A classroom concentration assessment system and method based on multi-modal data perception

Abstract

This invention discloses a classroom attention assessment system and method based on multimodal data perception. The invention relates to the field of classroom attention assessment, and includes a module for anchoring classroom teaching timelines and collecting multimodal, non-intrusive data; a module for time-anchored multimodal data fusion and attention feature extraction; and a module for hierarchical attention assessment and closed-loop feedback optimization. The classroom attention assessment system and method described in this invention automatically identifies the timeline nodes of four core teaching segments by analyzing the teacher's voiceprint features, speech rate changes, pause durations, and semantic keywords. This divides the complete classroom into segmented timeline intervals that perfectly match the teaching behavior. It not only accurately outputs individual student and overall class attention data for each teaching segment but also allows teachers to monitor the changes in student attention under different teaching content and methods in real time.

Description

Technical Field

[0001] This invention relates to the field of classroom attention assessment, and in particular to a classroom attention assessment system and method based on multimodal data perception. Background Technology

[0002] With the deep implementation of smart education systems, the refined and personalized management of classroom teaching has become a core direction of digital transformation in education. Student classroom focus, as a core indicator directly reflecting teaching effectiveness and student engagement, requires precise, efficient, and seamless assessment. This is a crucial prerequisite for optimizing classroom instruction design, improving teaching quality, and achieving personalized learning intervention. Currently, technologies for assessing classroom focus are gradually being applied, but existing solutions all suffer from insurmountable technical defects and application limitations. They cannot adapt to the full-scenario needs of routine classroom teaching, nor can they achieve deep integration between assessment and teaching processes.

[0003] However, existing technologies generally suffer from the core problem of lacking time-series anchoring, resulting in a complete disconnect between assessment results and actual teaching processes. Most current attention assessment technologies use a uniform assessment standard for the entire lesson, only outputting the average attention level for the whole class. They cannot accurately link these results to different teaching stages such as teacher explanations of knowledge points, classroom questioning, group discussions, and in-class exercises. They cannot identify the dynamic changes in student attention levels across different teaching stages, leading to assessment results that cannot effectively support teachers in adjusting the pace of instruction in real time, resulting in poor practicality of the assessment results.

[0004] Therefore, it is necessary to propose a classroom attention assessment system and method based on multimodal data perception to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide a classroom attention assessment system and method based on multimodal data perception, which can effectively solve the problems in the background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A classroom attention assessment system based on multimodal data perception includes a classroom teaching time-series anchoring and multimodal non-sensory data acquisition module, a time-series anchored multimodal data fusion and attention feature extraction module, and an attention stratified assessment and closed-loop feedback optimization module. The classroom teaching time-series anchoring and multimodal non-sensory data acquisition module is used to provide benchmark anchor points and standardized data sources. The time-anchored multimodal data fusion and attention feature extraction module is based on the anchor points and data of the classroom teaching time-anchored and multimodal non-sensory data acquisition module, and is used to achieve accurate extraction of effective attention features. The attention level stratification assessment and closed-loop feedback optimization module is used to implement the assessment results and iteratively optimize the entire system.

[0007] Preferably, the classroom teaching time sequence anchoring and multimodal seamless data acquisition module includes a classroom teaching key time sequence node calibration submodule and a multimodal seamless synchronous data acquisition submodule, wherein the classroom teaching key time sequence node calibration submodule: By collecting the teacher's voice data throughout the entire teaching process using the classroom's existing audio pickup equipment, and based on the teacher's voiceprint characteristics, speech rate changes, pause duration, and semantic keywords, four fixed teaching time sequence nodes are automatically identified: the start of knowledge point explanation, the initiation of classroom questions, the start of group discussions, and the start of classroom exercises. A unique timestamp is generated for each identified node, dividing the complete classroom into segmented time sequence intervals that perfectly match the teaching process. All subsequent data collection, processing, and evaluation operations are based on this time sequence node as the unique benchmark anchor point.

[0008] Preferably, the multimodal non-sensory synchronous data acquisition submodule relies on the monitoring cameras, desktop pressure sensing pads, and omnidirectional sound pickup devices already deployed in the classroom to synchronously collect three types of data bound to corresponding timestamps within each calibrated time sequence node interval: extracting only the facial action area and not collecting the student's facial micro-movement sequence data of the complete face; Data on the distribution of sitting pressure, which is collected by a desktop pressure sensing pad and reflects changes in students' body posture. The voice interaction data of students speaking in class and participating in group discussions achieves complete synchronization of the timeline of multimodal data.

[0009] Preferably, the time-series anchored multimodal data fusion and focus feature extraction module includes a temporal node matching modality weight adaptive allocation submodule and a multimodal data association feature extraction submodule, wherein the temporal node matching modality weight adaptive allocation submodule includes: Based on the pre-defined teaching timeline nodes, differentiated calculation weights are automatically assigned to the three types of collected data for different types of teaching nodes. Specifically, these weights include: Within the knowledge point explanation nodes, facial micro-movement sequence data is assigned the highest weight; Within classroom question-and-answer sessions, voice interaction data is assigned the highest weight. Within the group discussion node, changes in body posture and voice interaction data are assigned the same highest weight; Within classroom practice sessions, data on changes in body posture are assigned the highest weight.

[0010] Preferably, the multimodal data association feature extraction submodule extracts cross-modal association features for three types of synchronous data within the same time-series node interval based on the weights already allocated by the modal weight adaptive allocation submodule for temporal node matching. Specifically, it extracts the linkage change features of different modal data within the same time-series interval.

[0011] Preferably, the attention level stratification assessment and closed-loop feedback optimization module includes a time-series segmented attention level stratification assessment submodule and a dual-ended closed-loop feedback optimization submodule, wherein the time-series segmented attention level stratification assessment submodule specifically includes: Based on the effective correlation features extracted in the preceding sequence, the node focus level of individual students and the node focus distribution data of the whole class are generated according to each labeled time-series node interval. At the same time, the accurate attribution results of focus abnormality are generated simultaneously, clearly marking the core triggering factors of focus abnormality of individual students in the corresponding teaching node.

[0012] Preferably, the dual-ended closed-loop feedback optimization submodule specifically includes: The stratified assessment results of the time-series segmented attention level assessment submodule are synchronously output to the teacher's end and the student's end. The teacher's end pushes the class attention level distribution data of the current teaching node in real time, providing a real-time basis for teachers to adjust the teaching pace. After class, students receive a personalized full-term focus report and improvement suggestions.

[0013] A classroom attention assessment method based on multimodal data perception includes the following steps: S1: Pre-class adaptation and rule initialization, complete the time axis synchronous calibration of the existing classroom surveillance camera, omni-channel sound pickup device and desktop pressure sensor pad, pre-configure the recognition rules of the core teaching links, clarify the speech feature judgment criteria of the four links of knowledge point instruction, classroom questioning, group discussion and in-class exercise, and complete the baseline parameter configuration before system operation; S2: Real-time classroom data collection and dynamic anchoring of each segment. After the lesson begins, the system seamlessly collects three types of synchronous data: facial micro-movement sequence data of students without complete facial information, desktop pressure distribution data mapping body posture, and student classroom interaction voice data. At the same time, based on the voiceprint, speech rate, pauses and semantic features of the teacher's lecture voice, the start and end timestamps of each teaching segment are dynamically marked in real time, and all collected data are bound to the timestamps of the corresponding segments. S3: Feature extraction and hierarchical evaluation for stage adaptation. For the currently calibrated teaching stages, the system automatically adapts to differentiated multimodal data weight allocation rules, performs cross-modal linkage feature extraction on multi-source synchronous data within the same stage, distinguishes effective focus behavior from non-focus behavior through multi-data linkage matching, generates individual student focus level and class overall focus distribution data for each teaching stage, and simultaneously marks the core triggering factors of focus abnormality. S4: Dual-end feedback and closed-loop optimization. During the teaching process, the system pushes the class's focus data for the current segment to the teacher in real time, providing a real-time basis for adjusting the teaching pace. After class, a personal full-process focus report and improvement suggestions are pushed to the student's device. At the same time, the full-process evaluation data is fed back to the process calibration unit to optimize the recognition accuracy of subsequent teaching processes.

[0014] Compared with existing technologies, this invention provides a classroom attention assessment system and method based on multimodal data perception, which has the following beneficial effects: This classroom attention assessment system and method based on multimodal data perception automatically identifies the temporal nodes of four core teaching segments by analyzing the teacher's voiceprint features, speech rate changes, pause duration, and semantic keywords. It divides the complete classroom into segmented temporal intervals that perfectly match the teaching behavior. All data collection, processing, and assessment operations use these temporal nodes as the sole benchmark anchor point. This not only accurately outputs individual student and overall class attention data for each teaching segment but also allows teachers to monitor the changes in student attention under different teaching content and methods in real time. This provides teachers with precise and real-time data support for adjusting the teaching pace and optimizing instructional design, significantly improving the practicality and guidance of the assessment results.

[0015] This classroom attention assessment system and method based on multimodal data perception binds all multimodal data to the timestamps of each teaching segment through a time-anchored collection method. This provides a precise benchmark for subsequent adaptive weight allocation and cross-modal feature extraction, laying the core foundation for the high-precision assessment of the entire system and avoiding the distortion problem of fixed-duration segmented assessment from the source.

[0016] This classroom attention assessment system and method based on multimodal data perception fundamentally optimizes the processing logic of multimodal data by linking a temporal node matching modality weight adaptive allocation submodule with a multimodal data association feature extraction submodule. This significantly improves the accuracy of attention assessment while completely avoiding the risk of student privacy leakage. By adaptively adjusting the weights of different modal data based on the core behavioral characteristics of different teaching stages, the system assigns the highest weight to facial micro-movement sequence data at knowledge point explanation stages, the highest weight to voice interaction data at classroom questioning stages, equal and highest weight to body posture changes and voice interaction data at group discussion stages, and the highest weight to body posture change data at classroom practice stages. This completely solves the assessment bias problem under different teaching scenarios and significantly reduces the misjudgment rate of scenario adaptation.

[0017] This classroom attention assessment system and method based on multimodal data perception only extracts data from the facial movement regions of students during visual data collection, without collecting or storing complete facial images, thus avoiding the risk of student facial privacy leakage from the source of data collection. By analyzing the interconnected changes in different modalities within the same time interval, it accurately distinguishes between the effective attention characteristics of looking down while taking notes (corresponding to stable posture and downward gaze) and the inattentive characteristics of looking down without hand movements (corresponding to frequent changes in posture and shifting gaze).

[0018] This classroom attention assessment system and method based on multimodal data perception achieves bidirectional closed-loop optimization and full-process value release through the design of a dual-end closed-loop feedback optimization submodule. By providing real-time feedback to teachers and personalized feedback to students after class, a bidirectional closed loop is realized, enabling real-time intervention in classroom teaching and personalized improvement for students. During the lesson, teachers are provided with real-time data on the class's attention distribution at the current teaching stage, providing real-time basis for dynamically adjusting the teaching pace and switching teaching methods. After class, students receive individual attention reports and improvement suggestions for each stage, directly transforming attention assessment results into effective tools for teaching optimization and learning improvement, fully releasing the application value of the assessment data.

[0019] This classroom attention assessment system and method based on multimodal data perception can feed back the assessment results of the entire process to the calibration sub-module of key time nodes in classroom teaching. This can continuously optimize the calibration accuracy of teaching nodes. The improvement in calibration accuracy will further drive the overall improvement of the adaptability of weight allocation, the accuracy of feature extraction, and the accuracy of assessment results. Based on this, the system's assessment capabilities can be continuously optimized with the increase in usage frequency, and can adapt to the classroom teaching characteristics of different teachers, different subjects, and different grade levels. Attached Figure Description

[0020] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0022] Example 1: like Figure 1 As shown, a classroom attention assessment system based on multimodal data perception includes a classroom teaching time-series anchoring and multimodal non-sensory data acquisition module, a time-series anchored multimodal data fusion and attention feature extraction module, and an attention stratified assessment and closed-loop feedback optimization module. The classroom teaching time-series anchoring and multimodal non-sensory data acquisition module is used to provide benchmark anchor points and standardized data sources. The time-anchored multimodal data fusion and attention feature extraction module is based on the anchor points and data of the classroom teaching time-anchored and multimodal non-sensory data acquisition module, and is used to achieve accurate extraction of effective attention features; The focus level assessment and closed-loop feedback optimization module is used to implement the assessment results and iteratively optimize the entire system.

[0023] The classroom teaching time sequence anchoring and multimodal seamless data acquisition module includes a classroom teaching key time sequence node calibration submodule and a multimodal seamless synchronous data acquisition submodule. The classroom teaching key time sequence node calibration submodule includes: By collecting the teacher's voice data throughout the entire teaching process using the classroom's existing audio pickup equipment, and based on the teacher's voiceprint characteristics, speech rate changes, pause duration, and semantic keywords, four fixed teaching time sequence nodes are automatically identified: the start of knowledge point explanation, the initiation of classroom questions, the start of group discussions, and the start of classroom exercises. A unique timestamp is generated for each identified node, dividing the complete classroom into segmented time sequence intervals that perfectly match the teaching process. All subsequent data collection, processing, and evaluation operations use this time sequence node as the unique benchmark anchor point, achieving deep integration of the entire process data with the teaching process.

[0024] The multimodal seamless synchronous data acquisition submodule relies on the surveillance cameras, desktop pressure sensing pads, and omni-channel audio pickup devices already deployed in the classroom. It does not require students to be equipped with any wearable or terminal devices, and achieves completely seamless data acquisition. Within each calibrated time sequence node interval, it synchronously collects three types of data bound to the corresponding timestamps: extracting only the facial action area and collecting student facial micro-movement sequence data without collecting the complete face. Data on the distribution of sitting pressure, which is collected by a desktop pressure sensing pad and reflects changes in students' body posture. The voice interaction data of students speaking in class and participating in group discussions achieves complete synchronization of the timeline of multimodal data.

[0025] The temporally anchored multimodal data fusion and focus feature extraction module includes a temporal node matching modality weight adaptive allocation submodule and a multimodal data association feature extraction submodule. The temporal node matching modality weight adaptive allocation submodule includes: Based on the pre-defined teaching timeline nodes, differentiated calculation weights are automatically assigned to the three types of collected data for different types of teaching nodes. Specifically, these weights include: Within the knowledge point explanation nodes, facial micro-movement sequence data is assigned the highest weight; Within classroom question-and-answer sessions, voice interaction data is assigned the highest weight. Within the group discussion node, changes in body posture and voice interaction data are assigned the same highest weight; Within classroom practice sessions, data on changes in body posture are assigned the highest weight. The weighting is automatically adjusted synchronously as teaching sessions switch, with no fixed weighting settings.

[0026] The multimodal data association feature extraction submodule uses the modal weight adaptive allocation submodule based on temporal node matching to perform cross-modal association feature extraction on three types of synchronous data within the same temporal node interval. Specifically, it extracts the linkage change features of different modal data within the same temporal interval. Through linkage feature matching, the linkage feature of looking down while taking notes and stable sitting posture is marked as effective focus feature, while the linkage feature of looking down without hand movements and frequent changes in sitting posture is marked as non-focus feature, thus completing the accurate selection of effective features.

[0027] The attention stratification assessment and closed-loop feedback optimization module includes a time-series segmented attention stratification assessment submodule and a dual-ended closed-loop feedback optimization submodule. The time-series segmented attention stratification assessment submodule specifically includes: Based on the effective correlation features extracted in the preceding sequence, the system generates individual student attention level and class-wide attention distribution data for each labeled time-series node interval. Simultaneously, it generates accurate attribution results for attention abnormalities, clearly identifying the core triggering factors for individual student attention abnormalities within the corresponding teaching node, rather than outputting a single overall classroom attention score, thus achieving refined stratified assessment based on teaching segments.

[0028] The dual-ended closed-loop feedback optimization submodule specifically includes: The stratified assessment results of the time-series segmented attention level assessment submodule are synchronously output to the teacher's end and the student's end. The teacher's end pushes the class attention level distribution data of the current teaching node in real time, providing a real-time basis for teachers to adjust the teaching pace. After class, students receive a personalized full-process focus report and improvement suggestions. Simultaneously, this sub-module feeds back the evaluation results of this entire process to the key time-series node calibration sub-module, optimizing the calibration accuracy of subsequent teaching nodes. This forms a closed-loop process of data collection, processing, evaluation, feedback, and optimization, achieving synergistic efficiency among all modules.

[0029] Example 2: A classroom attention assessment method based on multimodal data perception includes the following steps: S1: Pre-class adaptation and rule initialization, complete the time axis synchronous calibration of the existing classroom surveillance camera, omni-channel sound pickup device and desktop pressure sensor pad, pre-configure the recognition rules of the core teaching links, clarify the speech feature judgment criteria of the four links of knowledge point instruction, classroom questioning, group discussion and in-class exercise, and complete the baseline parameter configuration before system operation; S2: Real-time classroom data collection and dynamic anchoring of each segment. After the lesson begins, the system seamlessly collects three types of synchronous data: facial micro-movement sequence data of students without complete facial information, desktop pressure distribution data mapping body posture, and student classroom interaction voice data. At the same time, based on the voiceprint, speech rate, pauses and semantic features of the teacher's lecture voice, the start and end timestamps of each teaching segment are dynamically marked in real time, and all collected data are bound to the timestamps of the corresponding segments. S3: Feature extraction and hierarchical evaluation for stage adaptation. For the currently calibrated teaching stages, the system automatically adapts to differentiated multimodal data weight allocation rules, performs cross-modal linkage feature extraction on multi-source synchronous data within the same stage, distinguishes effective focus behavior from non-focus behavior through multi-data linkage matching, generates individual student focus level and class overall focus distribution data for each teaching stage, and simultaneously marks the core triggering factors of focus abnormality. S4: Dual-end feedback and closed-loop optimization. During the teaching process, the system pushes the class's focus data for the current segment to the teacher in real time, providing a real-time basis for adjusting the teaching pace. After class, a personal full-process focus report and improvement suggestions are pushed to the student's device. At the same time, the full-process evaluation data is fed back to the process calibration unit to optimize the recognition accuracy of subsequent teaching processes.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A classroom attention assessment system based on multimodal data perception, comprising a classroom teaching time-series anchoring and multimodal non-sensory data acquisition module, a time-series anchored multimodal data fusion and attention feature extraction module, and a attention stratified assessment and closed-loop feedback optimization module, characterized in that: The classroom teaching time sequence anchoring and multimodal non-sensory data acquisition module is used to provide benchmark anchor points and standardized data sources; The time-anchored multimodal data fusion and attention feature extraction module is based on the anchor points and data of the classroom teaching time-anchored and multimodal non-sensory data acquisition module, and is used to achieve accurate extraction of effective attention features. The attention level stratification assessment and closed-loop feedback optimization module is used to implement the assessment results and iteratively optimize the entire system.

2. The classroom attention assessment system based on multimodal data perception according to claim 1, characterized in that: The classroom teaching time sequence anchoring and multimodal seamless data acquisition module includes a classroom teaching key time sequence node calibration submodule and a multimodal seamless synchronous data acquisition submodule, wherein the classroom teaching key time sequence node calibration submodule: By collecting the teacher's voice data throughout the entire teaching process using the classroom's existing audio pickup equipment, and based on the teacher's voiceprint characteristics, speech rate changes, pause duration, and semantic keywords, four fixed teaching time sequence nodes are automatically identified: the start of knowledge point explanation, the initiation of classroom questions, the start of group discussions, and the start of classroom exercises. A unique timestamp is generated for each identified node, dividing the complete classroom into segmented time sequence intervals that perfectly match the teaching process. All subsequent data collection, processing, and evaluation operations are based on this time sequence node as the unique benchmark anchor point.

3. The classroom attention assessment system based on multimodal data perception according to claim 2, characterized in that: The multimodal non-sensory synchronous data acquisition submodule relies on the surveillance cameras, desktop pressure sensing pads, and omnidirectional sound pickup devices already deployed in the classroom to synchronously collect three types of data bound to corresponding timestamps within each calibrated time sequence node interval: extracting only the facial action area and collecting student facial micro-movement sequence data without collecting the complete face. Data on the distribution of sitting pressure, which maps changes in students' body posture, was collected using a desktop pressure-sensing pad. The voice interaction data of students speaking in class and participating in group discussions achieves complete synchronization of the timeline of multimodal data.

4. A classroom attention assessment system based on multimodal data perception according to claim 2, characterized in that: The time-series anchored multimodal data fusion and focus feature extraction module includes a temporal node matching modality weight adaptive allocation submodule and a multimodal data association feature extraction submodule, wherein the temporal node matching modality weight adaptive allocation submodule includes: Based on the pre-defined teaching timeline nodes, differentiated calculation weights are automatically assigned to the three types of collected data for different types of teaching nodes. Specifically, these weights include: Within the knowledge point explanation nodes, facial micro-movement sequence data is assigned the highest weight; Within classroom question-and-answer sessions, voice interaction data is assigned the highest weight. Within the group discussion node, changes in body posture and voice interaction data are assigned the same highest weight; Within classroom practice sessions, data on changes in body posture are assigned the highest weight.

5. A classroom attention assessment system based on multimodal data perception according to claim 4, characterized in that: The multimodal data association feature extraction submodule, based on the modal weight adaptive allocation submodule assigned weights by the temporal node matching submodule, performs cross-modal association feature extraction on three types of synchronous data within the same temporal node interval. Specifically, it extracts the linkage change features of different modal data within the same temporal interval.

6. A classroom attention assessment system based on multimodal data perception according to claim 2, characterized in that: The attention level stratification assessment and closed-loop feedback optimization module includes a time-series segmented attention level stratification assessment submodule and a dual-ended closed-loop feedback optimization submodule, wherein the time-series segmented attention level stratification assessment submodule specifically includes: Based on the effective correlation features extracted in the preceding sequence, the node focus level of individual students and the node focus distribution data of the whole class are generated according to each labeled time-series node interval. At the same time, the accurate attribution results of focus abnormality are generated simultaneously, clearly marking the core triggering factors of focus abnormality of individual students in the corresponding teaching node.

7. A classroom attention assessment system based on multimodal data perception according to claim 6, characterized in that: The dual-ended closed-loop feedback optimization submodule specifically includes: The stratified assessment results of the time-series segmented attention level assessment submodule are synchronously output to the teacher's end and the student's end. The teacher's end pushes the class attention level distribution data of the current teaching node in real time, providing a real-time basis for teachers to adjust the teaching pace. After class, students receive a personalized full-term focus report and improvement suggestions.

8. A classroom attention assessment method based on multimodal data perception, employing a classroom attention assessment system based on multimodal data perception as described in any one of claims 1-7, characterized in that: The following steps are included: S1: Pre-class adaptation and rule initialization, complete the time axis synchronous calibration of the existing classroom surveillance camera, omni-channel sound pickup device and desktop pressure sensor pad, pre-configure the recognition rules of the core teaching links, clarify the speech feature judgment criteria of the four links of knowledge point instruction, classroom questioning, group discussion and in-class exercise, and complete the baseline parameter configuration before system operation; S2: Real-time classroom data collection and dynamic anchoring of each segment. After the lesson begins, the system seamlessly collects three types of synchronous data: facial micro-movement sequence data of students without complete facial information, desktop pressure distribution data mapping body posture, and student classroom interaction voice data. At the same time, based on the voiceprint, speech rate, pauses and semantic features of the teacher's lecture voice, the start and end timestamps of each teaching segment are dynamically marked in real time, and all collected data are bound to the timestamps of the corresponding segments one by one. S3: Feature extraction and hierarchical evaluation for stage adaptation. For the currently calibrated teaching stages, the system automatically adapts to differentiated multimodal data weight allocation rules, performs cross-modal linkage feature extraction on multi-source synchronous data within the same stage, distinguishes effective focus behavior from non-focus behavior through multi-data linkage matching, generates individual student focus level and class overall focus distribution data for each teaching stage, and simultaneously marks the core triggering factors of focus abnormality. S4: Dual-end feedback and closed-loop optimization. During the teaching process, the system pushes the class's focus data for the current segment to the teacher in real time, providing a real-time basis for adjusting the teaching pace. After class, a personal full-process focus report and improvement suggestions are pushed to the student's device. At the same time, the full-process evaluation data is fed back to the process calibration unit to optimize the recognition accuracy of subsequent teaching processes.

Images