Emotion recognition and multi-dimensional teaching quality evaluation system and method based on artificial intelligence

Abstract

The invention discloses an emotion recognition and multi-dimensional teaching quality evaluation system and method based on artificial intelligence, and relates to the new technical field of artificial intelligence and teaching, and the system comprises an emotion recognition model which can efficiently extract and fuse emotion feature information in multi-dimensional peripheral physiological signals so as to realize accurate emotion classification. Based on the analysis results, the classroom teacher and student state analysis and teaching quality evaluation system constructs a classroom state analysis framework by quantifying student emotion, teacher emotion and classroom mutual dynamic characteristics, provides abundant and quantified classroom state data for educators, and generates multi-dimensional teaching quality evaluation indexes. According to the method, the teaching adjustment strategy can be dynamically generated and executed according to the evaluation result, and meanwhile, the structured teaching classroom state report is generated, so that the core problems of distraction of students, insufficient learning effect and the like in the current teaching scene can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and new teaching technologies. Specifically, it relates to an emotion recognition and multi-dimensional teaching quality evaluation system and method based on artificial intelligence. Background Art

[0002] With the development of educational informatization, improving the quality of classroom teaching in the information age has become a key goal in the field of education. However, most of the existing classroom evaluation methods rely on simple statistical analysis and fail to fully utilize the emotional state information of students and teachers for comprehensive evaluation. Traditional emotion classification methods usually adopt machine learning algorithms, but these methods often require manual feature design and rely on domain expert knowledge, having limitations in terms of scalability and adaptability. These methods rarely involve the recognition of emotion data through peripheral physiological signals, and these peripheral physiological signals have the advantages of being conveniently obtainable through wearable devices and objectively reflecting the emotional state of the human body.

[0003] In addition, the existing emotion classification models based on peripheral physiological signals are insufficient in extracting and fusing complex features between different physiological signals, resulting in the accuracy of emotion classification not meeting the requirements of classroom teacher-student state analysis and teaching quality evaluation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an emotion recognition and multi-dimensional teaching quality evaluation system and method based on artificial intelligence. The system includes an emotion recognition model that can efficiently extract and fuse emotion feature information in multi-dimensional peripheral physiological signals, thereby achieving accurate emotion classification. Based on these analysis results, the classroom teacher-student state analysis and teaching quality evaluation system constructs a classroom state analysis framework by quantifying student emotions, teacher emotions, and classroom interaction characteristics, provides rich and quantitative classroom state data for educators, and generates multi-dimensional teaching quality evaluation indicators. The present invention can dynamically generate and execute teaching adjustment strategies according to the evaluation results, and at the same time generate a structured teaching classroom state report, which can effectively solve core problems such as students' distracted attention and insufficient learning effectiveness in the current teaching scenario. It provides a scientific basis for teaching evaluation and at the same time enables teachers to optimize teaching strategies, having significant practical application value.

[0005] The present invention adopts the following technical solutions to achieve the invention purpose: An emotion recognition and multi-dimensional teaching quality evaluation system based on artificial intelligence, characterized by comprising: A physiological signal perception and fusion module, an emotion feature engineering and preprocessing module, an emotion classification module, an emotion-driven classroom state analysis module, a dynamic teaching intervention module, a teaching quality evaluation and optimization module, and an intelligent analysis report generation and feedback module connected in sequence; the physiological signal perception and fusion module includes a multi-modal physiological sensing integration unit and a peripheral physiological signal data fusion unit; the emotion feature engineering and preprocessing module includes a signal preprocessing unit, a data segmentation and annotation unit, and a label encoding and feature vectorization unit; the emotion classification module includes a multi-scale feature extraction layer, a spatial attention optimization layer, a channel global information interaction layer, a parameter-free feature weighting layer, and an adaptive pooling and classification layer; the emotion-driven classroom state analysis module includes a student emotion dimension analysis unit, a teacher emotion dimension analysis unit, and a classroom interaction dimension analysis unit; the dynamic teaching intervention module includes a real-time monitoring and warning unit, an intelligent strategy generation unit, a multi-modal intervention execution unit, and an intervention effect evaluation unit; the teaching quality evaluation and optimization module includes a data integration and analysis unit, a multi-dimensional evaluation generation unit, and an optimization suggestion generation unit; the intelligent analysis report generation and feedback module includes a data integration and visualization unit, an interactive analysis unit, and an intelligent distribution and feedback unit.

[0006] An evaluation method for an emotion recognition and multi-dimensional teaching quality evaluation system based on artificial intelligence, comprising the following steps: S01: Collect and fuse human peripheral physiological signal data in real time; S02: Perform data preprocessing on the peripheral physiological signals; The signal preprocessing unit preliminarily processes the peripheral physiological signals of teachers and students through a noise filter and a Min-Max normalization processor to improve the accuracy and efficiency of subsequent analysis. Among them, the noise filter uses a Butterworth low-pass filter to remove sensor jitter or electromagnetic interference, and a Butterworth high-pass filter to remove baseline drift interference. The transfer function formula of the Butterworth low-pass filter is shown in formula (1), and the transfer function formula of the Butterworth high-pass filter is shown in formula (2): (1); Where: N represents the filter order; ω c represents the cut-off frequency; s represents the complex frequency variable of the input signal; (2); Where: D0 represents the cut-off frequency; represents the distance from the point to the spectrum center in the image frequency domain; is the abscissa of the point in the frequency domain; is the ordinate of the point in the frequency domain; The Min-Max normalization processor uses the Min-Max method to unify all physiological signals into the interval [0, 1]; let the input sequence be , where n represents the length of the sequence, and the output sequence after Min-Max normalization is , as shown in formula (3): (3); where: represents the -th value in the sequence, and are the maximum and minimum values of all samples in the input sequence x, respectively; S03: Student and teacher emotion classification; S04: Student and teacher classroom state judgment; S05: Teacher-student classroom state adjustment; S06: Classroom teaching quality evaluation; S07: Output of classroom teaching state report.

[0007] As a further limitation of this technical solution, the specific steps of the S03 are as follows: S031: Use the multi-scale feature extraction layer to obtain the first emotion data information; S032: Use the channel global information interaction layer to obtain the second emotion data information; S033: Use the spatial attention optimization layer to obtain the third emotion data information; S034: Use the parameter-free feature weighting layer to obtain the fourth emotion data information; S035: Use the residual connection layer to obtain the fifth emotion data information; S036: Use the adaptive pooling and classification layer to obtain the emotion classification output.

[0008] As a further limitation of this technical solution, the specific steps of the S04 are as follows: S041: Analyze the learning emotions of students using the student emotion dimension analysis unit; The student emotion dimension analysis unit focuses on the emotional states of individual and group students, quantifies the attention, emotional stability, positive and negative learning emotions in the learning process through the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index and emotional fluctuation index, and assists in identifying potential problems in teaching content or environment. The learning focus weight is a parameter used to quantify external factors affecting students' concentration. The calculation formula of the positive learning emotion intensity index is shown in formula (4): (4); where: nIndicates the number of time segments into which the class time can be divided; Indicates the i th time segment, the confidence level of the k th type of positive learning emotion; Indicates the i th time segment, the confidence level of the k th type of negative learning emotion; Indicates the i th time segment, the confidence level of the k th type of neutral learning emotion; Indicates the weight of the k th type of positive learning emotion; Indicates the weight of the k th type of negative learning emotion, Indicates the weight of the k th type of neutral learning emotion; Indicates the external positive learning weight in the i th time segment, used to quantify the influence of external factors at the i th time point on the positive learning emotion intensity index, weighted and calculated by formula (5): (5); Where: Indicates the difficulty of the teaching content at the i th time point, Indicates the quality of the teaching method at the i-th time point, Indicates the adaptability of the classroom session at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The negative learning emotion intensity index is used to quantify the intensity of negative emotions of students in the classroom, helping teachers identify potential problems in teaching content or environment and adjust teaching strategies in a timely manner. The external negative learning weight is used to quantify the influence weight of the negative external environment of students on the classroom effect. The calculation formula of the negative learning emotion intensity index is shown in formula (6): (6); Where: Indicates the external negative learning weight in the i th time segment, used to quantify the influence of external factors at the i th time point on the negative learning emotion intensity index, weighted and calculated by formula (7): (7); Where: Indicates the iThe difficulty of the teaching content at a certain time point represents the i interference factors in the classroom environment at the th time point, i represents the negativity of the teacher's feedback at the , and are the weight coefficients of these three external influencing factors respectively; The learning motivation index is used to evaluate the learning enthusiasm of students driven by positive emotions, excluding the interference of neutral emotions, and quantifying the incentive effect of classroom activities on students. The learning motivation weight is a parameter used to quantify the external or internal factors that affect students' learning motivation. The formula for the learning motivation index is shown in Equation (8): (8); where: represents the learning motivation weight in the i th time segment, used to quantify the influence of external factors at the i th time point on the learning motivation index, and is calculated by weighted averaging using Equation (9): (9); where: represents the interest of students in the subject at the i th time point, represents the teacher's incentive measures at the i th time point, represents the family expectations and support at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The emotional fluctuation index is used to analyze the stability of students' emotional states, and judges the impact of classroom rhythm or external interference on students' emotions through the amplitude of emotional value fluctuations. For each time point, calculate the square of the difference between the emotional value and the average emotional value, multiply by the weight, sum up and take the mean and then take the square root to obtain the emotional fluctuation index. The formula for the emotional fluctuation index is shown in Equation (10): (10); where: represents the weight coefficient in the t th time segment; represents the emotional value of students at the t th time point; represents the total classroom time; represents the weighted average emotional value; calculated by Equation (11): (11); S042: Analyze the teaching emotions of teachers using the teacher emotion dimension analysis unit; The positive teaching emotion intensity index is used to quantify the positive emotion performance of teachers in the classroom, reflecting their teaching investment and infectivity. The external positive teaching weight is used to quantify the influence weight of teachers' positive emotions on the classroom atmosphere. The calculation formula of the positive teaching emotion intensity index is shown in Equation (12): (12); Where: represents the confidence of the i th type of positive teaching emotion in the k th time segment; represents the confidence of the i th type of negative teaching emotion in the k th time segment; represents the confidence of the i th type of neutral teaching emotion in the k th time segment; represents the weight of the k th type of positive teaching emotion; represents the weight of the k th type of negative teaching emotion; represents the weight of the k th type of neutral teaching emotion; represents the external positive teaching weight in the i th time segment, used to quantify the influence of external factors on the positive teaching emotion intensity index at the i th time point, weighted and calculated by formula (13): (13); Where: represents the vividness of the teaching language at the i th time point; represents the enthusiasm of classroom interaction at the i th time point; and are the weight coefficients of these two external influencing factors respectively; The negative teaching emotion intensity index is used to evaluate the negative emotions generated by teachers due to classroom challenges, assisting teachers in adjusting teaching strategies. The external negative teaching weight is used to quantify the influence weight of teachers' negative emotions on teaching effects. The calculation formula of the negative teaching emotion intensity index is shown in Equation (14): (14); Where: represents the external negative teaching weight in the i th time segment, used to quantify the influence of external factors on the negative teaching emotion intensity index at the iThe influence of external factors on the negative learning emotion intensity index at each time point is calculated by weighted calculation using formula (15): (15); Where: represents the situation of classroom progress pressure at the i th time point, represents the situation of insufficient preparation of teaching content at the i th time point, represents the situation of external interference at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; S043: Use the classroom interaction dimension analysis unit to analyze the classroom interaction situation; The teacher-student emotional synchronization index is used to analyze the matching degree of the teacher-student emotional state, reflect the consistency and interaction quality of emotional communication in the classroom. Multiply the teacher's emotional sequence by the synchronization weight and calculate the correlation with the student's emotional sequence to obtain the teacher-student emotional synchronization index. The calculation formula of the teacher-student emotional synchronization index is shown in formula (16): (16); Where: represents the emotional value of the teacher at the t th time point; represents the emotional value of the student at the t th time point; represents the weight at the t th time point, reflecting the importance of this moment; represents the weighted average of the teacher's emotional values, represents the weighted average of the student's emotional values; The interaction frequency index is used to count the density of teacher-student interactions per unit time and evaluate whether the classroom activity design effectively promotes student participation. The calculation formula of the interaction frequency index is shown in formula (17): (17); Where: represents the number of teacher-student interactions at the t th time point; The interaction quality index is based on the proportion of interactions with positive emotion markers and quantifies the positive effect of classroom interactions. The calculation formula of the interaction quality index is shown in formula (18): (18); Where: represents the number of teacher-student interactions based on positive emotions at the t th time point.

[0009] As a further limitation of this technical solution, the specific steps of S06 are as follows: The classroom participation index reflects the activity level and emotional synchronization of teacher-student interaction through the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, and measures the quality of classroom participation. The calculation formula of the classroom participation index is shown in Equation (19): (19); Where: , and are the weights of the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index respectively; The teaching efficiency score measures the achievement of teaching goals and the learning effect driven by emotions through the positive learning emotion intensity index, negative learning emotion intensity index, positive teaching emotion intensity index, and negative teaching emotion intensity index. The calculation formula of the teaching efficiency score is shown in Equation (20): (20); Where: and are the weights of the positive learning emotion intensity index, negative learning emotion intensity index, positive teaching emotion intensity index, and negative teaching emotion intensity index respectively; The teaching motivation index measures the driving effect of students' learning motivation and teachers' emotions in the classroom through the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index. The calculation formula of the teaching motivation index is shown in Equation (21): (21); Where: , and are the weights of the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index respectively; The interaction-driven emotional stability index evaluates the inhibitory effect of classroom interaction on the emotional fluctuation index by combining the interaction frequency index and the interaction quality index, and quantifies the ability of teacher-student interaction in teaching activities to stabilize the classroom emotional state. The calculation formula of the interaction-driven emotional stability index is shown in Equation (22): (22); Where: , and are the weights of the interaction frequency index, interaction quality index, and emotional fluctuation index respectively.

[0010] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The present invention designs a neural network model for emotion classification using a variety of peripheral physiological information to achieve efficient and accurate emotion classification, providing real-time and comprehensive teacher-student emotion data for classroom teaching. The present invention designs a classroom teacher-student state analysis and teaching quality evaluation system, which dynamically monitors and quantitatively evaluates the emotional states of students and teachers in the classroom environment and the classroom teaching and learning effects based on the emotional data identified from the neural network model. This evaluation system can provide real-time feedback, timely adjust the teaching strategies of teachers and the learning states of students, optimize the classroom atmosphere, and thus improve the teaching effect and the learning experience of students. At the same time, individual evaluation is carried out for each student, the classroom performance of each student is monitored, and a personalized learning optimization plan is provided, and a structured teaching classroom state report can be generated for the entire classroom. Through this innovative method, not only can the emotional interaction in the classroom be understood more comprehensively, but also the continuous improvement of teaching quality can be effectively promoted. The present invention deeply integrates the collection of peripheral physiological data of students and teachers, the classification and recognition of emotions, the classroom teacher-student state analysis and teaching quality evaluation system with the classroom teaching process, and proposes an emotion recognition and multi-dimensional teaching quality evaluation system and method based on artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a system framework diagram of the present invention.

[0012] Figure 2 It is an evaluation system diagram of the present invention.

[0013] Figure 3 It is a working flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The following will describe in detail a specific embodiment of the present invention in conjunction with the drawings, but it should be understood that the protection scope of the present invention is not limited by the specific embodiment.

[0015] The present invention includes a physiological signal perception and fusion module, an emotion feature engineering and preprocessing module, an emotion classification module, an emotion-driven classroom state analysis module, a dynamic teaching intervention module, a teaching quality evaluation and optimization module, and an intelligent analysis report generation and feedback module, which are connected in sequence.

[0016] The physiological signal perception and fusion module includes a multi-modal physiological sensing integration unit and a peripheral physiological signal data fusion unit; the physiological signal perception and fusion module collects the peripheral physiological signals of teachers and students in real time and performs data fusion; a customized peripheral physiological signal bracelet is used to collect the peripheral physiological signals of teachers and students in the classroom environment in real time and perform data fusion.

[0017] The multi-modal physiological sensing integration unit uses a customized peripheral physiological signal bracelet to collect the peripheral physiological signals of teachers and students in the classroom environment in real time. The customized peripheral physiological signal bracelet consists of a skin temperature sensor, a pulse wave sensor, a skin resistance sensor, and a charging electrode / data transmission port, where: the sampling frequency of the skin temperature sensor is 1 Hz, and the measurement range is 0 to 50 degrees Celsius; the sampling frequency of the pulse wave sensor is 100 Hz; the sampling frequency of the skin resistance sensor is 4 Hz, and the AC excitation source frequency is 24 Hz; the charging electrode / data transmission port is used for charging and data transmission. The peripheral physiological signal data includes one-axis data of skin resistance, one-axis data of pulse wave, and one-axis data of skin temperature. The one-axis data of heart rate can be calculated through the one-axis data of pulse wave. The peripheral physiological signal data fusion unit adjusts the data of the peripheral physiological signals of teachers and students through an interpolation algorithm, and converts the sensor data with different sampling frequencies into a unified time series format to ensure that the data of different sensors have consistent time resolution. The peripheral physiological signal data fusion unit splices and combines the peripheral physiological signals of teachers and students. The combined data is in the format of a two-dimensional array, with the pulse wave signal, skin resistance signal, heart rate signal, and skin temperature signal arranged horizontally in sequence, and arranged vertically in chronological order. The sampling frequency of the sensor data collection f can be set to an appropriate value to facilitate the subsequent data preprocessing process.

[0018] The emotion feature engineering and preprocessing module includes a signal preprocessing unit, a data segmentation and annotation unit, and a label encoding and feature vectorization unit; the emotion feature engineering and preprocessing module preprocesses the peripheral physiological signals, generates emotion data segments with labels, and converts them into a form convenient for the neural network model to extract features. The signal preprocessing unit is used to preliminarily process the peripheral physiological signals of teachers and students to improve the accuracy and efficiency of subsequent analysis. The signal preprocessing unit includes a noise filter and a minimum-maximum (Min-Max) normalization processor. The noise filter uses a low-pass filter to remove high-frequency noise and a high-pass filter to remove baseline drift and other low-frequency interferences. The Min-Max normalization processor unifies all physiological signals into the range of [0, 1] to ensure the numerical unity of different types of signals; the data segmentation and annotation unit is used to segment the continuous physiological signals into data segments of a fixed length to ensure that each data segment can reflect the complete emotion state within a period of time; the label encoding and feature vectorization unit annotates the segmented data segments containing the feature information of the peripheral physiological signals with corresponding emotion labels, and then converts each data segment into a two-dimensional vector form by directly expanding the dimension, which is convenient for the neural network model to extract features.

[0019] The emotion classification module includes a multi-scale feature extraction layer, a spatial attention optimization layer, a channel global information interaction layer, a parameter-free feature weighting layer, and an adaptive pooling and classification layer; the emotion classification module inputs the peripheral physiological signal feature information in batches into the emotion classification model, constructs the emotion classification model and determines the corresponding model parameters, and obtains the human emotion classification and recognition output after training; the emotion classification module extracts multi-scale spatio-temporal features from the peripheral physiological signal feature information segments through the multi-scale feature extraction layer using parallel convolutional kernels, enhances the diversity of feature expression, and obtains the first emotion data information; performs channel grouping and dynamic convolution processing on the first emotion data information through the channel global information interaction layer to enhance the global information interaction between channels and obtains the second emotion data information; performs reduction of parameter dependence and enhancement of feature selectivity on the second emotion data information through the spatial attention optimization layer using a lightweight channel attention mechanism to obtain the third emotion data information; generates an adaptive weight for the third emotion data information through the parameter-free feature weighting layer by dynamically calculating the spatial distribution characteristics of the features, enhances the discriminability of channel responses, and obtains the fourth emotion data information; uses a residual connection layer to retain the original feature information while fusing the peripheral physiological signal features with the fourth emotion data information through residual connection, reduces information loss and enhances the complementarity of multi-source features, and obtains the fifth emotion data information; compresses the feature dimension of the fifth emotion data information through adaptive pooling in the adaptive pooling and classification layer to achieve emotion classification output.

[0020] The multi-scale feature extraction layer includes a parallel convolutional kernel unit and an adaptive attention weighting unit. The two units use a selective kernel residual network to process the peripheral physiological signal feature information segments, and extract multi-scale spatio-temporal features through parallel convolutional kernels to enhance the diversity of feature expression and generate the first emotion data information. Among them, the parallel convolutional kernel unit uses four different sizes of convolutional kernels (3×1), (5×1), (7×1), and (9×1), and each convolutional kernel independently processes the input feature map to capture different scale features from local to global. Each convolutional kernel is followed by batch normalization to ensure the consistency and stability of each scale feature. The adaptive attention weighting unit performs global average pooling after summing the feature maps output by the parallel convolutional kernels, and then generates channel attention weights through a fully connected layer to obtain the first emotion data information.

[0021] The channel global information interaction layer includes a dual pooling feature extraction unit, a receptive field feature generation unit, and a receptive field feature fusion unit. The three units process the first emotional data information in real time through a dynamic multi-branch receptive field attention module, and enhance the global information interaction between channels through channel grouping and dynamic convolution to generate the second emotional data information. Among them, the dual pooling feature extraction unit extracts the average feature and the main feature of the first emotional data information through the average pooling path and the max pooling path respectively: after the average pooling path performs the average pooling operation on the first emotional data information, it expands the channel dimension through 1×1 grouped convolution to generate the global average feature; the max pooling path captures the local main feature through the max pooling operation, followed by 1×1 grouped convolution to expand the channel dimension to generate the local main feature. The weights generated by the two paths are then weighted and fused. The receptive field feature generation unit performs spatial expansion and non-linear activation on the input features of the weighted fusion through convolution operations. Specifically, after the input features are processed by convolution, they are normalized and non-linearly mapped through batch normalization and the Rectified Linear Unit (ReLU) activation function. The receptive field feature fusion unit first multiplies the feature data output by the receptive field feature generation unit and the dynamic weights generated by the dual pooling feature extraction unit element by element to generate weighted features. Subsequently, the receptive field feature fusion unit expands the spatial dimension of the weighted features through tensor reconstruction to restore the data format of the original input. This reconstruction operation rearranges the spatial dimension of the features to ensure that the cross-channel feature information can be interacted and fused in the data format of the original input for subsequent processing. Finally, the receptive field feature fusion unit compresses the channel dimension to the target output channel number through a convolutional layer, and is processed by batch normalization and the ReLU activation function to generate the second emotional data information.

[0022] The spatial attention optimization layer includes a global spatial information aggregation unit and an adaptive channel weighting unit. The two units process the global spatial information of the second sentiment data information in real time through lightweight parameter design, and generate channel attention weights through one-dimensional convolution to enhance the adaptive feature weighting between channels and generate the third sentiment data information. Among them, the global spatial information aggregation unit compresses the number of channels of the second sentiment data information through adaptive average pooling, eliminates the spatial dimension difference, and retains the global statistical information of the channel features. The adaptive channel weighting unit dynamically generates channel attention weights through one-dimensional convolution operation on the global descriptor, combines symmetric padding with the Sigmoid activation function, and its dimension is consistent with the channel dimension of the second sentiment data information. Finally, the weight is extended to the original spatial dimension through the broadcast mechanism and multiplied element by element with the input feature map to dynamically strengthen the key channel features while suppressing redundant information. The spatial attention optimization layer replaces the fully connected layer with parameter-efficient one-dimensional convolution to achieve adaptive attention regulation between channels and generate the third sentiment data information.

[0023] The parameter-free feature weighting layer includes a spatial distribution quantization unit and a statistics-driven weight generation unit. The two units process the spatial distribution characteristics of the third sentiment data information in real time through parameter-free design, generate adaptive weights by dynamically calculating the spatial dispersion of the features, and enhance the feature response in the spatially significant regions to generate the fourth sentiment data information. Among them, the spatial distribution quantization unit extracts the spatial distribution features of the features through global mean elimination and variance normalization. First, the spatial mean of each channel is calculated, and then the squared difference between the feature and the mean is calculated to quantify the spatial dispersion degree of the feature. The statistics-driven weight generation unit generates channel weights by normalizing the spatial statistical results. After normalizing the weights to the range of [0, 1] through the Sigmoid activation function, they are multiplied element by element with the feature map of the third sentiment data information to suppress the low-variance regions. The parameter-free feature weighting layer realizes spatial adaptive weighting through zero-parameter calculation and generates the fourth sentiment data information.

[0024] The residual connection layer includes a multi-scale feature processing unit and a dynamic dimension matching unit. While retaining the original feature information through the residual connection, the two units perform feature fusion on the peripheral physiological signal features and the fourth emotional data information, reducing information loss and enhancing the complementarity of multi-source features. Among them, the multi-scale feature processing unit uses a convolutional kernel with a variable receptive field to perform a non-linear transformation on the input feature map, and strengthens the non-linear expression ability of the features through batch normalization and activation functions. The dynamic dimension matching unit dynamically matches the feature dimensions according to the difference between the number of input channels and the number of output channels: if the input and output dimensions are inconsistent or the stride is not 1, a linear transformation is performed on the input features through convolutional kernels and batch normalization operations to align them with the output dimensions of the multi-scale feature processing unit; if the dimensions are consistent and the stride is 1, the original input features are directly passed. The residual connection layer adds the output of the multi-scale feature processing unit and the output of the dynamic dimension matching unit element by element, and processes it through the ReLU activation function to finally generate the fifth emotional data information.

[0025] The adaptive pooling and classification layer includes a dynamic spatial compression unit and a fully connected classification unit. The two units compress the feature dimensions of the fifth emotional data information through adaptive pooling and classification layer to achieve emotional classification output. Among them, the dynamic spatial compression unit dynamically adjusts the spatial dimensions of the input feature map through adaptive average pooling, compressing the height of the feature map to 1 and keeping the width as the original time steps of the input sequence. The fully connected classification unit flattens and maps the compressed features to a preset number of categories through a linear transformation layer to finally generate the emotional classification output.

[0026] The emotional classification module can classify and statistically identify three positive learning emotions, three negative learning emotions, and two neutral learning emotions of students recognized through the peripheral physiological signals of students; the emotional classification module can classify and statistically identify three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions of teachers recognized through the peripheral physiological signals of teachers. It should be noted that this setting is not fixed, and the number and specific dimensions of emotional classification can be flexibly adjusted according to the course type, subject characteristics, or teaching needs in actual applications.

[0027] The emotional driving classroom state analysis module includes a student emotional dimension analysis unit, a teacher emotional dimension analysis unit, and a classroom interaction dimension analysis unit. After receiving the emotional classification results of teachers and students from the emotional classification module, the emotional driving classroom state analysis module constructs a classroom state analysis framework by calculating and quantifying the characteristics of student emotions, teacher emotions, and classroom interaction dynamics, providing rich and quantitative classroom state data for educators. After receiving the emotional classification results of teachers and students from the emotional classification module, the three units analyze and obtain nine types of indicators for the student emotional dimension, teacher emotional dimension, and classroom interaction dimension in the classroom teaching process by combining some statistical information. The student emotional dimension analysis unit focuses on the emotional states of individual and group students, quantifies the attention, emotional stability, and positive and negative learning emotion impacts in the learning process through the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index, and emotional fluctuation index, and helps identify potential problems in teaching content or the environment. The teacher emotional dimension analysis unit is used to evaluate the emotional state and teaching adaptability of teachers in the classroom using the positive teaching emotion intensity index and negative teaching emotion intensity index, providing a reference basis for optimizing teachers' teaching behaviors and professional development. The classroom interaction dimension analysis unit is used to evaluate the quality of emotional interaction between teachers and students and the effectiveness of classroom activity design using the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, providing a basis for enhancing two-way communication in teaching.

[0028] The student emotional dimension analysis unit focuses on the emotional states of individual and group students, and quantifies the attention, emotional stability, positive and negative learning emotion impacts in the learning process through the Positive Emotion Learning Index (PELI), Negative Emotion Learning Index (NELI), Learning Motivation Index (LMI) and Emotion Fluctuation Index (EFI), to assist in identifying potential problems in teaching content or the environment. The Positive Emotion Learning Index measures the degree of attention of students driven by positive emotions in the classroom, reflecting the attractiveness of teaching content and the environment to students. For each time point, multiply the positive emotion value of the student by the corresponding external positive learning weight, sum them up, divide by the total sum of emotions at all time points, and finally multiply by 100% to obtain the percentage result. The external positive learning weight is a parameter used to quantify the external factors affecting students' concentration. The Negative Emotion Learning Index is used to quantify the intensity of negative emotions of students in the classroom, helping teachers identify potential problems in teaching content or the environment and adjust teaching strategies in a timely manner. For each time point, multiply the negative emotion value of the student by the external negative learning weight, sum them up, divide by the total sum of emotions at all time points, and finally multiply by 100% to obtain the percentage result. The external negative learning weight is used to quantify the impact weight of the negative external environment of students on the classroom effect. The Learning Motivation Index is used to evaluate the learning enthusiasm of students driven by positive emotions, excluding the interference of neutral emotions, and quantifying the incentive effect of classroom activities on students. For each time point, multiply the positive emotion value by the learning motivation weight, sum them up, divide by the total sum of emotions excluding neutral emotions at all time points, and finally multiply by 100% to obtain the percentage result. The learning motivation weight is a parameter used to quantify the external or internal factors affecting students' learning motivation. The Emotion Fluctuation Index is used to analyze the stability of students' emotional states, and judges the impact of classroom rhythm or external interference on students' emotions through the amplitude of emotional value fluctuations. For each time point, calculate the square of the difference between the emotional value and the average emotional value, multiply by the weight, sum them up, take the mean and take the square root to obtain the Emotion Fluctuation Index.

[0029] The teacher emotion dimension analysis unit uses the Positive Emotion Teaching Index (PETI) and the Negative Emotion Teaching Index (NETI) to evaluate the teacher's emotional state and teaching adaptability in the classroom. The Positive Emotion Teaching Index is used to quantify the teacher's positive emotional performance in the classroom, reflecting their teaching engagement and infectivity. For each time point, the teacher's positive emotion value is multiplied by the external positive teaching weight, the sum is divided by the total sum of the teacher's emotions at all time points, and finally multiplied by 100% to obtain the percentage result. The external positive teaching weight is used to quantify the influence weight of the teacher's positive emotion on the classroom atmosphere. The Negative Emotion Teaching Index is used to evaluate the negative emotions generated by the teacher due to classroom challenges, assisting the teacher in adjusting teaching strategies. For each time point, the teacher's negative emotion value is multiplied by the external negative teaching weight, the sum is divided by the total sum of the teacher's emotions at all time points, and finally multiplied by 100% to obtain the percentage result. The external negative teaching weight is used to quantify the influence weight of the teacher's negative emotion on the teaching effect.

[0030] The classroom interaction dimension analysis unit is used to evaluate the quality of teacher-student emotional interaction and the effectiveness of classroom activity design using the Teacher-Student Emotional Synchronization Index (TSESI), the Interaction Frequency Index (IFI), and the Interaction Quality Index (IQI), providing a basis for improving teaching quality. The Teacher-Student Emotional Synchronization Index is used to analyze the matching degree of the teacher-student emotional state, reflecting the consistency and interaction quality of emotional communication in the classroom. Multiply the teacher's emotion sequence by the synchronization weight, calculate the correlation with the student's emotion sequence, and obtain the Teacher-Student Emotional Synchronization Index. The Interaction Frequency Index is used to count the density of teacher-student interactions per unit time, evaluating whether the classroom activity design effectively promotes student participation. For each time point, the sum of the interaction times is divided by the total classroom duration to obtain the Interaction Frequency Index. The Interaction Quality Index is based on the proportion of interactions marked with positive emotions, quantifying the positive effect of classroom interactions. For each time point, the number of interactions based on positive emotions is multiplied by the interaction quality weight, the sum is divided by the total sum of the total number of interactions, and finally multiplied by 100% to obtain the percentage result.

[0031] The dynamic teaching intervention module includes a real-time monitoring and warning unit, an intelligent strategy generation unit, a multi-modal intervention execution unit, and an intervention effect evaluation unit; the dynamic teaching intervention module dynamically generates and executes teaching adjustment strategies by receiving nine types of classroom state analysis indicators output by the emotion-driven classroom state analysis module in real time and combining the real-time classification results of the emotion classification model; the four units dynamically generate and execute teaching adjustment strategies by receiving the output index results of the emotion-driven classroom state analysis module in real time and combining the real-time classification result data of the emotion classification model, transform the emotion analysis results into precise intervention actions, and directly optimize the classroom state. The real-time monitoring and warning unit continuously receives the output index results of the emotion-driven classroom state analysis module, combines the real-time classification result data of the emotion classification model, and uses preset threshold rules to identify key problems in the classroom in real time and immediately send intervention signals to the subsequent units. The intelligent strategy generation unit dynamically generates differentiated teaching adjustment strategies based on the triggered warning type and real-time data. The generation process of the adjustment strategy comprehensively considers multi-dimensional parameters to ensure that the strategy is accurately matched with the current classroom situation. The generation process of the strategy comprehensively considers multi-dimensional parameters to ensure that the strategy is accurately matched with the current classroom situation. The multi-modal intervention execution unit transforms the generated strategy into multi-channel execution actions. The teacher-side device receives intervention suggestions in real time and displays them sorted by urgency. The student-side receives personalized feedback or learning resource support through intelligent devices. In terms of environmental linkage, the system automatically controls classroom devices to achieve coordinated adjustment. The intervention effect evaluation unit verifies the effectiveness of the strategy by recalculating relevant indicators after the intervention and comparing them with the data before the intervention.

[0032] The teaching quality evaluation and optimization module includes a data integration and analysis unit, a multi-dimensional evaluation generation unit, and an optimization suggestion generation unit. The teaching quality evaluation and optimization module generates multi-dimensional teaching quality evaluation indicators by integrating the classroom state analysis indicators output by the emotion-driven classroom state analysis module, the intervention strategy execution records and effect data of the dynamic teaching intervention module, and combining the classification results of the emotion classification module, and proposes optimization suggestions based on the data-driven analysis results. The three units generate multi-dimensional teaching quality evaluation indicators by integrating the output results of the classroom state judgment module and the dynamic teaching intervention module, combining the classification results of the emotion classification module, and propose optimization suggestions based on the data-driven analysis results. The data integration and analysis unit receives and processes the results of nine types of indicators output by the classroom state judgment module, the intervention strategy execution records and effect data of the dynamic teaching intervention module, and the real-time classification results of the emotion classification module. Through data cleaning, feature fusion, and multi-source information alignment, the data integration and analysis unit transforms the scattered original data into a structured evaluation data set, providing a unified data basis for subsequent analysis. The multi-dimensional evaluation generation unit quantifies and generates multi-dimensional teaching quality indicators based on the integrated data, and quantifies the classroom teaching effect through the classroom participation index, teaching efficiency score, teaching motivation index, and interaction-driven emotional stability index.

[0033] The data integration and analysis unit receives and processes the nine types of index results of the judgment of the classroom states of students and teachers output by the classroom state judgment module, the execution records and effect data of the intervention strategies of the dynamic teaching intervention module, and the real-time classification results of the emotion classification module. Through data cleaning, feature fusion, and multi-source information alignment, the data integration and analysis unit transforms the scattered original data into a structured evaluation data set, providing a unified data basis for subsequent analysis. The multi-dimensional evaluation generation unit quantifies and generates multi-dimensional teaching quality indicators based on the integrated data. Among them, the Class Participation Index (CPI) reflects the activity level and emotional synchronization of teacher-student interaction through the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, and measures the quality of classroom participation. The Teaching Efficiency Score (TES) measures the achievement of teaching objectives and the learning effect driven by emotions through the positive learning emotion intensity index, negative learning emotion intensity index, positive teaching emotion intensity index, and negative teaching emotion intensity index. The Teaching Momentum Index (TMI) measures the driving effect of students' learning motivation and teachers' emotions in the classroom through the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index. The Interaction-Driven Stability Index (IDSI) evaluates the inhibitory effect of classroom interaction on emotional fluctuations by combining the interaction frequency index and the interaction quality index, and quantifies the ability of teacher-student interaction in teaching activities to stabilize the classroom emotional state.

[0034] The optimization suggestion generation unit further analyzes the abnormal or inefficient links of the evaluation indicators, and generates targeted optimization suggestions by combining educational theories and historical teaching data.

[0035] The intelligent analysis report generation and feedback module includes a data integration and visualization unit, an interactive analysis unit, and an intelligent distribution and feedback unit; the intelligent analysis report generation and feedback module generates a structured visualization report by integrating the output results of the emotion-driven classroom state analysis module, the dynamic teaching intervention module, and the teaching quality evaluation and optimization module, and combining the classification results of the emotion classification module. The three units generate a structured visualization report by integrating the output results of the classroom state judgment module, the dynamic teaching intervention module, and the teaching quality evaluation and optimization module, and combining the classification results of the emotion classification module. The data integration and visualization unit receives and processes multi-source data from the classroom state judgment module, the dynamic teaching intervention module, and the teaching quality evaluation and optimization module, including the results of nine types of indicators for judging the classroom states of students and teachers, the execution records and effect data of intervention strategies, multi-dimensional teaching quality indicators, and optimization suggestions, and generates a structured visualization report by combining the classification results of the emotion classification module. This unit converts the scattered original data into structured visualization elements, generates core report components, presents the spatio-temporal distribution of emotional states, shows the dynamic change trends of key indicators through a timeline, and presents the optimization suggestions in a list form. The interactive analysis unit supports users to deeply explore data correlations through interactive operations such as filtering and comparison. This interactive design transforms the data presentation from a static report to a dynamic analysis tool, helping educators quickly locate the root causes of problems and verify the intervention effects. The intelligent distribution and feedback unit accurately pushes the report to the educator, administrator, or parent side through multiple channels. At the same time, this unit collects the feedback from users on the report content, integrates it into the system, and uses it to optimize the subsequent data processing logic and report generation strategy.

[0036] An evaluation method for an emotion recognition and multi-dimensional teaching quality evaluation system based on artificial intelligence includes the following steps: S01: Real-time collect and fuse human peripheral physiological signal data; The real-time collection and fusion of the peripheral physiological signals are carried out in the context of a classroom teaching system. Through the physiological signal perception and fusion module, a customized peripheral physiological signal bracelet is used to real-time collect the peripheral physiological signals of teachers and students in the classroom environment and perform data fusion.

[0037] The human peripheral physiological signal data includes one-axis data of skin resistance, one-axis data of pulse wave, and one-axis data of skin temperature. The one-axis data of heart rate can be calculated from the one-axis data of pulse wave. The peripheral physiological signal data fusion unit adjusts the teacher-student's peripheral physiological signals through an interpolation algorithm, converting sensor data with different sampling frequencies into a unified time series format to ensure that data from different sensors has consistent time resolution. The peripheral physiological signal data fusion unit splices and combines the teacher-student's peripheral physiological signals. The combined data is in the format of a two-dimensional array, with pulse wave signals, skin resistance signals, heart rate signals, and skin temperature signals arranged horizontally in sequence, and sensor data acquisition frequencies arranged vertically in chronological order. f It can be set to an appropriate frequency, such as 100Hz, that is, the size of the data generated per second is 100 rows and 4 columns, facilitating subsequent data preprocessing.

[0038] S02: Perform data preprocessing on peripheral physiological signals; The data preprocessing of the peripheral physiological signals is carried out through an emotion feature engineering and preprocessing module, which preprocesses the peripheral physiological signals to generate emotion data segments with labels and converts them into a form convenient for the neural network model to extract features. The signal preprocessing unit initially processes the teacher-student's peripheral physiological signals through a noise filter and a Min-Max normalization processor to improve the accuracy and efficiency of subsequent analysis. Among them, the noise filter uses a Butterworth low-pass filter to remove high-frequency noises such as sensor jitter or electromagnetic interference, and a Butterworth high-pass filter to remove baseline drift and other low-frequency interferences. The transfer function formula of the Butterworth low-pass filter is shown in formula (1), and the transfer function formula of the Butterworth high-pass filter is shown in formula (2): (1); Where: N represents the filter order, which determines the steepness of the transition band and the filtering characteristics. The higher the order, the faster the attenuation rate of the stopband, but the width of the transition band increases and the phase delay also increases; ω c represents the cut-off frequency; s represents the complex frequency variable of the input signal; (2); Where: D0 represents the cut-off frequency; represents the distance from the point to the spectrum center in the image frequency domain; is the abscissa of the point in the frequency domain; is the ordinate of the point in the frequency domain; The Min - Max normalization processor uses the Min - Max method to unify all physiological signals into the interval [0, 1], ensuring the numerical unity of different types of signals for subsequent fusion and analysis. Let the input sequence be , where n represents the length of the sequence. The output sequence after Min - Max normalization is , as shown in formula (3): (3); where: represents the th value in the sequence, and are the maximum and minimum values of all samples in the input sequence x respectively; The data segmentation and annotation unit uses a window of fixed length to segment continuous sensor data into data segments of fixed length for input into the recognition model. Set the sensor acquisition frequency to f = 100Hz. According to the characteristics of human peripheral physiological signals, a window of length W = 100 is selected, that is, each window contains W / f = 1 second of sensor data. The sliding step S can be set to 50, that is, the data coverage rate is S / W = 50%, thus segmenting the data into a size of W rows M columns, M is the number of columns after merging all sensor data. When using pulse wave signals, skin resistance signals, heart rate signals, and skin temperature signals, the size of one data segment is 100 rows and 4 columns.

[0039] The label encoding and feature vectorization unit then annotates the segmented data segments containing peripheral physiological signal feature information with corresponding emotion labels. The labels include three positive learning emotions, three negative learning emotions, and two neutral learning emotions of students, as well as three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions for identifying teachers. Finally, each data segment is transformed into a two - dimensional vector form by directly expanding the dimension, facilitating the neural network model to extract features.

[0040] S03: Student and teacher emotion classification; The emotional classification of students and teachers is to input emotional data segments into the emotional classification model in batches through the emotional classification module, construct the emotional classification model and determine the corresponding model parameters, and obtain the human emotional classification and recognition output after training. The emotional classification model extracts multi-scale spatio-temporal features from the peripheral physiological signal feature information segments through a multi-scale feature extraction layer using parallel convolutional kernels to enhance the diversity of feature expressions and obtain the first emotional data information. The first emotional data information is subjected to channel grouping and dynamic convolution processing through a channel global information interaction layer to enhance the global information interaction between channels and obtain the second emotional data information. The second emotional data information is processed through a spatial attention optimization layer using a lightweight channel attention mechanism to reduce parameter dependence and improve feature selectivity, and obtain the third emotional data information. The third emotional data information is weighted by a parameter-free feature weighting layer and an adaptive weight is generated by dynamically calculating the spatial distribution characteristics of the features to enhance the discriminability of channel responses and obtain the fourth emotional data information. Through the residual connection layer, while retaining the original feature information using the residual connection, the peripheral physiological signal features are fused with the fourth emotional data information to reduce information loss and enhance the complementarity of multi-source features, and obtain the fifth emotional data information. The fifth emotional data information is compressed in feature dimension through adaptive pooling by the adaptive pooling and classification layer to achieve emotional classification output.

[0041] Among them, the emotional classification module can identify and statistically classify three positive learning emotions, three negative learning emotions, and two neutral learning emotions of students through the peripheral physiological signals of students; the emotional classification module can identify and statistically classify three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions of teachers through the peripheral physiological signals of teachers. It should be noted that for the convenience of elaboration, the number of emotions set in this system here is three positive learning emotions, three negative learning emotions, two neutral learning emotions, three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions, but this setting is not fixed, and the number and specific dimensions of emotional classification can be flexibly adjusted according to the course type, subject characteristics, or teaching needs in actual applications.

[0042] S04: Judgment of the classroom states of students and teachers; The judgment of the classroom states of students and teachers is carried out through the emotion-driven classroom state analysis module. Based on the emotion classification results of teachers and students by the emotion classification module in S03, combined with some statistical information, such as classroom interaction situation, homework completion degree, and course difficulty, the student emotion dimension analysis unit, teacher emotion dimension analysis unit, and classroom interaction dimension analysis unit are used to analyze the classification results for various emotion-based indicators. Through the analysis, nine types of indicators in the student emotion dimension, teacher emotion dimension, and classroom interaction dimension in the classroom teaching process are obtained. The nine types of indicators jointly construct a classroom emotion analysis framework by quantifying the dynamic characteristics of student emotion, teacher emotion, and classroom interaction, providing accurate teaching optimization suggestions based on data for educators.

[0043] S05: Adjustment of the classroom states of teachers and students; The adjustment of the classroom states of teachers and students is carried out through the dynamic teaching intervention module, which conducts dynamic teaching intervention based on the output results in S03 and S04 to optimize the classroom teaching quality.

[0044] The real-time monitoring and early warning unit continuously receives the nine types of indicators output by S04, and combines the real-time classification result data of S03. Through the preset threshold rules, it real-time identifies key problems in the classroom, such as students' distracted attention, cumulative negative emotions, or the failure of teacher-student emotion synchronization, and immediately sends an intervention signal to the subsequent unit. The intelligent strategy generation unit dynamically generates differentiated teaching adjustment strategies based on the triggered early warning type and real-time data. The strategy generation process comprehensively considers multi-dimensional parameters to ensure the accurate matching of the strategy with the current classroom situation. The multi-modal intervention execution unit converts the generated strategy into multi-channel execution actions. The teacher-side device receives the intervention suggestions in real-time and displays them sorted by urgency. The student-side receives personalized feedback or learning resource support through intelligent devices. In terms of environment linkage, the system automatically controls classroom devices to achieve the coordinated adjustment of the teaching environment and teaching behavior. The intervention effect evaluation unit verifies the effectiveness of the strategy by recalculating the relevant indicators after the intervention and comparing them with the data before the intervention, forming a continuous improvement cycle of "monitoring - intervention - optimization".

[0045] S06: Evaluation of classroom teaching quality; The evaluation of classroom teaching quality is carried out through the teaching quality evaluation and optimization module. Based on the output results of S03, S04, and S05, it generates multi-dimensional teaching quality evaluation indicators, and comprehensively evaluates the classroom teaching quality through the classroom participation index, teaching efficiency score, teaching motivation index, and teaching motivation index in the classroom teaching quality evaluation, and puts forward optimization suggestions based on the data-driven analysis results.

[0046] S07: Output of the classroom state report of teaching.

[0047] The output of the teaching classroom status report is generated by the intelligent analysis report generation and feedback module, which generates a structured visual report based on the results of S03, S04, S05, and S06. The intelligent analysis report generation and feedback module includes a data integration and visualization unit, an interactive analysis unit, and an intelligent distribution and feedback unit.

[0048] The data integration and visualization unit receives and processes the results of the emotion classification module for teachers and students in S03, the nine types of index results of the classroom status judgment of students and teachers in S04, the intervention strategy execution records and effect data of the classroom status adjustment of teachers and students in S05, and the multi-dimensional teaching quality indicators generated by quantification in S06. This unit transforms the scattered original data into structured visual elements, generates core report components, presents the spatio-temporal distribution of the emotion state in a heat map, shows the dynamic change trend of key indicators through a time axis, and presents optimization suggestions in a list form. The interactive analysis unit supports users to deeply explore the data correlation through interactive operations such as filtering and comparison. For example, a teacher can click on a specific time period in the emotion heat map to view the corresponding emotion classification distribution and intervention strategy execution records; or locate key events through the time axis slider and compare the index changes before and after the intervention. The teaching classroom status report will display the relevant data of all students at the same time, and take the average after summing up all the data, which can accurately reflect the comprehensive level of this classroom. This interactive design transforms the data presentation from a static report to a dynamic analysis tool, helping educators quickly locate the root cause of problems and verify the intervention effect. The intelligent distribution and feedback unit then accurately pushes the report to educators, administrators, or the parent side through multiple channels. At the same time, this unit collects the feedback from users on the report content, integrates it into the system, and uses it to optimize the subsequent data processing logic and report generation strategy.

[0049] The specific steps of S03 are as follows: S031: Obtain the first emotion data information using a multi-scale feature extraction layer; The multi-scale feature extraction layer includes a parallel convolution kernel unit and an adaptive attention weighting unit. The two units use a selective kernel residual network to process peripheral physiological signal feature information segments in real time, and extract multi-scale spatio-temporal features through parallel convolution kernels to enhance the diversity of feature expression and generate first emotional data information. Among them, the parallel convolution kernel unit adopts four different sizes of convolution kernels: (3×1), (5×1), (7×1) and (9×1). The padding parameter of all convolution kernels is set to (kernel size – 1) / 2, where kernel size is the convolution kernel size. Each convolution kernel independently processes the input feature map to capture different scale features from local to global. After each convolution kernel, batch normalization is performed to ensure the consistency and stability of each scale feature. The adaptive attention weighting unit splices the feature maps output by the parallel convolution kernels along the channel dimension. First, it sums the features of each scale along the channel dimension to fuse the multi-scale information into a single feature map, and then compresses it into a feature vector with a spatial dimension of 1 through global average pooling. Subsequently, a weight vector of [b, out_channel×4] is generated through a fully connected layer and a Sigmoid activation function, and it is reshaped into an attention weight tensor of [b, 4, out_channel, 1, 1], where b is the batch size and out_channel is the number of output channels. Finally, the multi-scale features are dynamically fused by multiplying the weight tensor and the original feature tensor element by element and summing along the convolution kernel dimension, thereby generating first emotional data information.

[0050] S032: Obtain second emotional data information using a channel global information interaction layer; The channel global information interaction layer includes a dual pooling feature extraction unit, a receptive field feature generation unit, and a receptive field feature fusion unit. The three units process the first emotional data information in real time through a dynamic multi-branch receptive field attention module, and enhance the global information interaction between channels through channel grouping and dynamic convolution to generate the second emotional data information. Among them, the dual pooling feature extraction unit extracts the average feature and the main feature of the first emotional data information through the average pooling path and the max pooling path respectively: after the average pooling path performs spatial average pooling on the input feature, it expands the channel dimension to in_channel × kernel_size² through a 1×1 grouped convolution (groups = in_channel), where groups is the number of groups, in_channel is the number of input channels, to generate the global average feature weight; the max pooling path captures the local main feature through spatial max pooling, and then follows the grouped convolution with the same structure to generate the local main feature weight. The weights of the two paths are fused by weighted fusion (the weight coefficients of the average pooling path and the max pooling path are 0.925 and 0.075 respectively) to achieve the dynamic balance of global and local features. The receptive field feature generation unit expands the input feature spatially through depthwise separable convolution (groups = in_channel), and then processes it through batch normalization and the ReLU activation function to generate an intermediate feature tensor containing multi-scale spatial correlations. Subsequently, the receptive field feature fusion unit performs feature weighting on the weighted feature weight and the generated intermediate feature through element-wise multiplication to obtain a weighted feature tensor of [b, c, kernel_size², h, w], where c is the number of channels, h is the height of the feature map, and w is the width of the feature map. The receptive field feature fusion unit expands the spatial dimension from (h, w) to (h × kernel_size, w × kernel_size) through a tensor reconstruction operation to restore the spatial resolution of the original input, ensuring the continuity of cross-channel features in the spatial domain. Finally, the reconstructed feature compresses the channel dimension to the target output channel number through a 1×1 convolution, completing the complete feature processing process from channel interaction to spatial reconstruction, and outputting the second emotional data information.

[0051] S033: Obtain the third emotional data information using the spatial attention optimization layer; The spatial attention optimization layer includes a global spatial information aggregation unit and an adaptive channel weighting unit. The two units process the global spatial information of the second sentiment data information in real time through lightweight parameter design, generate channel attention weights through one-dimensional convolution, enhance the adaptive feature weighting between channels, and generate the third sentiment data information. Among them, the global spatial information aggregation unit uses adaptive average pooling to compress the spatial dimension of the input feature map to 1 × 1, forming a channel-level global statistical descriptor [b, c, 1, 1]. This process preserves the channel distribution information of the original features by eliminating spatial differences, while significantly reducing the computational amount. The adaptive channel weighting unit performs cross-channel interaction modeling on the global descriptor through one-dimensional convolution: first, the pooled feature [b, c, 1, 1] is converted into a channel dimension vector of [b, 1, c], and then a convolutional kernel with a kernel size is used, combined with symmetric padding padding=(kernel size – 1) / 2 for local feature interaction to generate candidate channel weights; after transposition and compression to the [0, 1] interval through the Sigmoid activation function, the weight tensor is restored to the original channel dimension of [b, c, 1, 1]. Finally, the channel weights are extended to the original spatial dimension [b, c, h, w] through the broadcast mechanism and multiplied element-wise with the input feature map to achieve dynamic enhancement of key channel features and suppression of redundant information. The spatial attention optimization layer replaces the fully connected layer with parameter-efficient one-dimensional convolution to achieve adaptive attention regulation between channels and generate the third sentiment data information.

[0052] S034: Obtain the fourth sentiment data information using a parameter-free feature weighting layer; The parameter-free feature weighting layer includes a spatial distribution quantization unit and a statistics-driven weight generation unit. The two units process the spatial distribution characteristics of the third sentiment data information in real time through parameter-free design, generate adaptive weights by dynamically calculating the spatial dispersion of features, enhance the feature response in spatially significant regions, and generate the fourth sentiment data information. Among them, the spatial distribution quantization unit first calculates the spatial mean of each channel through the global mean, and then quantifies the dispersion degree between the feature and the mean through the squared difference operation to generate an intermediate feature tensor representing the spatial distribution dispersion. Specifically, the squared difference between the input feature x and the channel global mean directly reflects the deviation degree between the local region and the global distribution, and the larger this value is, the higher the spatial dispersion. The statistics-driven weight generation unit generates channel weights by normalizing the spatial statistical results. After normalizing the weights to the [0, 1] range through the Sigmoid activation function, the weights are extended to the original feature dimension through broadcast and multiplied element-wise with the third sentiment data information feature map to suppress low-variance regions and enhance features with significant spatial dispersion. The parameter-free feature weighting layer achieves spatial adaptive weighting through zero-parameter statistical calculations and generates the fourth sentiment data information.

[0053] S035: Obtain the fifth emotional data information using a residual connection layer; The residual connection layer includes a multi-scale feature processing unit and a dynamic dimension matching unit. While retaining the original feature information through the residual connection, these two units perform feature fusion on the peripheral physiological signal features and the fourth emotional data information, reducing information loss and enhancing the complementarity of multi-source features. Among them, the multi-scale feature processing unit performs a non-linear transformation on the input feature map using a convolutional kernel with a variable receptive field, and strengthens the non-linear expression ability of the features through batch normalization and activation functions. The dynamic dimension matching unit dynamically matches the feature dimensions according to the difference between the number of input channels and the number of output channels: if the input and output dimensions are inconsistent or the stride is not 1, a linear transformation is performed on the input features through convolutional kernels and batch normalization operations to align them with the output dimensions of the multi-scale feature processing unit; if the dimensions are consistent and the stride is 1, the original input features are directly passed through. The residual connection layer adds the output of the multi-scale feature processing unit and the output of the dynamic dimension matching unit element-wise, and processes it through a ReLU activation function to finally generate the fifth emotional data information.

[0054] S036: Obtain the emotional classification output using an adaptive pooling and classification layer.

[0055] The adaptive pooling and classification layer includes a dynamic spatial compression unit and a fully connected classification unit. These two units compress the feature dimensions of the fifth emotional data information through the adaptive pooling and classification layer to achieve the emotional classification output. Among them, the dynamic spatial compression unit dynamically adjusts the spatial dimensions of the input feature map through adaptive average pooling, compressing the height of the feature map to 1 and keeping the width as the original time steps of the input sequence. The fully connected classification unit flattens and maps the compressed features to a preset number of categories through a linear transformation layer to finally generate the emotional classification output.

[0056] Among them, the emotional classification module can perform classification recognition and statistics on three positive learning emotions, three negative learning emotions, and two neutral learning emotions of students identified through the peripheral physiological signals of students; the emotional classification module can perform classification recognition and statistics on three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions of teachers identified through the peripheral physiological signals of teachers. It should be noted that for the convenience of elaboration, the number of emotions set in this system here is three positive learning emotions, three negative learning emotions, two neutral learning emotions, three positive teaching emotions, three negative teaching emotions, and two neutral teaching emotions, but this setting is not fixed, and the number and specific dimensions of emotional classification can be flexibly adjusted according to course types, subject characteristics, or teaching requirements in actual applications.

[0057] The specific steps of S04 are as follows: S041: Analyze students' learning emotions using the student emotion dimension analysis unit; The student emotion dimension analysis unit focuses on the emotional states of individual and group students, quantifies the attention, emotional stability, positive and negative learning emotion impacts in the learning process through the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index, and emotional fluctuation index, and helps identify potential problems in teaching content or environment. The positive learning emotion intensity index measures the degree of attention concentration driven by positive emotions of students in the classroom and reflects the attractiveness of teaching content and environment to students. For each time point, multiply the positive emotion value of the student by the corresponding external positive learning weight, sum them up, and then divide by the total sum of emotions at all time points. The learning focus weight is a parameter used to quantify external factors affecting students' concentration. The calculation formula for the positive learning emotion intensity index is shown in Equation (4): (4); Where: n represents the number of time segments that the classroom time can be divided into; represents the i th time segment, the confidence level of the k th type of positive learning emotion; represents the i th time segment, the confidence level of the k th type of negative learning emotion; represents the i th time segment, the confidence level of the k th type of neutral learning emotion; represents the weight of the k th type of positive learning emotion; represents the weight of the k th type of negative learning emotion, represents the weight of the k th type of neutral learning emotion; represents the external positive learning weight in the i th time segment, used to quantify the impact of external factors on the positive learning emotion intensity index at the i th time point, weighted and calculated by formula (5): (5); Where: represents the difficulty of teaching content at the i th time point, represents the quality of teaching method at the i-th time point, represents the adaptability of classroom links at the i th time point, , and The weight coefficients for these three external influencing factors respectively; The negative learning emotion intensity index is used to quantify the intensity of negative emotions of students in the classroom, helping teachers identify potential problems in teaching content or the environment and adjust teaching strategies in a timely manner. The external negative learning weight is used to quantify the influence weight of the negative external environment of students on the classroom effect. The difficulty of teaching content exceeds the ability of students; interference factors in the classroom environment; the negativity of teacher feedback. The calculation formula of the negative learning emotion intensity index is shown in Equation (6): (6); Where: represents the external negative learning weight in the i th time segment, used to quantify the influence of external factors on the negative learning emotion intensity index at the i th time point, weighted and calculated by formula (7): (7); Where: represents the difficulty of teaching content at the i th time point, represents the interference factors in the classroom environment at the i th time point, represents the negativity of teacher feedback at the i th time point, , and are the weight coefficients for these three external influencing factors respectively; The learning motivation index is used to evaluate the learning enthusiasm of students driven by positive emotions, excluding the interference of neutral emotions, and quantifying the incentive effect of classroom activities on students. For each time point, multiply the positive emotion value by the learning motivation weight, sum and divide by the sum of the total emotion values after excluding neutral emotions at all time points, and finally multiply by 100% to get the percentage result. The learning motivation weight is used to quantify the parameters of external or internal factors that affect students' learning motivation. The formula of the learning motivation index is shown in Equation (8): (8); Where: represents the learning motivation weight in the i th time segment, used to quantify the influence of external factors on the learning motivation index at the i th time point, weighted and calculated by formula (9): (9); Where: represents the interest of students in the subject at the i th time point, represents the teacher incentive measures at the i th time point, Represents the family expectations and support at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The emotional fluctuation index is used to analyze the stability of students' emotional states. By judging the amplitude of emotional value fluctuations, it analyzes the impact of classroom rhythm or external disturbances on students' emotions. For each time point, calculate the square of the difference between the emotional value and the average emotional value, multiply by the weight, sum them up, take the mean and then take the square root to obtain the emotional fluctuation index. The formula for the emotional fluctuation index is shown in Equation (10): (10); Where: Represents the weight coefficient in the t th time segment, reflecting the importance of this time point, and the value range is [0, 1]; Represents the t th student emotional value at the time point; Represents the total classroom time; Represents the weighted average emotional value, which is used as a reference value to calculate the emotional fluctuation amplitude at each time point; it is calculated by Equation (11): (11); S042: Use the teacher emotional dimension analysis unit to analyze the teacher's teaching emotions; The teacher emotional dimension analysis unit is used to evaluate the teacher's emotional state and teaching adaptability in the classroom using the positive teaching emotion intensity index and the negative teaching emotion intensity index, providing a reference basis for optimizing the teacher's teaching behavior and career development. The positive teaching emotion intensity index is used to quantify the teacher's positive emotion performance in the classroom, reflecting their teaching investment degree and infectivity. For each time point, multiply the teacher's positive emotion value by the external positive teaching weight, sum them up, divide by the total sum of the teacher's emotions at all time points, and finally multiply by 100% to obtain the percentage result. The external teaching positive weight is used to quantify the influence weight of the teacher's positive emotions on the classroom atmosphere. The calculation formula for the positive teaching emotion intensity index is shown in Equation (12): (12); Where: Represents the confidence level of the i th type of positive teaching emotion in the k th time segment; Represents the confidence level of the i th type of negative teaching emotion in the k th time segment; Represents the i th time segment, the kConfidence of neutral teaching emotion; Indicates the k weight of the positive teaching emotion of the k category; Indicates the k weight of the negative teaching emotion of the i category; the weights of the three teaching emotions are used to comprehensively evaluate the intensity of teachers' classroom teaching emotions by quantifying the impacts of different positive, negative, and neutral emotions on teaching behaviors; i Indicates the external positive teaching weight in the th time segment, which is used to quantify the impact of external factors on the positive teaching emotion intensity index at the th time point, and is calculated by weighted formula (13): where: i Indicates the vividness of teaching language at the th time point; i Indicates the enthusiasm of classroom interaction at the th time point; and are the weight coefficients of these two external influencing factors respectively; The negative teaching emotion intensity index is used to evaluate the negative emotions of teachers caused by classroom challenges and assist teachers in adjusting teaching strategies. For each time point, multiply the negative emotion value of the teacher by the teaching negative weight, sum them up, divide by the sum of all teachers' emotions at all time points, and finally multiply by 100% to get the percentage result. The external negative teaching weight is used to quantify the impact weight of teachers' negative emotions on teaching effects. The calculation formula of the negative teaching emotion intensity index is shown in formula (14): where: Indicates the external negative teaching weight in the i th time segment, which is used to quantify the impact of external factors on the negative learning emotion intensity index at the i th time point, and is calculated by weighted formula (15): (15); where: Indicates the situation of classroom progress pressure at the i th time point; Indicates the situation of insufficient preparation of teaching content at the i th time point; Indicates the situation of external interference at the i th time point; , and The weight coefficients for these three external influencing factors respectively; S043: Analyze the classroom interaction situation using the classroom interaction dimension analysis unit; The classroom interaction dimension analysis unit is used to evaluate the quality of the emotional interaction between teachers and students and the effectiveness of classroom activity design using the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, providing a basis for improving two-way teaching communication. The teacher-student emotional synchronization index is used to analyze the matching degree of the emotional states of teachers and students, reflecting the consistency and interaction quality of emotional communication in the classroom. Multiply the teacher's emotional sequence (positive, neutral, negative time-series data) by the synchronization weight and calculate the correlation with the student's emotional sequence to obtain the teacher-student emotional synchronization index. The calculation formula for the teacher-student emotional synchronization index is shown in Equation (16): (16); Where: represents the emotional value of the teacher at the t th time point; represents the emotional value of the student at the t th time point; represents the weight at the t th time point, reflecting the importance of this moment; represents the weighted average of the teacher's emotional values, represents the weighted average of the student's emotional values; The interaction frequency index is used to count the density of teacher-student interactions per unit time and evaluate whether the classroom activity design effectively promotes student participation. For each time point, sum the number of interactions and divide by the total classroom duration to obtain the interaction frequency index. The calculation formula for the interaction frequency index is shown in Equation (17): (17); Where: represents the number of teacher-student interactions at the t th time point; The interaction quality index is based on the proportion of interactions with positive emotional markers, quantifying the positive effects of classroom interactions (such as encouragement, effective communication). For each time point, multiply the number of interactions with positive emotions by the interaction quality weight, sum them up, divide by the total sum of the number of interactions, and finally multiply by 100% to obtain the percentage result. The calculation formula for the interaction quality index is shown in Equation (18): (18); Where: represents the number of teacher-student interactions with positive emotions at the t th time point.

[0058] The specific steps of the said S06 are as follows: The classroom teaching quality assessment generates multi-dimensional teaching quality assessment indicators through the teaching quality assessment and optimization module based on the output results of S03, S04, and S05, and proposes optimization suggestions based on the data-driven analysis results.

[0059] The data integration and analysis unit receives the results of the emotion classification of teachers and students by the emotion classification module in S03, the results of nine types of indicators for judging the classroom status of students and teachers in S04, and the execution records and effect data of the intervention strategies for adjusting the classroom status of teachers and students in S05. Through data cleaning, feature fusion, and multi-source information alignment, the data integration and analysis unit converts the scattered original data into a structured evaluation data set, providing a unified data basis for subsequent analysis.

[0060] The multi-dimensional evaluation generation unit quantifies and generates multi-dimensional teaching quality indicators based on the integrated data. The classroom participation index reflects the activity and emotional synchronization of teacher-student interaction through the teacher-student emotion synchronization index, interaction frequency index, and interaction quality index, and measures the quality of classroom participation. The calculation formula of the classroom participation index is shown in Equation (19): (19); Where: , and are the weights of the teacher-student emotion synchronization index, interaction frequency index, and interaction quality index respectively, which can be set according to the actual classroom situation; The teaching efficiency score measures the achievement of teaching goals and the learning effect driven by emotion through the positive learning emotion intensity index, negative learning emotion intensity index, positive teaching emotion intensity index, and negative teaching emotion intensity index. The calculation formula of the teaching efficiency score is shown in Equation (20): (20); Where: and are the weights of the positive learning emotion intensity index and negative learning emotion intensity index, and positive teaching emotion intensity index and negative teaching emotion intensity index respectively, which can be set according to the actual classroom situation; The teaching motivation index measures the driving effect of students' learning motivation and teachers' emotions in the classroom through the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index. The calculation formula of the teaching motivation index is shown in Equation (21): (21); Where: , and are the weights of the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index respectively, which can be set according to the actual classroom situation; The interaction-driven emotional stability index combines the interaction frequency index and the interaction quality index to evaluate the inhibitory effect of classroom interaction on the emotional fluctuation index, and quantifies the ability of teacher-student interaction in teaching activities to stabilize the classroom emotional state. The calculation formula of the interaction-driven emotional stability index is shown in Equation (22): (22); Where: , and are the weights of the interaction frequency index, the interaction quality index, and the emotional fluctuation index respectively, and can be set according to the actual classroom situation.

[0061] The optimization suggestion generation unit further analyzes the abnormal or inefficient links of the evaluation indicators, and combines educational theories and historical teaching data to generate targeted optimization suggestions. The optimization suggestion generation unit proposes an operable improvement plan based on the emotional classification results of teachers and students by the emotional classification module in S03, the results of nine types of indicators for judging the classroom status of students and teachers in S04, and the execution records and effect data of the intervention strategies for adjusting the classroom status of teachers and students in S05.

[0062] Since the weights are affected by different factors, each weight can be specifically determined according to the actual situation in practical applications. In the following examples of this patent, each weight is temporarily set to 1.

[0063] Example 1

[0064] Based on the above analysis of the emotion recognition and multi-dimensional teaching quality evaluation system based on artificial intelligence based on emotion classification, the relevant emotions in the classroom teaching of a certain professional course are classified and recognized, and the classified emotion data is intelligently processed and comprehensively analyzed through this system, and finally the output of the classroom status report is realized.

[0065] S01: Real-time collection and fusion of human peripheral physiological signal data First, during the teaching process of the professional course, a customized peripheral physiological signal bracelet is used to real-time collect the human peripheral physiological signals of teachers and students in the classroom environment. Among them, the human peripheral physiological signal data includes the one-axis data of skin resistance, the one-axis data of pulse wave, and the one-axis data of skin temperature. The one-axis data of heart rate can be calculated through the one-axis data of pulse wave. Then, the peripheral physiological signals of teachers and students are spliced and merged. The merged data is in the format of a two-dimensional array, with the pulse wave signal, skin resistance signal, heart rate signal, and skin temperature signal arranged horizontally in sequence, and the sensor data is arranged vertically in chronological order. The collection frequency f is set to 100Hz, that is, the size of the data generated per second is 100 rows and 4 columns.

[0066] S02: Data preprocessing of peripheral physiological signals The peripheral physiological signals of teachers and students are preliminarily processed through noise filters and Min-Max normalization processors to improve the accuracy and efficiency of subsequent analysis. Among them, the noise filter uses a Butterworth low-pass filter to remove high-frequency noise such as sensor jitter, and a Butterworth high-pass filter to remove low-frequency interference such as baseline drift. Subsequently, a fixed length of W =100, sliding step S =50, splitting the continuous sensor data into fixed-length data with a coverage of S / W= 50% of the data segments. Finally, the data segments of the peripheral physiological signal feature information are labeled with corresponding emotional labels. The students' positive learning emotions are labeled as: happy, excited, satisfied, the students' neutral learning emotions are labeled as: calm, indifferent, and the students' positive learning emotions are labeled as: sad, anxious, and angry. The teachers' positive teaching emotions are labeled as: happy, excited, and confident, the teachers' neutral teaching emotions are labeled as: calm, neutral, and the teachers' negative teaching emotions are labeled as: irritable, tired, and frustrated.

[0067] S03: Student and teacher sentiment classification The emotional data fragments after data preprocessing are input into the emotional classification model in batches, the emotional classification model is constructed and the corresponding model parameters are determined. After training, the human emotional classification recognition output is obtained, as follows: S031: Use the multi-scale feature extraction layer to obtain the first emotion data information The multi-scale feature extraction layer includes a parallel convolution kernel unit and an adaptive attention weighting unit. The two units use a selective kernel residual network to process peripheral physiological signal feature information fragments in real time, and extract multi-scale spatiotemporal features through parallel convolution kernels to enhance the diversity of feature expression and generate first emotion data information.

[0068] S032: Use the channel global information interaction layer to obtain the second emotion data information The channel global information interaction layer includes a dual pooling feature extraction unit, a receptive field feature generation unit and a receptive field feature fusion unit. The three units process the first emotion data information in real time through a dynamic multi-branch receptive field attention module, and enhance the global information interaction between channels through channel grouping and dynamic convolution to generate the second emotion data information.

[0069] S033: Use the spatial attention optimization layer to obtain the third emotion data information The spatial attention optimization layer includes a global spatial information aggregation unit and an adaptive channel weighting unit. The two units process the global spatial information of the second emotional data information in real time through lightweight parameter design, and generate channel attention weights through one-dimensional convolution to enhance the adaptive feature weighting between channels, generating the third emotional data information.

[0070] S034: Obtain the fourth emotional data information using a parameter-free feature weighting layer The parameter-free feature weighting layer includes a spatial distribution quantization unit and a statistic-driven weight generation unit. The two units process the spatial distribution characteristics of the third emotional data information in real time through parameter-free design, generate adaptive weights by dynamically calculating the spatial dispersion of features, enhance the feature response of the spatially significant region, and generate the fourth emotional data information.

[0071] S035: Obtain the fifth emotional data information using a residual connection layer The residual connection layer includes a multi-scale feature processing unit and a dynamic dimension matching unit. While retaining the original feature information through residual connection, the two units perform feature fusion on the peripheral physiological signal features and the fourth emotional data information, reduce information loss, and enhance the complementarity of multi-source features, generating the fifth emotional data information.

[0072] S036: Obtain the emotional classification output using an adaptive pooling and classification layer The adaptive pooling and classification layer includes a dynamic spatial compression unit and a fully connected classification unit. The two units compress the feature dimension of the fifth emotional data information through adaptive pooling by the adaptive pooling and classification layer to achieve emotional output.

[0073] S04: Judgment of the classroom states of students and teachers The professional class duration of Teacher A is 40 minutes. Through the collection of the peripheral physiological signals of teachers and students in the classroom by S01 and the preprocessing process of S02, the emotional data of the teachers and students in the class is obtained through the emotional classification model of S03. For Teacher A: the pleasant duration is 8 minutes, the excited duration is 6 minutes, and the confident duration is 2 minutes. That is, the positive teaching emotion duration is 16 minutes, accounting for 40%; the calm duration is 12 minutes, and the neutral duration is 4 minutes. That is, the neutral teaching emotion duration is 16 minutes, accounting for 40%; the irritable duration is 4 minutes, the tired duration is 2 minutes, and the frustrated duration is 2 minutes. That is, the negative teaching emotion duration is 8 minutes, accounting for 20%. Taking Student B as an example for the classroom emotion classification of students, other students will also get data in the same format. For Student B: the happy duration is 6 minutes, the excited duration is 4 minutes, and the satisfied duration is 4 minutes. That is, the positive learning emotion duration is 14 minutes, accounting for 35%; the calm duration is 12 minutes, and the emotionless duration is 8 minutes. That is, the neutral learning emotion duration is 20 minutes, accounting for 50%; the sad duration is 2 minutes, the anxious duration is 3.2 minutes, and the angry duration is 0.8 minutes. That is, the negative learning duration is 6 minutes, accounting for 15%. Teacher A and Student B interacted 30 times (such as asking questions, answering, blackboard writing interaction, eye contact), and 18 of them were accompanied by positive emotions. Assuming that the confidence level of each emotion is 1, for the convenience of calculating the emotional value of the emotional fluctuation index: the three types of positive emotions are all quantified as 1, the two types of neutral emotions are all quantified as 0, and the three types of negative emotions are all quantified as -1.

[0074] S041: Analyze the learning emotions of students using the student emotion dimension analysis unit The student emotion dimension analysis unit focuses on the emotional states of individual students and groups, and quantifies the attention, emotional stability, positive and negative learning emotion impacts in the learning process through the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index, and emotional fluctuation index, to assist in identifying potential problems in teaching content or the environment. Calculated respectively by formulas (4), (6), (8), and (10): ; ; ; ; S042: Analyze the teaching emotions of teachers using the teacher emotion dimension analysis unit The teacher emotion dimension analysis unit is used to evaluate the emotional state and teaching adaptation ability of teachers in the classroom using the positive teaching emotion intensity index and negative teaching emotion intensity index. Calculated respectively by formulas (12) and (14): ; ; S043: Analyze classroom interaction using the classroom interaction dimension analysis unit The classroom interaction dimension analysis unit is used to evaluate the quality of teacher-student emotional interaction and the effectiveness of classroom activity design using the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, providing a basis for improving two-way teaching communication. For the teacher-student emotional synchronization index, it is calculated respectively by formulas (16), (17), and (18): ; ; ; S05: Adjustment of teacher-student classroom state Conduct dynamic teaching intervention based on the output results in S03 and S04 to optimize classroom teaching quality. First, receive the nine types of indicators output by the emotion-driven classroom state analysis module, and combine the real-time classification result data of the emotion classification model. Through the preset threshold rules, key problems in the classroom are identified in real time and intervention signals are immediately sent to the subsequent unit. Subsequently, based on the triggered warning type and real-time data, differentially customized teaching adjustment strategies are dynamically generated. Finally, by recalculating the relevant indicators after the intervention and comparing them with the data before the intervention, the effectiveness of the strategy is verified.

[0075] S06: Evaluation of classroom teaching quality Based on the output results of S03, S04, and S05, multi-dimensional teaching quality evaluation indicators are generated, and optimization suggestions are proposed based on the data-driven analysis results. First, receive the nine types of indicator results of the emotion classification of teachers and students by the emotion classification module in S03, the judgment of the classroom state of students and teachers in S04, and the execution record and effect data of the intervention strategy for the adjustment of the teacher-student classroom state in S05. Through data cleaning, feature fusion, and multi-source information alignment, the data integration and analysis unit converts the scattered original data into a structured evaluation data set, providing a unified data basis for subsequent analysis. Then, based on the integrated data, multi-dimensional teaching quality indicators are quantitatively generated through the classroom participation index, teaching efficiency score, teaching motivation index, and interaction-driven emotional stability index, which are calculated respectively by formulas (19), (20), (21), and (22): ; ; ; ; Finally, further analyze the abnormal or inefficient links of the evaluation indicators, and combine educational theories and historical teaching data to generate targeted optimization suggestions.

[0076] S07: Output of teaching classroom state report Visual report based on the output results of S03, S04, S05, and S06. First, receive and process the results of the sentiment classification module in S03 for teacher and student sentiment classification, the nine-category index results of student and teacher classroom status judgment in S04, the intervention strategy execution records and effect data of teacher-student classroom status adjustment in S05, and the multi-dimensional teaching quality indicators generated quantitatively in S06. Through data cleaning, standardization, and feature alignment, this unit transforms the scattered raw data into structured visual elements, generates core report components, presents the spatio-temporal distribution of emotional states in a heat map, shows the dynamic change trends of key indicators through a timeline, and presents optimization suggestions in a list form. For Teacher A, the positive teaching emotion intensity index and the negative teaching emotion intensity index are 40.0% and 20.0% respectively, and the former is much greater than the latter, indicating that the teacher has a high teaching enthusiasm. The classroom participation index, teaching efficiency score, and teaching motivation index are 63.3%, 80.0%, and 52.5% respectively, indicating that the teacher's teaching has good teaching effects and teaching efficiency for Student B, but the teaching motivation can be further improved. For Student B, the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index, emotional fluctuation index, and interaction-driven emotional stability index are 35.0%, 15.0%, 70.0%, 6.7%, and 9.9% respectively, indicating that the proportion of positive learning emotions of Student B in this course is much greater than the proportion of negative learning emotions, the learning motivation reaches a relatively high level, and the emotional fluctuation is small. In addition, the teacher-student emotion synchronization index, interaction frequency index, and interaction quality index between Student B and Teacher A are 55%, 0.75 (times / minute), and 60% respectively, indicating that Student B can clearly feel Teacher A's teaching method in more than half of the cases, and both the quantity and quality of their interactions reach a relatively high level. The teaching classroom status report will also display the data of other students, and after summing up all the data and taking the average, it can accurately reflect the comprehensive level of this classroom.

[0077] Then, support users to deeply explore data correlations through interactive operations such as filtering and comparison. Finally, accurately push the report to educators, administrators, or parent terminals through multiple channels and collect users' feedback on the report content, and integrate it into the system for optimizing subsequent data processing logic and report generation strategies.

[0078] The specific embodiments of the present invention disclosed above are only for illustration, but the present invention is not limited thereto. Any changes that can be thought of by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. An emotion recognition and multi-dimensional teaching quality evaluation system based on artificial intelligence, characterized in that Including: A physiological signal perception and fusion module, an emotion feature engineering and preprocessing module, an emotion classification module, an emotion-driven classroom state analysis module, a dynamic teaching intervention module, a teaching quality evaluation and optimization module, and an intelligent analysis report generation and feedback module that are connected in sequence; The physiological signal perception and fusion module includes a multi-modal physiological sensing integration unit and a peripheral physiological signal data fusion unit; The emotion feature engineering and preprocessing module includes a signal preprocessing unit, a data segmentation and annotation unit, and a label encoding and feature vectorization unit; The emotion classification module includes a multi-scale feature extraction layer, a spatial attention optimization layer, a channel global information interaction layer, a parameter-free feature weighting layer, and an adaptive pooling and classification layer; The emotion-driven classroom state analysis module includes a student emotion dimension analysis unit, a teacher emotion dimension analysis unit, and a classroom interaction dimension analysis unit; The dynamic teaching intervention module includes a real-time monitoring and warning unit, an intelligent strategy generation unit, a multi-modal intervention execution unit, and an intervention effect evaluation unit; The teaching quality evaluation and optimization module includes a data integration and analysis unit, a multi-dimensional evaluation generation unit, and an optimization suggestion generation unit; The intelligent analysis report generation and feedback module includes a data integration and visualization unit, an interactive analysis unit, and an intelligent distribution and feedback unit.

2. An evaluation method using the artificial intelligence-based emotion recognition and multi-dimensional teaching quality evaluation system according to claim 1, characterized in that, Including the following steps: S01: Real-time collect and fuse human peripheral physiological signal data; S02: Perform data preprocessing on the peripheral physiological signals; The signal preprocessing unit preliminarily processes the peripheral physiological signals of teachers and students through a noise filter and a Min-Max normalization processor to improve the accuracy and efficiency of subsequent analysis. Among them, the noise filter uses a Butterworth low-pass filter to remove sensor jitter or electromagnetic interference, and uses a Butterworth high-pass filter to remove baseline drift interference. The transfer function formula of the Butterworth low-pass filter is shown in formula (1), and the transfer function formula of the Butterworth high-pass filter is shown in formula (2): （1）； Where: N represents the filter order; ω c represents the cut-off frequency; s represents the complex frequency variable of the input signal; （2）； Where: D0 represents the cut-off frequency; represents the distance from the point in the image frequency domain to the center of the spectrum; is the abscissa of the point in the frequency domain; is the ordinate of the point in the frequency domain; The Min-Max normalization processor uses the Min-Max method to unify all physiological signals into the interval [0, 1]; let the input sequence be , where n represents the length of the sequence, and the output sequence after Min-Max normalization is , as shown in formula (3): （3）； Wherein: represents the th value in the sequence, and are the maximum and minimum values of all samples in the input sequence x, respectively; S03: Classify the emotions of students and teachers; S04: Judge the classroom states of students and teachers; S05: Adjust the classroom states of teachers and students; S06: Evaluate the teaching quality of the classroom; S07: Output a report on the teaching classroom state.

3. The evaluation method according to claim 2, wherein: The specific steps of S03 are as follows: S031: Use the multi-scale feature extraction layer to obtain the first emotion data information; S032: Use the channel global information interaction layer to obtain the second emotion data information; S033: Use the spatial attention optimization layer to obtain the third emotion data information; S034: Use the parameter-free feature weighting layer to obtain the fourth emotion data information; S035: Use the residual connection layer to obtain the fifth emotion data information; S036: Use the adaptive pooling and classification layer to obtain the emotion classification output.

4. The evaluation method according to claim 2, wherein: The specific steps of S04 are as follows: S041: Use the student emotion dimension analysis unit to analyze the learning emotions of students; The student emotion dimension analysis unit focuses on the emotional states of individual and group students, quantifies the attention, emotional stability, positive and negative learning emotions in the learning process through the positive learning emotion intensity index, negative learning emotion intensity index, learning motivation index and emotional fluctuation index, and helps to identify potential problems in teaching content or environment. The learning focus weight is a parameter used to quantify external factors affecting students' concentration. The calculation formula of the positive learning emotion intensity index is shown in Equation (4): （4）； Wherein: n represents the number of time segments into which the class time can be divided; represents the i th time segment, and the confidence level of the k th type of positive learning emotion; represents the i th time segment, and the confidence level of the k th type of negative learning emotion; represents the i th time segment, and the confidence level of the k th type of neutral learning emotion; represents the weight of the k th type of positive learning emotion; represents the weight of the k th type of negative learning emotion, represents the weight of the k th type of neutral learning emotion; Indicates the external positive learning weight in the i th time segment, used to quantify the influence of external factors on the positive learning emotion intensity index at the i th time point, weighted and calculated by formula (5): （5）； Wherein: represents the teaching content difficulty at the i th time point, represents the teaching method quality at the i-th time point, represents the classroom session adaptability at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The negative learning emotion intensity index is used to quantify the intensity of negative emotions of students in the classroom, helps teachers identify potential problems in teaching content or environment, and timely adjust teaching strategies. The external negative learning weight is used to quantify the influence weight of students' negative external environment on classroom effects. The calculation formula of the negative learning emotion intensity index is shown in Equation (6): (6)； Wherein: represents the external negative learning weight in the i th time segment, which is used to quantify the influence of external factors on the negative learning emotion intensity index at the i th time point, and is calculated by weighted formula (7): (7)； Wherein: represents the difficulty of the teaching content at the i th time point, represents the interference factors in the classroom environment at the i th time point, represents the negativity of the teacher's feedback at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The learning motivation index is used to evaluate the learning enthusiasm of students driven by positive emotions, exclude the interference of neutral emotions, and quantify the incentive effect of classroom activities on students. The learning motivation weight is a parameter used to quantify external or internal factors affecting students' learning motivation. The formula of the learning motivation index is shown in Equation (8): （8）； Wherein: represents the learning motivation weight in the i th time segment, which is used to quantify the influence of external factors on the learning motivation index at the i th time point, and is calculated by weighted formula (9): （9）； Wherein: represents the interest of students in a subject at the i th time point, represents the teacher incentive measures at the i th time point, represents the family expectations and support at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; The emotional fluctuation index is used to analyze the stability of students' emotional states, and judge the influence of classroom rhythm or external interference on students' emotions through the fluctuation range of emotional values. For each time point, calculate the square of the difference between the emotional value and the average emotional value, multiply by the weight, sum up, take the mean and take the square root to obtain the emotional fluctuation index. The formula of the emotional fluctuation index is shown in Equation (10): （10）； Wherein: represents the weight coefficient in the t th time segment; represents the student emotion value at the t th time point; represents the total classroom time; represents the weighted average emotion value, calculated by formula (11): （11）； S042: Use the teacher emotion dimension analysis unit to analyze the teaching emotions of teachers; The positive teaching emotion intensity index is used to quantify the positive emotion performance of teachers in the classroom, reflect their teaching investment degree and infectivity. The external teaching positive weight is used to quantify the influence weight of teachers' positive emotions on classroom atmosphere. The calculation formula of the positive teaching emotion intensity index is shown in Equation (12): （12）； Wherein: represents the confidence level of the i th type of positive teaching emotion in the k th time segment; represents the confidence level of the i th type of negative teaching emotion in the k th time segment; represents the confidence level of the i th type of neutral teaching emotion in the k th time segment; represents the weight of the k th type of positive teaching emotion; represents the weight of the k th type of negative teaching emotion; represents the weight of the k th type of neutral teaching emotion; Indicates the external positive teaching weight in the i th time segment, used to quantify the influence of external factors on the positive teaching emotion intensity index at the i th time point, weighted and calculated by formula (13): （13）； Wherein: represents the vividness of the teaching language at the i th time point; represents the enthusiasm of classroom interaction at the i th time point; and are the weight coefficients of these two external influencing factors respectively; The negative teaching emotion intensity index is used to evaluate the negative emotions of teachers caused by classroom challenges, and helps teachers adjust teaching strategies. The external negative teaching weight is used to quantify the influence weight of teachers' negative emotions on teaching effects. The calculation formula of the negative teaching emotion intensity index is shown in Equation (14): （14）； Wherein: represents the external negative teaching weight in the i th time segment, which is used to quantify the influence of external factors on the negative learning emotion intensity index at the i th time point, and is weighted and calculated by formula (15): （15）； Wherein: represents the classroom progress pressure at the i th time point, represents the insufficient preparation of teaching content at the i th time point, represents the external interference at the i th time point, , and are the weight coefficients of these three external influencing factors respectively; S043: Use the classroom interaction dimension analysis unit to analyze classroom interaction situations; The teacher-student emotion synchronization index is used to analyze the matching degree of teacher-student emotional states, reflect the consistency of emotional communication and interaction quality in the classroom. Multiply the teacher emotion sequence by the synchronization weight, and calculate the correlation with the student emotion sequence to obtain the teacher-student emotion synchronization index. The calculation formula of the teacher-student emotion synchronization index is shown in Equation (16): （16）； Wherein: represents the emotional value of the teacher at the t th time point; represents the emotional value of the student at the t th time point; represents the weight at the t th time point, reflecting the importance of this moment; represents the weighted average of the teacher's emotional values, represents the weighted average of the student's emotional values; The interaction frequency index is used to count the density of teacher-student interactions per unit time, and evaluate whether the classroom activity design effectively promotes students' participation. The calculation formula of the interaction frequency index is shown in Equation (17): （17）； Wherein: represents the number of teacher-student interactions at the t th time point; The interaction quality index is based on the proportion of interactions marked with positive emotions, and quantifies the positive effect of classroom interactions. The calculation formula of the interaction quality index is shown in Equation (18): （18）； Wherein: represents the number of teacher-student interactions based on positive emotions at the t th time point.

5. The evaluation method according to claim 4, wherein: The specific steps of S06 are as follows: The classroom participation index reflects the activity level and emotional synchronization of teacher-student interaction through the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, and measures the quality of classroom participation. The calculation formula of the classroom participation index is shown in Equation (19): （19）； Wherein: , and are the weights of the teacher-student emotional synchronization index, interaction frequency index, and interaction quality index, respectively; The teaching efficiency score measures the achievement of teaching objectives and the learning effect driven by emotions through the positive learning emotion intensity index, negative learning emotion intensity index, positive teaching emotion intensity index, and negative teaching emotion intensity index. The calculation formula of the teaching efficiency score is shown in Equation (20): （20）； Wherein: and are the weights of the positive learning emotion intensity index, the negative learning emotion intensity index, the positive teaching emotion intensity index, and the negative teaching emotion intensity index, respectively; The teaching motivation index measures the driving effect of students' learning motivation and teachers' emotions in the classroom through the learning motivation index, positive learning emotion intensity index, and positive teaching emotion intensity index. The calculation formula of the teaching motivation index is shown in Equation (21): （21）； Wherein: , and are the weights of the learning motivation index, the positive learning emotion intensity index, and the positive teaching emotion intensity index, respectively; The interaction-driven emotional stability index combines the interaction frequency index and the interaction quality index to evaluate the inhibitory effect of classroom interaction on the emotional fluctuation index, and quantifies the ability of teacher-student interaction in teaching activities to stabilize the classroom emotional state. The calculation formula of the interaction-driven emotional stability index is shown in Equation (22): （22）； Wherein: , and are the weights of the interaction frequency index, the interaction quality index, and the emotional fluctuation index, respectively.

Images