the model is asked to predict these masked features. The loss function uses mean square error (MSE). The first stage of training is 50 epochs, with a batch size of 32. In the second stage, the model is fine-tuned using a supervised learning method, with the student attention state label as the supervision signal to optimize the model parameters. The loss function uses a weighted combination of cross-entropy loss and mean square error, with a weight ratio of 0.7:0.3. During the training process, the cosine annealing learning rate scheduling strategy is used, with an initial learning rate of 0.001 and a minimum learning rate of 10 -6, the performance on the validation set is gradually reduced during the training process. At the same time, the gradient clipping technique is used to prevent gradient explosion, and the upper limit of the gradient norm is set to 5.0. To prevent overfitting, in addition to using L2 regularization (weight decay coefficient is 10 -4 ), a Dropout layer is also added after each transformer layer, with a dropout rate of 0.1. The early stopping strategy is adopted to avoid overfitting, and the training is stopped when the performance indicators on the validation set do not improve for 5 consecutive epochs. The entire training process is carried out in parallel on four GPUs, using data parallel strategy, with a batch size of 32 for each GPU, and a total of 100 epochs of training, saving a model checkpoint every 10 epochs, and finally selecting the model with the best performance on the validation set as the final model. During the training process, multiple performance indicators are monitored, including accuracy, precision, recall, F1 score, etc., to comprehensively evaluate the classification and regression performance of the model.

[0065] Step S09 is an optional step, and its specific implementation is to carry out class learning state clustering analysis and teaching feedback system construction. The improved K-means++ clustering algorithm is used to analyze the dynamic portrait feature map of all students in the class, and the clustering number K is adaptively determined by the silhouette coefficient method, usually between 3-5. The clustering feature dimensions include attention state score, expression change frequency, eye concentration index, etc. Core indicators, and the feature vector is standardized by Z-score to eliminate dimensional differences. Calculate the distance between each cluster center and the ideal learning state (high attention, moderate expression change, high eye concentration), define the cluster with the smallest distance as the best learning state group, and the cluster with the largest distance as the group to be improved. Calculate the proportion of the number of students in each cluster to form a class learning state distribution graph. Based on the clustering analysis results, calculate the classroom interaction effect evaluation indicators, including the class average attention state score (value range 0-100), attention synchronization index (defined as the standard deviation of the attention state scores of all students in the class, the smaller the better), and emotion distribution balance (defined as the entropy value of the proportion of each emotion, the larger the better). Set the evaluation index threshold, the average attention state score threshold is 70 points, the attention synchronization index threshold is 15, and the emotion distribution balance threshold is 0.8. When the indicators are lower than the threshold, trigger the corresponding teaching adjustment suggestions, such as increasing the interaction link, adjusting the teaching rhythm, and refining the difficult point explanation, etc. Establish a teaching suggestion pushing mechanism to provide real-time teaching optimization suggestions through the teacher's display screen without interfering with normal teaching, including class overall state analysis, problem area identification, improvement strategy recommendation, etc. to assist teachers to adjust teaching strategies in a timely manner, and realize closed-loop optimization of the teaching process.

[0066] The second aspect of the present application provides a computer readable storage medium, wherein program instructions are stored in the computer readable storage medium, and the program instructions are used to execute the above-mentioned interactive classroom student dynamic portrait generation method when running in a computer.

[0067] The third aspect of the present application provides an interactive classroom student dynamic portrait generation system, comprising the above-mentioned computer readable storage medium, the system being any one of a computer, a server, and a single-chip microcomputer, the computer readable storage medium being arranged in the system, and the system being provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0068] The mathematical model or calculation process involved in the present application is described in detail below.

[0069] The process of collecting a student face image sequence in step S01 and extracting face key point data involves the construction of an expression feature matrix. The specific representation of the expression feature matrix is as follows:

[0070]

[0071] In the formula, F is a sequence of expression feature matrices; F t is the expression feature matrix of the t-th frame; and n is the total number of video frames.

[0072] The construction formula of a single-frame expression feature matrix is as follows:

[0073]

[0074] In the formula, (x i，t , y i，t ) is the normalized coordinate of the i-th feature point in the t-th frame image; and 68 represents the total number of face feature points.

[0075] The calculation formula of coordinate normalization processing is as follows:

[0076]

[0077] In the formula, (x i，t ′, y i，t ′) is the normalized coordinate; (x c，t , y c，t ) is the face center point coordinate; d t is a face scale factor, usually taking the face width.

[0078] The Kalman filter algorithm is used for feature point trajectory smoothing, and the state equation and observation equation are as follows:

[0079] s t+1 = As t +w t ;

[0080] z t = Hs t + v t ;

[0081] wherein, is a state vector, containing feature point positions and velocities; z t = [x i，t , y i，t ] T is an observation vector; A is a state transition matrix; H is an observation matrix; w t is process noise, obeying a Gaussian distribution with zero mean and covariance Q; v t is observation noise, obeying a Gaussian distribution with zero mean and covariance R.

[0082] The feature point displacement vector calculation formula is:

[0083] d i，t = [x i，t+1 -x i，t , y i，t+1 -y i，t ];

[0084] wherein, d i，t is the displacement vector of the i-th feature point from time t to time t+1.

[0085] The calculation of the expression similarity vector in step S02 involves similarity calculation with the standard expression library. The specific representation of the expression similarity vector is as follows:

[0086] S t = [s t，1 , s t，2 ,..., s t，10 ] T ;

[0087] wherein, S t is the expression similarity vector at time t; s t，j is the similarity of the expression at time t with the j-th expression in the standard expression library, with a value range of [0, 1].

[0088] The calculation of the expression similarity adopts a weighted cosine similarity, and the calculation formula is:

[0089]

[0090] wherein, p i，t is the position vector of the i-th feature point at time t; p i，j is the position vector of the i-th feature point of the j-th expression in the standard expression library; w iis the weight coefficient of the ith feature point, the weight of eye and mouth feature points is higher than other regions, and the value range is [0.5, 2.0].

[0091] The expression change frequency calculation formula is:

[0092]

[0093] In the formula, F is the expression change frequency; D(S t , S t+1 ) is the Euclidean distance of the adjacent time window expression similarity vector; δ(·) is an indicator function, which takes a value of 1 when the condition is met, otherwise 0; τ d is the expression change threshold, which takes a value of 0.3; T is the number of time windows.

[0094] The Euclidean distance calculation formula of the expression similarity vector is:

[0095]

[0096] The calculation of the eye gaze concentration index in step S03 involves processing and analysis of eye movement data. The fixation point recognition uses the I-VT algorithm, and the calculation formula is:

[0097]

[0098] In the formula, v t is the eye movement speed at time t; (x t , y t ) is the gaze position at time t; Δt is the sampling time interval, and for an eye tracker with a sampling rate of 120Hz, Δt = 1 / 120 seconds.

[0099] When v t < τ v , it is determined as a fixation state, where τ v is the speed threshold, which takes a value of 30° / s.

[0100] The gaze heat map generation uses a two-dimensional Gaussian kernel function, and the calculation formula is:

[0101]

[0102] In the formula, H(x, y) is the heat value at coordinates (x, y); (x i , y i ) is the coordinate of the ith fixation point; t i is the duration of the ith fixation point; σ is the standard deviation of the Gaussian kernel function, corresponding to the pixel value of the viewing angle 1.5°; N is the total number of fixation points.

[0103] The eye gaze concentration index calculation formula is:

[0104]

[0105] In the formula, E is the eye concentration index, and the value range is [0, 1]; Ω is the region in which the heat value in the heat map exceeds the threshold value; T f is the total time of the fixation state; T is the total observation time.

[0106] In step S05, the Mahalanobis distance is used to measure the deviation of the current expression of the student from the baseline or the average state of the class, and the calculation formula is as follows:

[0107]

[0108] In the formula, D M (F t , F b ) is the Mahalanobis distance between the expression feature matrix F t and the baseline expression matrix F b ; Σ is the covariance matrix, reflecting the correlation between the features.

[0109] The calculation formula of the specific expression frequency is as follows:

[0110]

[0111] In the formula, f a is the specific expression frequency; δ(·) is the indicator function; τ m is the Mahalanobis distance threshold, and the value is 2.5; T is the length of the time window.

[0112] In step S07, the state transition function is based on the hidden Markov model, and the calculation formulas of the state transition matrix and the observation probability matrix are as follows:

[0113] P ij (S t+1 t+1 t =j|S ij t t =i);

[0114] In the formula, P ij is the state transition matrix element, representing the probability of transitioning from state i to state j; S t represents the attention state at time t.

[0115] B ij (O t t t =j|S ij t t =i);

[0116] In the formula, B ij is the observation probability matrix element, representing the probability of observing the feature vector j in state i; O t represents the observation vector at time t.

[0117] The observation vector contains multiple features, denoted as:

[0118] O t =[S t , E t , f a，t , A t-1 , d t ];

[0119] In the formula, S t is the expression similarity vector; E t is the eye concentration index; f a，t is the specific expression frequency; A t-1 is the attention state score of the previous moment; and d t is the difficulty coefficient of the classroom content.

[0120] The calculation formula of the attention state score is:

[0121]

[0122] In the formula, A t is the attention state score at moment t, with a value range of [0, 100]; α i is the score weight corresponding to state i, α1=10, α2=30, α3=50, α4=70, and α5=90; P(S t |i|O 1：t ) is the posterior probability of state i at moment t under the condition of a given observation sequence O 1：t , which is calculated by the forward-backward algorithm.

[0123] The calculation formula of the dynamic threshold adjustment mechanism is:

[0124]

[0125] In the formula, τ A (d t ) is the adjusted attention state threshold; θ t is the basic threshold, with a value of 60; β is the adjustment coefficient, with a value of 0.3; and d k is the difficulty coefficient of the classroom content, with a value range of [0, 1].

[0126] The multi-head attention mechanism for learning the state evaluation network model in step S08 involves the calculation of the number of heads, dimensions, and layer depth, and the calculation formula is as follows:

[0127]

[0128] In the formula, h is the number of heads of the multi-head attention mechanism; F is the expression change frequency; and represents the floor operation.

[0129]

[0130] where d is the dimension of each attention head; E0 is the eye concentration threshold, which is 0.6.

[0131]

[0132] where L is the depth of the attention layer; R is the attention fluctuation rate.

[0133] The self-attention calculation formula of the transformer network is:

[0134]

[0135] where Q is the query matrix; K is the key matrix; V is the value matrix; d k is the dimension of the key vector.

[0136] The calculation formula of multi-head attention is:

[0137] MultiHead(Q, K, V)=Concat(head1, …, head h )W O ;

[0138] where and W O are learnable weight matrices.

[0139] The calculation of the classroom interaction effect evaluation index in step S09 involves cluster analysis and statistical index calculation, which is specifically represented as follows:

[0140] The calculation formula of the attention synchronization index is:

[0141]

[0142] where ASI is the attention synchronization index; A i is the attention state score of the i-th student; is the average attention state score of the class; N is the total number of students.

[0143] The calculation formula of the emotion distribution balance degree is:

[0144]

[0145] where EDB is the emotion distribution balance degree; p j is the proportion of the j-th emotion in the class; 10 is the total number of emotion categories.

[0146] The K-means++ algorithm is used for cluster analysis, and the objective function is:

[0147]

[0148] In the formula, J is the clustering objective function; x i Let μ be the feature vector of the i-th student; j Let K be the j-th cluster center; K be the number of clusters; and N be the total number of students.

[0149] The formula for calculating the optimal number of clusters K using the silhouette coefficient method is as follows:

[0150]

[0151] In the formula, s(i) is the silhouette coefficient of the i-th sample; a(i) is the average distance between sample i and other samples in the same cluster; b(i) is the average distance between sample i and the nearest sample in a different cluster.

[0152] The formula for calculating the average profile coefficient is:

[0153]

[0154] In the formula, The average profile coefficient has a value range of [-1, 1]. A larger value indicates a better clustering effect.

[0155] The formula for calculating the distance between the ideal learning state and the cluster center is:

[0156]

[0157] In the formula, D(c j c * c is the j-th cluster center. j Compared to the ideal learning state c * The weighted Euclidean distance between them; c j，k The k-th feature of the j-th cluster center; The k-th feature of the ideal learning state; w k d represents the weight of the k-th feature; d represents the feature dimension.

[0158] Most of the parameters involved in these formulas were obtained experimentally. For example, the eye focus threshold E0 was set to 0.6 based on eye movement data analysis; values ​​above this were considered a state of focused attention. The Mahalanobis distance threshold τ... m The threshold was set at 2.5 based on statistical analysis; values ​​exceeding this threshold were marked as unusual facial expressions. Attention state scoring threshold. Based on the experience of educational experts, a value of 60 was set as the baseline for judging students' attention levels. Other parameters, such as the weight of facial feature points w, were also considered. i The state transition matrix P and the observation probability matrix B need to be determined through statistical analysis and machine learning methods using a large amount of historical data.

[0159] Specifically, the principle of the present application is that the core technical principle of the present application is to establish a multi-level and multi-dimensional student attention evaluation framework. First, the facial key point data of the student is extracted through the facial feature point positioning technology, an expression feature matrix is constructed, the positions and changes of the key points such as eyebrows, eyes and corners of the mouth are captured, and the expression similarity vector is calculated by comparison with the standard expression library to quantify the emotional state of the student.

[0160] At the same time, the eye tracking technology is used to collect the student's visual focus coordinates, calculate the eye concentration area and residence time, generate a gaze heat map, and quantitatively represent the stability and distribution characteristics of the student's visual attention. This dual-channel data acquisition mechanism ensures comprehensive perception of the student's attention state.

[0161] The innovation of the present application lies in the introduction of deep learning to construct an individual correlation model, which solves the individual difference problem of the expression-attention mapping relationship of different students. Through the multi-layer bidirectional transformer network architecture, the time sequence feature extraction and fusion of the expression feature matrix and the eye concentration degree index are realized, and the multi-head attention mechanism parameters are dynamically determined by the expression change frequency, the eye concentration degree threshold and the attention fluctuation rate, so that the model can adapt to the behavior patterns of different students.

[0162] Another core innovation is to apply a state transition function to dynamically model the student's attention fluctuation, which comprehensively considers the expression similarity vector, the eye concentration degree index, the specific expression appearance frequency, the attention history value, the classroom content difficulty coefficient and other multi-dimensional parameters, and realizes continuous tracking and prediction of the student's attention state through recursive calculation, overcoming the limitations of static evaluation methods.

[0163] Through the organic combination of these technical principles, the present application realizes precise, dynamic and personalized evaluation of the student's attention state, and provides a scientific basis for optimizing classroom interaction.

[0164] A specific embodiment 1 of the present application is provided below, and the specific implementation manner of each step in the embodiment 1 is described in detail as follows.

[0165] The specific implementation of step S01 is to use a deep convolutional neural network model to detect and track the facial key points of students. First, a high-definition camera is used to collect video of students in the classroom, with a frame rate of 30 frames per second and a resolution of 1080P to ensure clear and identifiable facial features. After video collection is complete, a face detection algorithm is used to locate the face region in each frame of image, and an improved MTCNN (Multi-Task Convolutional Neural Network) algorithm is used to detect the face region and the positions of eyes, nose, mouth and other main organs at the same time, with a detection accuracy of more than 98%. The improved 68-point facial feature point positioning algorithm in the Dlib library is applied to the detected face region to extract a complete set of facial feature points including eyebrows (6 points on the left and right), eyes (6 points on the left and right), nose (9 points), mouth (20 points) and contour (17 points). The extracted facial feature points form an expression feature matrix sequence, denoted as:

[0166]

[0167] In the formula, F is the expression feature matrix sequence; F t is the expression feature matrix of the t-th frame; and n is the total number of video frames. A single-frame expression feature matrix is constructed as:

[0168]

[0169] In the formula, (x i，t , y i，t ) is the normalized coordinate of the i-th feature point in the t-th frame of image. The coordinate normalization process uses the following formula:

[0170]

[0171] In the formula, (x i，t ', y i，t ') is the normalized coordinate; (x c，t , y c，t ) is the center point coordinate of the face; and d t is the face scale factor. The Kalman filter algorithm is used to smooth the feature point trajectory, and the state equation and observation equation are as follows:

[0172] s t+1 = As t +w t ;

[0173] z t = Hs t +v t ;

[0174] In the formula, is the state vector; and z t = [x i，t , y i，t] T A is the observation vector; A is the state transition matrix; H is the observation matrix; w t For process noise; v t To account for observation noise, the final step is to calculate the feature point displacement vector:

[0175] d i，t =[x i，t+1 -x i，t y i，t+1 -y i，t ];

[0176] In the formula, d i，t Let be the displacement vector of the i-th feature point from time t to time t+1.

[0177] The specific implementation of step S02 involves constructing a framework for calculating and analyzing facial expression similarity. First, a standard facial expression library is established, containing standard expression templates corresponding to 10 typical learning states, such as focus, confusion, understanding, boredom, and fatigue. Principal component analysis is used to reduce the dimensionality of facial feature point data and extract the main patterns of expression changes. For the facial expression feature matrix extracted from each frame of a student's facial image, its cosine similarity with each template in the standard facial expression library is calculated to form an expression similarity vector.

[0178] S t =[s t，1 s t，2 , ..., s t，10 ] T ;

[0179] In the formula, S t Let s be the facial expression similarity vector at time t; t，j Let be the similarity between the expression at time t and the j-th type of expression in the standard expression library. The expression similarity is calculated using weighted cosine similarity:

[0180]

[0181] In the formula, p i，t p is the position vector of the i-th feature point at time t; i，j w is the position vector of the i-th feature point of the j-th type of expression in the standard emoji library; i Let be the weight coefficient for the i-th feature point. Statistical analysis of facial expression change frequency:

[0182]

[0183] In the formula, F represents the frequency of facial expression changes; D(S) t S t+1 ) represents the Euclidean distance between the expression similarity vectors of adjacent time windows; δ(·) is the indicator function; τ dThe expression change threshold is 0.3, and T is the number of time windows. The Euclidean distance is calculated as:

[0184]

[0185] The specific implementation of step S03 is to realize an eye movement tracking and gaze analysis system. A high-precision eye movement tracking device (with a sampling rate of 120 Hz) is used to collect student eye movement data. After pre-processing the eye movement data, the I-VT (velocity threshold recognition) algorithm is used to identify the fixation points and saccadic movements in the gaze trajectory:

[0186]

[0187] where v t is the eye movement speed at time t, (x t , y t ) is the gaze position at time t, and Δt is the sampling time interval. The identified fixation points form a sequence of gaze focal point coordinates, and the DBSCAN algorithm is used for cluster analysis of the fixation points to identify the eye light concentration area. A gaze heat map is generated:

[0188]

[0189] where H(x, y) is the heat value at coordinates (x, y), (x i , y i ) is the coordinates of the i-th fixation point, t i is the duration of the i-th fixation point, σ is the standard deviation of the Gaussian kernel function, and N is the total number of fixation points. The eye light concentration index E is calculated as:

[0190]

[0191] where E is the eye light concentration index, Ω is the area in the heat map where the heat value exceeds the threshold, T f is the total time in the fixation state, and T is the total observation time.

[0192] The specific implementation of step S04 is the same as described above, and will not be repeated here.

[0193] The specific implementation of step S05 is to perform specific expression recognition and learning state anomaly detection. A student individual expression baseline and a class average expression state reference model are established. The Mahalanobis distance is used to measure the degree of deviation of the student's current expression from the baseline or the class average state:

[0194]

[0195] where D M (F t , F b ) is the expression feature matrix Ft Mahalanobis distance between the baseline expression matrix F b ;∑ is the covariance matrix. Calculate the specific expression frequency:

[0196]

[0197] where f a is the specific expression frequency; τ m is the Mahalanobis distance threshold, taking a value of 2.5; T is the length of the time window. Analyze the specific expression type and establish the time correspondence with the classroom content.

[0198] The specific implementation of step S06 is the same as the foregoing, and will not be described here again.

[0199] The specific implementation of step S07 is to realize the state transition function and the attention state evaluation system. The attention state transition function based on the hidden Markov model is constructed, and the attention state space S = {S1, S2, S3, S4, S5} is defined. The state transition matrix and the observation probability matrix are calculated as:

[0200] P ij = P(S t+1 = j | S t = i);

[0201] B ij = P(O t = j | S t = i);

[0202] where P ij is the state transition matrix element; B ij is the observation probability matrix element; S t represents the attention state at time t; O t represents the observation vector at time t. The observation vector contains multiple features:

[0203] O t = [S t , E t , f a，t , A t-1 , d t ];

[0204] where S t is the expression similarity vector; E t is the eye concentration index; f a，t is the specific expression frequency; A t-1 is the attention state score of the previous time; d t is the difficulty coefficient of the classroom content. Calculate the attention state score:

[0205]

[0206] wherein A t is the attention state score at time t; a i is the weight of state i; P(S t | O 1：t ) is the posterior probability of being in state i at time t given the observation sequence O 1：t . The formula of the dynamic threshold adjustment mechanism is:

[0207]

[0208] wherein τ A (d t ) is the adjusted attention state threshold; is the base threshold, which is 60; β is the adjustment coefficient, which is 0.3; d t is the difficulty coefficient of the classroom content.

[0209] The specific implementation of step S08 is the same as the foregoing, and will not be described here again.

[0210] The specific implementation of step S09 is to carry out class learning state clustering analysis and teaching feedback system construction. The improved K-means++ clustering algorithm is used to analyze the dynamic portrait feature map of all students in the class, and the clustering number K is adaptively determined by the silhouette coefficient method. The attention synchronization index is calculated:

[0211]

[0212] wherein ASI is the attention synchronization index; A i is the attention state score of the i-th student; is the average attention state score of the class; N is the total number of students. The emotion distribution balance degree is calculated:

[0213]

[0214] wherein EDB is the emotion distribution balance degree; p j is the proportion of the j-th emotion in the class. The objective function of the clustering analysis is:

[0215]

[0216] wherein J is the clustering objective function; x i is the feature vector of the i-th student; μ j is the j-th clustering center; K is the number of clusters. The silhouette coefficient is calculated as:

[0217]

[0218] In the formula, s(i) is the silhouette coefficient of the i-th sample; a(i) is the average distance between sample i and other samples in the same cluster; b(i) is the average distance between sample i and samples in the nearest different clusters. The distance between the ideal learning state and the cluster center is calculated as follows:

[0219]

[0220] In the formula, D(c j c * c is the j-th cluster center. j Compared to the ideal learning state c * The weighted Euclidean distance between them; c j，k The k-th feature of the j-th cluster center; The k-th feature of the ideal learning state; w k Let be the weight of the k-th feature. Based on the cluster analysis results, calculate the evaluation index for classroom interaction effectiveness, set the threshold for the evaluation index, and construct a teaching suggestion push mechanism.

[0221] Optionally, the Kalman filter algorithm plays a key role in smoothing facial feature point trajectories, and its state equation s t+1 =As t +w t The motion model of the feature points is described, and the observation equation z t =Hs t +v t The algorithm connects actual observations with state variables. Through a recursive prediction-update process, it effectively filters out noise caused by factors such as changes in illumination and minute head movements, ensuring the continuity and stability of feature point trajectories. (State vector) It also includes position and velocity information, enabling the algorithm to predict the future location of feature points and adapt to rapid changes in facial expressions.

[0222] Optional, facial expression similarity vector S t =[s t，1 s t，2 , ..., s t，10 ] T This reflects the degree of similarity between a student's current facial expression and various expressions in a standard facial expression database, forming the basis of emotion state analysis. Weighted cosine similarity formula. By introducing the feature point weight coefficient w i This study emphasizes the importance of key facial areas (such as the eyes and mouth) in emotional expression, thus improving the accuracy of facial expression recognition. The weighting coefficients are determined based on facial anatomy and the rules of emotional expression; the weights for the eyes and mouth are typically 1.5–2.0, while other areas are 0.5–1.0.

[0223] Optional, frequency of facial expression changes The fluctuation degree of emotional state is an important indicator of students' cognitive and emotional engagement. The Euclidean distance between adjacent time window expression similarity vectors is calculated, and a threshold τ d = 0.3 is used to determine whether it constitutes an effective expression change. The indicator function δ(·) converts the continuous distance value into a discrete count, which is convenient for statistical analysis.

[0224] Optionally, the speed threshold identification (I-VT) algorithm in eye movement analysis calculates the speed of visual line movement to distinguish between fixation state and saccade movement. The speed threshold τ v = 30° / s is determined based on the physiological characteristics of the human eye, and eye movements below this threshold are classified as fixation, indicating that the student is extracting information and processing cognition in the specified area; above this threshold, it is classified as saccade, representing the rapid transfer process of the visual line.

[0225] Optionally, the gaze heat map visually presents the distribution of visual attention and is the core tool for eye concentration analysis. The formula is based on a two-dimensional Gaussian kernel function, which spreads each fixation point into a hot spot area, with fixation time t i as the weight coefficient, reflecting the intensity of fixation; the standard deviation σ corresponds to the pixel value of a visual angle of 1.5°, usually between 30 and 50 pixels, determining the diffusion range of the hot spot. High-intensity areas on the heat map represent the focus of visual attention, making it easy to identify the teaching content that the student is paying attention to.

[0226] Optionally, the eye concentration index considers both spatial concentration and temporal stability. Spatial concentration is reflected by the proportion of high-intensity areas Ω on the heat map, and temporal stability is represented by the proportion of fixation state time . The index takes values in the range [0, 1], with a larger value indicating more concentrated fixation. In practical applications, the eye concentration threshold E0 = 0.6 is used as the basis for judging the state of attention concentration.

[0227] Optionally, the Mahalanobis distance has unique advantages in specific expression recognition. It considers the correlation between features (through the covariance matrix Σ), which can accurately capture abnormal states in a multi-dimensional feature space. Compared with the Euclidean distance, the Mahalanobis distance is not sensitive to feature scales, can balance the contribution of different facial regions, and improve the accuracy of anomaly detection. The frequency of specific expressions is quantified by counting the proportion of time when the Mahalanobis distance exceeds the threshold τ m = 2.5, which quantifies the degree of students' cognitive conflict or emotional fluctuations.

[0228] Optionally, the hidden Markov model is applied in attention state evaluation, through state transition matrix P ij =

[0229] P(S t+1 =j|S t =i) and observation probability matrix B ij =P(O t =j|S t =i), the probability relationship between observation characteristics and potential attention state is established. This model considers the time sequence dependence of attention state, avoids frequent state jumps due to instantaneous feature fluctuations, and enhances the stability and reliability of the evaluation results. The observation vector O t =[S t , E t , f a，t , A t-1 , d t ] fuses multi-dimensional information such as expression, eye movement, and historical state, fully capturing various aspects of student cognitive engagement.

[0230] Optionally, attention state score converts the discrete state probability distribution into a continuous score value, and α i is the score weight corresponding to each state, which is determined based on educational psychology research. Dynamic threshold adjustment mechanism takes into account the impact of content difficulty on attention evaluation criteria. When the content difficulty coefficient d t increases, the evaluation threshold is appropriately reduced, reflecting the application of cognitive load theory.

[0231] Optionally, the parameter design of the multi-head attention mechanism in the learning state evaluation network reflects a deep consideration of individual characteristics of students. The number of heads is related to the expression change frequency F, and a higher expression change frequency corresponds to more attention heads, enhancing the model's ability to capture complex expression change patterns. The dimension of each attention head is related to the eye concentration threshold E0, reflecting the importance of eye movement features in attention evaluation; the depth of the attention layer is related to the attention fluctuation rate R, and a higher fluctuation rate corresponds to a deeper hierarchical structure, enhancing the model's ability to handle unstable attention states. This adaptive parameter design allows the model to dynamically adjust the network structure according to the student's behavioral characteristics, improving the accuracy and individualization level of the evaluation.

[0232] Optionally, the attention synchronization index ASI is the standard deviation of students' attention state scores within a class, quantifying the consistency of the class learning state. A lower ASI value indicates high synchrony, reflecting that the teaching content or activity can attract the attention of most students simultaneously; a higher ASI value indicates low synchrony, which may mean that the teaching difficulty is uneven or some students cannot keep up with the teaching pace. This indicator provides a quantitative assessment of the overall engagement of the class for teachers, with a threshold set at 15, which should be considered for adjusting teaching strategies if exceeded.

[0233] Optional, Emotion Distribution Balance Degree Using the concept of information entropy, the diversity and balance of the class emotion state are measured. j EDB represents the proportion of the jth emotion in the class, and a high EDB value corresponds to a uniform emotion state, while a low EDB value corresponds to a concentrated distribution. The ideal range of this indicator is 0.7-0.9, and a too low value indicates a single emotion (which may be negative or positive), while a too high value may indicate a chaotic classroom atmosphere and lack of clear emotional direction. Combined with the ASI and EDB indicators, the effectiveness of classroom interaction can be comprehensively evaluated.

[0234] Optional, K-means++ clustering algorithm objective function By minimizing the sum of the squares of the distances from each sample to the nearest cluster center, automatic grouping of student learning states is achieved. Compared with the traditional K-means algorithm, K-means++ improves the selection strategy of the initial center point, avoids local optimal solution, and improves the clustering quality. As an evaluation indicator of clustering effectiveness, it helps to determine the optimal number of clusters K. An average silhouette coefficient close to 1 indicates good clustering results, with each cluster being tight and separated from each other; close to 0 indicates that the sample is near the cluster boundary; and a negative value indicates possible misallocation.

[0235] Optional, Distance between ideal learning state and cluster center The weighted Euclidean distance quantifies the difference between each cluster and the best learning state. The weight coefficient w k Reflects the importance of different feature dimensions, with the attention state score weight being the highest (about 0.5), followed by the expression change frequency and eye concentration degree indicators (each about 0.25). This distance measure helps to identify the best learning state group and the group to be improved, providing teachers with targeted teaching intervention basis.

[0236] Optional, The key difference between the expression feature matrix extraction technology and the traditional facial expression recognition method is that the traditional method is mostly based on static expression image analysis, while the expression feature matrix It contains complete timing information and can capture the dynamic changes of facial expressions. This dynamic expression analysis method is more in line with the real characteristics of human emotional expression, effectively avoids the ambiguity problem in static expression recognition, and improves the accuracy of emotional state judgment.

[0237] Optional weighted cosine similarity formula Compared with the traditional cosine similarity, the feature point weight coefficient w is introduced i This improvement is derived from the study of facial expression, which shows that different facial regions contribute unequally to emotional expression. The eyes and mouth play a leading role in most basic emotional expressions, so they are given a higher weight (usually 1.5-2.0), while the weight of the facial contour and other regions is lower (usually 0.5-1.0). This weighting strategy significantly improves the sensitivity and accuracy of expression recognition, especially for subtle emotional changes in the classroom environment.

[0238] Further, the eye concentration index calculation formula Innovatively combines spatial concentration and temporal stability in two dimensions. Traditional eye movement analysis focuses on fixation point distribution or fixation time, but ignores the comprehensive effect of these two dimensions. Spatial concentration reflects the size of the gaze focus range, and temporal stability reflects the degree of gaze duration. Both determine the depth of cognitive investment. The multiplication of the two parts rather than simple addition reflects the mutual enhancement of the two dimensions. Both spatial concentration and temporal stability are required to determine a state of high concentration.

[0239] Further, the application of Hidden Markov Model in attention state evaluation reflects a deep understanding of the dynamic change law of attention. Attention state has obvious time sequence dependence, that is, the current state is affected by the previous state, and it will not jump frequently in a short time. The state transition matrix P ij = P(S t+1 = j | S t = i) captures this timing rule, making the evaluation result more smooth and stable. The observation vector O t = [S t , E t , f a，t , A t-1 , d t ] integrates multiple heterogeneous features, fully reflects the cognitive and emotional state of students, and avoids the one-sidedness that a single indicator may bring.

[0240] Further, the dynamic threshold adjustment mechanism Based on the cognitive load theory, the impact of classroom content difficulty on attention performance is considered. When the difficulty of learning materials increases, students need to invest more cognitive resources for information processing, and the external characteristics of attention performance may be weakened, such as increased blink frequency, tense facial expressions, etc. By dynamically reducing the judgment threshold, the problem of incorrect judgment of students' attention state due to content difficulty is avoided, making the evaluation result more fair and reasonable.

[0241] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: researchers applied the interactive classroom student dynamic portrait generation method in a middle school eighth grade mathematics classroom, and observed and analyzed 32 students in a class for 4 weeks. The class is equipped with a high-definition camera and an eye tracking device, the camera has a frame rate of 30 frames per second and a resolution of 1080P; the eye tracking device has a sampling rate of 120Hz. Before the start of the study, the researchers collected the students' basic data and established the initial expression baseline. During the classroom observation, the system collects student facial image sequences and eye movement data in real time, constructs expression feature matrix and gaze focus coordinate sequence, and then generates student dynamic portrait feature map.

[0242] In a lesson about quadratic functions, the researchers recorded the changes in students' attention states during the entire teaching process in detail. The course is divided into five stages: review introduction (10 minutes), new concept explanation (15 minutes), example analysis (15 minutes), group discussion (15 minutes), and summary test (10 minutes). The system constructs the expression feature matrix F t by extracting the data of 68 key points on the students' faces. For a representative student Zhang, the calculation results of his expression similarity vectors at different times are shown in Table 1:

[0243] Table 1 Expression similarity vectors of student Zhang at different teaching stages

[0244]

[0245] By analyzing the expression similarity vector data, the system identifies that Zhang has higher confusion expressions (similarities are 0.79 and 0.83 respectively) during the new concept explanation and example analysis stages, which are significantly higher than the threshold value of 0.75, indicating that he has difficulty understanding the concept of quadratic functions. At the same time, the frequency of expression changes F reaches the highest value of 3.6 times per minute during the example analysis stage, which is much higher than the average value of 2.1 times per minute in the class, further confirming the confusion state.

[0246] In terms of eye movement analysis, the system identifies Zhang's fixation points and saccadic movements through the I-VT algorithm, and the calculated eye concentration index E is shown in Table 2:

[0247] Table 2 Eye movement characteristic data of student Zhang in different teaching stages

[0248]

[0249]

[0250] Table 2 shows that the eye concentration index of Zhang in the new concept explanation and example analysis stage is 0.52 and 0.48 respectively, which is lower than the threshold value 0.6, indicating that the attention is not concentrated. Combined with the heat map analysis, it is found that the line of sight jumps frequently between the blackboard and the teaching material, and cannot form effective fixation on the key formula and graph.

[0251] The researchers use Mahalanobis distance formula to calculate the deviation of Zhang's expression feature matrix from the personal baseline, and identify the specific expression. In the 8th minute of the example analysis stage, when the teacher explains the calculation method of the maximum value of the quadratic function, the Mahalanobis distance value is detected to be 3.2, which exceeds the threshold value 2.5, and the system marks it as a specific expression point. Through analysis of the corresponding relationship of the expression similarity vector, it is determined to be a composite emotion of "confusion + surprise", indicating that this knowledge point may be the learning obstacle of Zhang.

[0252] Based on the state transition function of the hidden Markov model, the system integrates the expression features, eye movement data and historical state to calculate the attention state score A of Zhang in the whole course t As shown in Table 3:

[0253] Table 3 Attention state score change of student Zhang and comparison with class average level

[0254] Time point (minutes) Zhang attention state score Class average attention state score Attention state evaluation 5 78 72 High 15 56 68 Medium 25 43 65 Low 35 61 74 Medium 45 75 76 High 55 82 79 Extremely high

[0255] Table 3 data shows that the attention state of Zhang decreased significantly in the new concept explanation and example analysis stage (15-25 minutes), and the gap with the class average level widened, and then rebounded in the group discussion and summary test stage, finally reached a very high level. This is highly consistent with the analysis results of the expression similarity vector and the eye concentration index.

[0256] The researchers use a multi-layer bidirectional transformer network to learn the state evaluation, and the network parameters are set as follows: the number of multi-head attention heads h = 6 (calculated according to the expression change frequency F = 2.8), the dimension of each attention head d = 38 (calculated according to the eye concentration threshold E0 = 0.6), and the attention layer depth L = 7 (calculated according to the attention fluctuation rate R = 0.18). The model training adopts a two-stage strategy, the first stage is self-supervised learning for 50 epochs, and the second stage is supervised fine-tuning for 100 epochs, the initial learning rate is 0.001, and the batch size is 32.

[0257] The dynamic portrait feature map of 32 students in a class was analyzed by K-means++ clustering algorithm, and the optimal cluster number K was determined as 4 by contour coefficient method. The clustering results and the characteristics of each group are shown in Table 4.

[0258] Table 4 Clustering analysis results of learning state of students in a class

[0259]

[0260]

[0261] Zhang is classified as group C (to be promoted group), and the system provides targeted teaching suggestions for teachers: (1) focus on the difficulty of the maximum value calculation of quadratic function; (2) increase the intuitive graphics to assist understanding; (3) arrange group cooperation learning between group A and group C students. The teacher adjusts the teaching according to the suggestion, and in the subsequent course, the attention state score of Zhang is significantly improved, with an average of 72.5 points, the expression change frequency is reduced to 2.0 times per minute, and the eye concentration degree index is increased to 0.68.

[0262] Compared with the traditional student attention evaluation method, the interactive classroom student dynamic portrait generation method shown in the embodiment has obvious advantages. The traditional method mainly relies on teacher subjective observation or simple behavior counting (such as the number of hands raised, the frequency of nodding, etc.), which cannot capture subtle expression changes and eye movement characteristics, and the evaluation result is subjective and time-consuming. And the method extracts 68 key point data through facial feature point positioning technology, combines eye tracking and multi-dimensional feature fusion, realizes objective, accurate and real-time evaluation of student attention state. Especially, the Mahalanobis distance is used to identify specific expressions and state transition function modeling attention fluctuation, which successfully solves the problem that the traditional method cannot quantify emotional changes and predict attention trends. K-means++ clustering analysis enables teachers to quickly identify different learning state groups and achieve precise teaching intervention. The implementation results show that the teaching adjustment based on the method makes the attention state score of the target student improve by an average of 25.9%, and the learning effect is significantly improved, which confirms the practical value of the method in improving teaching quality and promoting personalized education.

[0263] It should be noted that the variables involved in the present application are explained in detail as shown in Tables 5 and 6.

[0264] Table 5 Variable explanation table (first part)

[0265]

[0266]

[0267] Table 6 Variable explanation table (second part)

[0268]

[0269] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An interactive classroom student dynamic image generation method, characterized by, The application relates to a learning state evaluation method and system. A sequence of facial images of students in a classroom is collected, and facial key point data is extracted to construct an expression feature matrix; The similarity of the expression feature matrix to a standard expression library is calculated to generate an expression similarity vector, the frequency of expression changes is recorded, and an expression dynamic change curve is established; a sequence of eye focus coordinates is collected, the eye concentration area and residence time are calculated, and the eye concentration degree index is quantified; An individual correlation model is established; the expression feature matrix of students at different time periods in the classroom is statistically analyzed, and a specific expression deviating greatly from the average expression state in the classroom is identified and marked as a learning state abnormal point; The correlation between the attention concentration degree and the content nodes in the classroom is analyzed, and an attention fluctuation curve is constructed; The attention fluctuation is modeled to generate an attention state score; A learning state evaluation network model is used to fuse multi-dimensional features to generate a dynamic portrait feature map, which is used as a student dynamic portrait; the specific structure of the learning state evaluation network model is a multi-layer bidirectional transformer network architecture, which comprises an encoder layer and a decoder layer, wherein the encoder layer is responsible for extracting the time sequence features of the expression feature matrix of the student and the eye concentration degree index, the decoder layer is responsible for fusing the time sequence features and outputting the attention state score of the student, the model core is a multi-head attention mechanism, the model adopts a residual connection structure to alleviate the gradient disappearance problem, and a layer normalization operation is added after each layer to improve the training stability, and finally, high-dimensional features are mapped to the student attention state score space through a full connection layer; The multi-dimensional features fused by the learning state evaluation network model include the spatial coordinate data of 68 facial key points and the trajectory index of eye movement; Wherein, the number of heads of the multi-head attention mechanism is determined by the expression change frequency, and the calculation formula is Wherein, F is the expression change frequency; the dimension d of each attention head is determined by the eye concentration threshold, and the calculation formula is Wherein is the eye concentration threshold; the depth L of the attention layer is calculated by the attention fluctuation rate R, and the calculation formula is ; The individual correlation model is a mapping function between the expression feature matrix and the attention concentration degree index for each student, which is trained by a deep learning algorithm, and is used to adjust the evaluation parameter weight according to the individual differences of the students to improve the accuracy of the attention state judgment; The specific expression is an expression change of the student in the classroom, which has a significant statistical difference from the personal expression baseline or the average expression state of the class, and is identified by the Mahalanobis distance or Z-score method, and reflects the irregular reaction of the student to the teaching content; The construction process of the attention fluctuation curve is as follows: firstly, the time variation characteristics of the attention concentration degree index of the student are analyzed by using a sliding time window technology, the statistical characteristics of the mean, standard deviation and change rate of the index in each time window are calculated to form an attention state description vector; meanwhile, the teaching content is divided into multiple content nodes with start and end time stamps by combining the teaching content progress information, and the time correlation of the attention concentration degree index and the content nodes is calculated by using the Pearson correlation coefficient; finally, the dynamic attention fluctuation curve is constructed based on the obtained attention state description vector and the content node time stamp.

2. The method of claim 1, wherein, The expression change frequency is the number of times that the expression similarity vector of the student changes significantly within a unit time, which is obtained by calculating the difference between adjacent time points, and reflects the activity degree of the emotional state change of the student.

3. The method of claim 2, wherein: The eye concentration degree index is a proportion of a time that a student's visual line stays in a specified area to a total observation time, is obtained by calculating a spatial aggregation degree and a time density of eye movement track points, and reflects a visual attention stability of the student.

4. The method of claim 3, wherein the method further comprises: The eye concentration degree threshold is a critical value for judging whether the student's visual line is in a concentrated state, is determined by analyzing historical data, and is used for converting continuous eye concentration degree indexes into discrete attention state judgments.

5. The method of claim 4, wherein, The eye concentration area is a spatial range where the student's visual line frequently stays, is obtained by analyzing visual line focus point coordinate sequences through a density clustering algorithm, and is represented as a hot spot area coordinate set and a weight distribution on a two-dimensional plane.

6. The method of claim 5, wherein, The attention concentration degree index is a cognitive input degree quantitative index of the student, which is calculated by comprehensively considering expression feature stability and the eye concentration degree index, is obtained by weightedly fusing multiple physiological behavior feature parameters, and reflects a content attention degree and a cognitive processing depth of the student in a classroom.

7. A computer readable storage medium characterized in that, The computer readable storage medium stores program instructions, and the program instructions are used for executing the interactive classroom student dynamic portrait generation method in any one of claims 1-6 when running in the computer.

Images