Teaching task management system based on Internet

Abstract

The invention, which relates to the technical field of the teaching task management system, discloses an internet-based teaching task management system comprising a student classroom concentration state identification module, a teaching video slice pushing module, a personalized teaching resource generation module, and an education index analysis module. The student classroom concentration state recognition module reasonably collects and processes images and adopts an improved YOLOv8-SEAM algorithm to improve the recognition precision; the teaching video slice pushing module carries out audio processing and the like on the teaching video, and video clips related to the concentration state of the student are screened out; the personalized teaching resource generation module customizes a personalized teaching resource package in combination with student states and learning histories; and the educational index analysis module generates an individual and class educational index analysis report. According to the invention, the problems that the student class concentration state identification precision is not high and the teaching resource pushing is not targeted are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching task management systems, and in particular to an Internet-based teaching task management system. Background Art

[0002] With the development of Internet technology, the education field is also constantly exploring innovative teaching management models. At present, there have been some studies and applications on smart teaching task management at home and abroad, which assist teachers in completing teaching tasks by monitoring students' classroom concentration status;

[0003] However, the existing smart teaching task management technology still has the following shortcomings: when the student density is high and occlusion is serious, it is easy to misjudge the student's classroom concentration state; the teaching resource push is not targeted, and it is difficult to accurately screen out specific fragments related to the student's classroom concentration state and push relevant targeted knowledge points to each student. Therefore, a solution is urgently needed to solve the shortcomings of the existing technology. Summary of the invention

[0004] The purpose of the present invention is to provide an Internet-based teaching task management system to solve the problems in the above-mentioned background technology.

[0005] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: an Internet-based teaching task management system, comprising: a student classroom concentration state recognition module, a teaching video slice push module, a personalized teaching resource generation module, and an education indicator analysis module;

[0006] The student classroom concentration status recognition module is used to identify the student's concentration status in class and generate a student concentration status report after comprehensive analysis.

[0007] The teaching video slicing push module is used to report on the student's concentration status, filter out specific video clips related to the student's concentration status, focus on marking and extracting the corresponding teaching content clips when the student is distracted or inefficient, and push personalized content based on the student's concentration status;

[0008] The personalized teaching resource generation module combines the students' classroom concentration status recognition results and students' learning history data to analyze students' learning characteristics, knowledge weaknesses and learning styles, and automatically generates personalized teaching resources including teaching video slice resources and related exercise resources in the teaching resource library;

[0009] The education indicator analysis module is used to collect data on students' classroom concentration status, teaching video viewing records, and other aspects, and generate detailed education indicator analysis reports, including individual student reports and class overall reports, to provide teachers with feedback on teaching effectiveness.

[0010] As a further solution of the present invention: the student classroom concentration state recognition module includes the following units and steps:

[0011] The data acquisition unit collects the classroom videos of students recorded by the high-definition camera in the classroom, and extracts frames of the video data according to the set threshold to obtain the classroom images of students;

[0012] The super-resolution image preprocessing unit inputs the student classroom images into a pre-set super-resolution image enhancement model for processing, and outputs a clear enhanced image with more details;

[0013] A data set annotation unit, annotating the enhanced image to generate an enhanced image data set;

[0014] The student classroom concentration state detection unit is based on the YOLOv8 detector. The occlusion-aware attention module (SEAM) and the repulsion loss function (Repulsion Loss) are added to reconstruct the network to obtain the YOLOv8-SEAM detector. The constructed YOLOv8-SEAM model is used to detect the student classroom state, obtain the student's learning state data, and based on this, determine the concentration state of each student during the class time, including distraction, low efficiency, concentration, etc.

[0015] The student concentration status report generating unit comprehensively analyzes the status data obtained through the test and generates a concentration status report for each student.

[0016] Furthermore, in step (1), according to the frame rate of the video and the dynamic degree of the teaching scene, a threshold may be set to extract one frame every three frames and store it as a classroom image;

[0017] Furthermore, in step (2), the super-resolution image enhancement model can use the Swin IR algorithm, take the classroom image as a low-resolution image input, extract shallow features through convolution operations, obtain the basic outline and simple texture information of the image, and further enter the residual Swin Transformer (RSTB) structure formed by multiple series to obtain deep features, focus on the local and global features of the image from different angles, such as the student's face, eyes, and body movements, and combine the convolution layer and the sub-pixel convolution layer to upsample the features to achieve image reconstruction, and output a clear enhanced image with more details, providing a better data basis for subsequent student classroom status detection;

[0018] Furthermore, in step (3), the student status is roughly divided into states such as distraction, inefficiency, and concentration according to the actual situation in the classroom. The LabelImg open source software is used to annotate the student status. The specific annotation is subdivided into states such as listening, raising hands, standing up to answer, reading, writing, whispering, looking around, and looking down in a daze. The enhanced image dataset is stored for training and verification of the constructed student classroom status detection model.

[0019] Furthermore, in step (4), the constructed student classroom status detection model adopts the improved YOLOv8-SEAM algorithm, which is based on the YOLOv8 detector. In order to improve the recognition accuracy of student classroom status under the condition of high student density and severe occlusion, the occlusion-aware attention module (SEAM) and the repulsion loss function (Repulsion Loss) are added. YOLOv8-SEAM consists of four parts: input end, backbone structure, neck and head. The enhanced image dataset is input at the input end, and operations such as Mosaic data enhancement and adaptive anchor frame calculation are used. The backbone structure with C2f (CSPLayer_2Conv) is used for feature extraction to reduce the amount of calculation and accelerate the convergence of the algorithm. The PANet structure of the neck structure is further used to transmit information bidirectionally to enhance the model's detection ability for student targets of different scales. The head part is replaced with a decoupled head structure with a SEAM module. By integrating the occlusion-aware attention mechanism into the detection head design, the network can detect occluded objects while focusing on the decoupled classification and regression tasks. At the same time, a small target detection layer is added to extract the location information and feature details of smaller targets.

[0020] Furthermore, in step (4), the SEAM module learns multi-scale information in space and channels through three depthwise separable convolution operations of different sizes, and outputs the features of the depthwise separable convolution after pointwise convolution. After residual connection, the average pooling operation is input to integrate the global information of the image, and then the expression ability of the target is further enhanced by multiplication after channel expansion to deal with the target occlusion problem.

[0021] Furthermore, the rejection loss function formula in step (4) is as follows:

[0022] Loss = L Attr +αL RepGT +βL RepBox

[0023] Where, L Attr Represents the attraction loss, which is used to reduce the distance between the predicted box and the corresponding real box. RepGT Represents the exclusion loss of the predicted box to other real boxes, which is used to reduce the aggregation of predicted boxes of different targets. RepBoxRepresents the exclusion loss between prediction boxes, which is used to improve the aggregation of prediction boxes of the same target. α and β represent the weight coefficients of the loss.

[0024]

[0025] In the formula, P+ represents the set of positive samples taken according to the set intersection-over-union ratio threshold. Here the threshold is set to 0.5. is the real target box with the maximum intersection-over-union ratio with the detection box p, B P is the predicted box after regression offset from the detection box p. The function formula is as follows:

[0026]

[0027] Where γ is the smoothing parameter;

[0028]

[0029] In the formula, It represents the largest real target box except the real target box that has been matched with the detection box p, and IoG represents the intersection area of ​​the predicted area and the real area divided by the area of ​​the real area. The function can achieve robustness while improving fast convergence. The formula is as follows:

[0030]

[0031] In the formula, δ is a smoothing parameter, and its value range is [0,1);

[0032]

[0033] Where i, j represent the positive sample numbers of the mutually exclusive subsets divided by the two P+ sets traversed respectively, ε is the minimum value, Ι is the identity function, and IoU is the intersection over union ratio calculation, which is the intersection area of ​​the predicted area and the true area divided by their union area;

[0034] Furthermore, in step (5), the learning status data of each student is collated, and the various states of the students in the classroom, such as distraction, inefficiency, and concentration, are screened and marked according to the timeline using the time series analysis method; at the same time, they are classified and summarized, and the duration of each state of the students is calculated by counting the duration of each state, and based on this, a report on the concentration status of each student is generated for the push of teaching video slices.

[0035] As a further solution of the present invention: the teaching video slice generation module includes the following units and steps:

[0036] The audio acquisition unit collects the teacher's classroom teaching video recorded by the high-definition camera in the classroom, extracts the audio track from the video, and uses audio preprocessing to obtain clear and stable audio;

[0037] The speech recognition unit recognizes clear and stable audio, converts the audio into text, and establishes a temporal association between the text and the video;

[0038] The knowledge point acquisition unit extracts keywords from the text to obtain classroom teaching knowledge points, uses natural language processing technology to perform topic analysis on the text, marks the classroom teaching knowledge points in the video, and generates and stores the corresponding information of the classroom knowledge points;

[0039] The teaching video slice resource generation unit selects the video clips of classroom teaching knowledge points related to students’ distraction and inefficiency based on each student’s concentration status report, and generates teaching video slice resources. Students can focus on learning the video clips of classroom teaching knowledge points related to distraction and inefficiency based on the teaching resources.

[0040] Furthermore, in step (1), an audio track is extracted from the video and audio preprocessing is performed. The preprocessing step includes using noise reduction to process the fan sound, air conditioner sound, table and chair moving sound, etc. in the classroom, echo elimination to eliminate the influence of the sound amplification equipment, and gain adjustment to avoid the volume being too high or too low to affect the speech recognition effect. The preprocessed audio is divided into small audio files at a certain interval to facilitate subsequent parallel processing and improve the speech recognition efficiency.

[0041] Furthermore, in step (2), the pre-processed audio file is input into a speech recognition tool to perform speech-to-text conversion, and the text results obtained are stored in chronological order and associated with the corresponding video timestamps;

[0042] Furthermore, in step (3), the text is analyzed using a keyword extraction algorithm to extract key terms and phrases that can represent classroom teaching knowledge points. The TF-IDF algorithm is used to calculate the importance of each word in the text, and words with higher importance are selected as keywords. Natural language processing technology is used to perform topic analysis on the text, and topic models such as the LDA algorithm are used for topic analysis. The text is clustered to determine the topic category to which it belongs, so as to accurately mark the classroom teaching knowledge points of the video. The generated knowledge point corresponding information is stored in a structured manner in the database, including the name of the knowledge point, the subject to which it belongs, the time range of occurrence, etc., to facilitate subsequent query and call;

[0043] Furthermore, in step (4), each student's concentration status report is read, the time period of the student's distraction and low efficiency is screened out, and the video clip corresponding to the time period is accurately located for the subsequent personalized teaching resource generation module;

[0044] As a further solution of the present invention: the personalized teaching resource generation module includes the following units and steps:

[0045] (1) A data acquisition unit, which collects the attention status report data of each student generated by the student classroom attention status recognition module and collects the content of web exercises to generate a teaching resource library;

[0046] (2) The teaching video slice resource push unit analyzes students’ concentration in different subjects and teaching links based on students’ classroom concentration status report data, determines students’ interest in and difficulty in accepting various types of knowledge content, and pushes teaching video slice resources in a targeted manner;

[0047] (3) The unit for pushing relevant exercise resources selects relevant exercises from the teaching resource library based on students’ weak knowledge points and points of interest, and pushes relevant exercise resources in a targeted manner;

[0048] (4) The personalized teaching resource package generation unit customizes each student’s personalized teaching resource package by integrating the teaching video slice resources when students are distracted or inefficient and the relevant exercise resources in the teaching resource library. The resources are classified and sorted according to subjects, knowledge points and learning stages, and resource usage instructions and learning suggestions are attached to guide students to use resources reasonably for learning.

[0049] As a further solution of the present invention: the education indicator analysis module includes units and steps:

[0050] The individual indicator analysis unit calculates the concentration status indicators for each student based on the classroom concentration status data, including the percentage of concentration time, average concentration time, etc. Through the teaching video viewing records, the student's review needs and independent learning ability for specific knowledge points are judged, and the mastery of knowledge points is judged based on the number of views of the knowledge point video;

[0051] The class overall indicator analysis unit determines the key points and difficulties of teaching based on the average indicators of the class's overall classroom concentration status, including the average proportion of class concentration time, the distribution of distraction frequency, etc., and adjusts the teaching process in a timely manner, while understanding the overall learning atmosphere and concentration of the class. The teacher can view the knowledge map and adjust the key points and difficulties. Based on these analysis of the class's performance, a corresponding teaching plan is formulated.

[0052] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0053] (1) Aiming at the problem of low accuracy in identifying students' concentration status in complex teaching scenarios with high student density and severe occlusion, a clear enhanced image with more details is obtained through reasonable image acquisition method and super-resolution image enhancement model (Swin IR algorithm). The improved YOLOv8-SEAM algorithm is used to detect students' classroom status after image enhancement and annotation. The addition of occlusion-aware attention module (SEAM) and repulsion loss function (RepulsionLoss) can accurately handle the misjudgment problem caused by factors such as high density and occlusion, effectively improving the recognition accuracy of students' concentration status in class.

[0054] (2) In order to address the problem of untargeted teaching resource push, we can accurately screen out specific video clips related to students' classroom concentration by performing audio processing, text conversion, knowledge point extraction and annotation on teaching videos. We can then customize a personalized teaching resource package for each student by integrating teaching video slice resources and related exercise resources to meet the different needs of students. At the same time, we can also provide individual education indicator analysis reports and class overall education indicator analysis reports to assist teachers in making teaching decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0056] Figure 1 A flow chart of an Internet-based teaching task management system provided by the present invention;

[0057] Figure 2 A schematic diagram of a module for identifying students' classroom concentration status provided by the present invention;

[0058] Figure 3 A schematic diagram of a teaching video slice push module provided by the present invention;

[0059] Figure 4 A schematic diagram of a personalized teaching resource generation module provided by the present invention;

[0060] Figure 5 This is a schematic diagram of the education indicator analysis module provided by the present invention. DETAILED DESCRIPTION

[0061] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0062] The present invention provides Figure 1 As shown, an Internet-based teaching task management system includes: a student classroom concentration state recognition module, a teaching video slice push module, a personalized teaching resource generation module, and an education indicator analysis module;

[0063] The student classroom concentration status recognition module is used to identify the student's concentration status in class and generate a student concentration status report after comprehensive analysis.

[0064] The teaching video slicing push module is used to report on the student's concentration status, filter out specific video clips related to the student's concentration status, focus on marking and extracting the corresponding teaching content clips when the student is distracted or inefficient, and push personalized content based on the student's concentration status;

[0065] The personalized teaching resource generation module combines the students' classroom concentration status recognition results and students' learning history data to analyze students' learning characteristics, knowledge weaknesses and learning styles, and automatically generates personalized teaching resources including teaching video slice resources and related exercise resources in the teaching resource library;

[0066] The education indicator analysis module is used to collect data on students' classroom concentration status, teaching video viewing records, and other aspects, and generate detailed education indicator analysis reports, including individual student reports and class overall reports, to provide teachers with feedback on teaching effectiveness.

[0067] See also Figure 2 ,The module for identifying students’ classroom concentration status includes the following units and steps:

[0068] The data acquisition unit collects the classroom videos of students recorded by the high-definition camera in the classroom, and extracts frames of the video data according to the set threshold to obtain the classroom images of students;

[0069] The super-resolution image preprocessing unit inputs the student classroom images into a pre-set super-resolution image enhancement model for processing, and outputs a clear enhanced image with more details;

[0070] A data set annotation unit, annotating the enhanced image to generate an enhanced image data set;

[0071] The student classroom concentration state detection unit is based on the YOLOv8 detector. The occlusion-aware attention module (SEAM) and the repulsion loss function (Repulsion Loss) are added to reconstruct the network to obtain the YOLOv8-SEAM detector. The constructed YOLOv8-SEAM model is used to detect the student classroom state, obtain the student's learning state data, and based on this, determine the concentration state of each student during the class time, including distraction, low efficiency, concentration, etc.

[0072] The student concentration status report generating unit comprehensively analyzes the status data obtained through the test and generates a concentration status report for each student.

[0073] In step (1), according to the frame rate of the video and the dynamic degree of the teaching scene, a threshold can be set to extract one frame every three frames and store it as a classroom image;

[0074] In step (2), the super-resolution image enhancement model can use the Swin IR algorithm, take the classroom image as a low-resolution image input, extract shallow features through convolution operation, obtain the basic outline and simple texture information of the image, and further enter the residual Swin Transformer (RSTB) structure formed by multiple series to obtain deep features. Focus on the local and global features of the image from different angles, such as the student's face, eyes, and body movements, and combine the convolution layer and the sub-pixel convolution layer to upsample the features to achieve image reconstruction, and output a clear enhanced image with more details, providing a better data basis for subsequent student classroom status detection;

[0075] In step (3), the student status is roughly divided into states such as distraction, inefficiency, and concentration according to the actual situation in the classroom. The LabelImg open source software is used to annotate the student status. The specific annotation is subdivided into states such as listening, raising hands, standing up to answer, reading, writing, whispering, looking around, and looking down in a daze. The enhanced image dataset is stored for training and verification of the constructed student classroom status detection model;

[0076] In step (4), the constructed student classroom status detection model adopts the improved YOLOv8-SEAM algorithm, which is based on the YOLOv8 detector. In order to improve the recognition accuracy of student classroom status under the conditions of high student density and severe occlusion, the occlusion-aware attention module (SEAM) and the repulsion loss function (Repulsion Loss) are added. YOLOv8-SEAM consists of four parts: input end, backbone structure, neck and head. The enhanced image dataset is input at the input end, and operations such as Mosaic data enhancement and adaptive anchor frame calculation are used. The backbone structure with C2f (CSPLayer_2Conv) is used for feature extraction to reduce the amount of calculation and accelerate the convergence of the algorithm. The PANet structure of the neck structure is further used to transmit information bidirectionally to enhance the model's detection ability for student targets of different scales. The head part is replaced with a decoupled head structure with a SEAM module. By integrating the occlusion-aware attention mechanism into the detection head design, the network can focus on the decoupled classification and regression tasks while realizing the detection of occluded objects. At the same time, a small target detection layer is added to extract the location information and feature details of smaller targets.

[0077] In step (4), the SEAM module learns multi-scale information in space and channels through three depth-separable convolution operations of different sizes, and outputs the features of the depth-separable convolution after point-by-point convolution. After residual connection, the average pooling operation is input to integrate the global information of the image, and then the expression ability of the target is further enhanced by multiplication after channel expansion to deal with the target occlusion problem.

[0078] The rejection loss function formula in step (4) is as follows:

[0079] Loss = L Attr +αL RepGT +βL RepBox

[0080] Where, L Attr Represents the attraction loss, which is used to reduce the distance between the predicted box and the corresponding real box. RepGT Represents the exclusion loss of the predicted box to other real boxes, which is used to reduce the aggregation of predicted boxes of different targets. RepBox Represents the exclusion loss between prediction boxes, which is used to improve the aggregation of prediction boxes of the same target. α and β represent the weight coefficients of the loss.

[0081]

[0082] In the formula, P+ represents the set of positive samples taken according to the set intersection-over-union ratio threshold. Here the threshold is set to 0.5. is the real target box with the maximum intersection-over-union ratio with the detection box p, B P is the predicted box after regression offset from the detection box p. The function formula is as follows:

[0083]

[0084] Where γ is the smoothing parameter.

[0085]

[0086] In the formula, It represents the largest real target box except the real target box that has been matched with the detection box p, and IoG represents the intersection area of ​​the predicted area and the real area divided by the area of ​​the real area. The function can achieve robustness while improving fast convergence. The formula is as follows:

[0087]

[0088] In the formula, δ is a smoothing parameter, and its value range is [0,1);

[0089]

[0090] Where i, j represent the positive sample numbers of the mutually exclusive subsets divided by the two P+ sets traversed respectively, ε is the minimum value, Ι is the identity function, and IoU is the intersection over union ratio calculation, which is the intersection area of ​​the predicted area and the true area divided by their union area;

[0091] In step (5), the learning status data of each student is collated, and the various states of the students in the classroom, such as distraction, inefficiency, and concentration, are screened and marked according to the timeline using the time series analysis method; at the same time, they are classified and summarized, and the duration of each state of the students is calculated by counting the duration of distraction, inefficiency, and concentration. Based on this, a report on the concentration status of each student is generated for the push of teaching video slices.

[0092] See also Figure 3 ,The teaching video slice generation module includes the following units and steps:

[0093] The audio acquisition unit collects the teacher's classroom teaching video recorded by the high-definition camera in the classroom, extracts the audio track from the video, and uses audio preprocessing to obtain clear and stable audio;

[0094] The speech recognition unit recognizes clear and stable audio, converts the audio into text, and establishes a temporal association between the text and the video;

[0095] The knowledge point acquisition unit extracts keywords from the text to obtain classroom teaching knowledge points, uses natural language processing technology to perform topic analysis on the text, marks the classroom teaching knowledge points in the video, and generates and stores the corresponding information of the classroom knowledge points;

[0096] The teaching video slice resource generation unit selects video clips of relevant classroom teaching knowledge points when students are distracted or inefficient based on each student's concentration status report, and generates teaching video slice resources.

[0097] In step (1), an audio track is extracted from the video and audio preprocessing is performed. The preprocessing steps include noise reduction to process the fan sound, air conditioner sound, table and chair movement sound, etc. in the classroom, echo elimination to eliminate the influence of the sound amplification equipment, and gain adjustment to avoid excessive or low volume affecting the speech recognition effect. The preprocessed audio is divided into small audio files at intervals of 10 minutes to facilitate subsequent parallel processing and improve speech recognition efficiency.

[0098] In step (2), the pre-processed audio file is input into a speech recognition tool to convert speech into text. The text results are stored in chronological order and associated with the corresponding video timestamps.

[0099] In step (3), the text is analyzed using a keyword extraction algorithm to extract key terms and phrases that can represent classroom teaching knowledge points. The TF-IDF algorithm is used to calculate the importance of each word in the text, and words with higher importance are selected as keywords. Natural language processing technology is used to perform topic analysis on the text, and topic models such as the LDA algorithm are used for topic analysis. The text is clustered to determine the topic category to which it belongs, so as to accurately mark the classroom teaching knowledge points of the video. The generated knowledge point corresponding information is stored in a structured manner in the database, including the name of the knowledge point, the subject to which it belongs, the time range of occurrence, etc., to facilitate subsequent query and call;

[0100] In step (4), read each student's concentration status report, filter out the time period of students' distraction and low efficiency, and accurately locate the video clip corresponding to the time period. If the distraction state lasts for more than 30 seconds, filter out the video clip of the classroom knowledge point corresponding to the time period for the subsequent personalized teaching resource generation module;

[0101] See also Figure 4 ,The personalized teaching resource generation module includes the following units and steps:

[0102] (1) A data acquisition unit, which collects the attention status report data of each student generated by the student classroom attention status recognition module and collects the content of web exercises to generate a teaching resource library;

[0103] (2) The teaching video slice resource push unit analyzes students’ concentration in different subjects and teaching links based on students’ classroom concentration status report data, determines students’ interest in and difficulty in accepting various types of knowledge content, and pushes teaching video slice resources in a targeted manner;

[0104] (3) The unit for pushing relevant exercise resources selects relevant exercises from the teaching resource library based on students’ weak knowledge points and interest points, and pushes relevant exercise resources in a targeted manner. For example, for the weak function part of mathematics, we select function exercises of different difficulty levels and question types, including basic concept questions, application questions, and extension and improvement questions, to meet students’ needs to consolidate their foundation and improve their abilities;

[0105] (4) The personalized teaching resource package generation unit customizes each student’s personalized teaching resource package by integrating the teaching video slice resources when students are distracted or inefficient and the relevant exercise resources in the teaching resource library. The resources are classified and sorted according to subjects, knowledge points and learning stages, and resource usage instructions and learning suggestions are attached to guide students to use resources reasonably for learning.

[0106] See also Figure 5 ,The education indicator analysis module includes the following units and steps:

[0107] The individual indicator analysis unit calculates the concentration status indicators for each student based on the classroom concentration status data, including the percentage of concentration time, average concentration time, etc. Through the teaching video viewing records, the student's review needs and independent learning ability for specific knowledge points are judged, and the mastery of knowledge points is judged based on the number of views of the knowledge point video;

[0108] The class overall indicator analysis unit determines the key points and difficulties of teaching and adjusts the teaching process in a timely manner based on the average indicators of the class's overall classroom concentration status, including the average proportion of class concentration time, the distribution of distraction frequency, etc., while understanding the overall learning atmosphere and concentration of the class.

[0109] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0110] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A teaching task management system based on the Internet, characterized in that: It includes a module for identifying students' concentration status in class, a module for pushing teaching video slices, a module for generating personalized teaching resources, and a module for analyzing educational indicators. The student classroom concentration state recognition module is used to identify the student's concentration state in class and generate a student concentration state report after comprehensive analysis; The teaching video slice push module is used to report the student's concentration status, filter out specific video clips related to the student's concentration status, focus on marking and extracting the teaching content clips corresponding to when the student is distracted or inefficient, and push personalized content according to the student's concentration status; The personalized teaching resource generation module combines the student's classroom concentration state recognition results and the student's learning history data to analyze the student's learning characteristics, knowledge weaknesses and learning style, and automatically generates personalized teaching resources including teaching video slice resources and related exercise resources in the teaching resource library; The educational indicator analysis module is used to collect data on students' classroom concentration status, teaching video viewing records and other aspects, generate detailed educational indicator analysis reports, including individual student reports and class overall reports, and provide teachers with teaching effect feedback.

2. The Internet-based teaching task management system according to claim 1, characterized in that: The student classroom concentration state recognition module includes the following units and steps: (1) A data acquisition unit collects the classroom video of students recorded by a high-definition camera in the classroom, and extracts frames of the video data according to a set threshold to obtain the classroom image of students; (2) a super-resolution image preprocessing unit, which inputs the student classroom image into a pre-set super-resolution image enhancement model for processing, and outputs a clear enhanced image with more details; (3) a data set annotation unit, annotating the enhanced image to generate an enhanced image data set; (4) The student classroom concentration state detection unit is based on the YOLOv8 detector. The occlusion-aware attention module (SEAM) and the repulsion loss function (Repulsion Loss) are added to reconstruct the network to obtain the YOLOv8-SEAM detector. The constructed YOLOv8-SEAM model is used to detect the student classroom state, obtain the student's learning state data, and based on this, determine the concentration state of each student during class time, including distraction, low efficiency, and concentration. (5) A student concentration status report generation unit, which comprehensively analyzes the status data obtained from the test and generates a concentration status report for each student.

3. The Internet-based teaching task management system according to claim 2, characterized in that: In the step (1), according to the frame rate of the video and the dynamic degree of the teaching scene, the threshold value may be set to extract one frame every three frames and store it as a classroom image; In step (2), the super-resolution image enhancement model can use the SwinIR algorithm, take the classroom image as a low-resolution image input, extract shallow features through convolution operation, obtain the basic outline and simple texture information of the image, and further enter the residual Swin Transformer (RSTB) structure formed by multiple series to obtain deep features, focus on the local and global features of the image from different angles, such as the student's face, eyes, and body movements, combine the convolution layer and the sub-pixel convolution layer to upsample the features to achieve image reconstruction, and output a clear enhanced image with more details, providing a better data basis for subsequent student classroom status detection; In step (3), the student status is roughly divided into states such as distraction, inefficiency, and concentration according to the actual situation in the classroom. The LabelImg open source software is used to annotate the student status. The specific annotation is subdivided into states such as listening, raising hands, standing up to answer, reading, writing, whispering, looking around, and looking down in a daze. The states are stored as an enhanced image dataset for the training and verification of the constructed student classroom status detection model.

4. The Internet-based teaching task management system according to claim 2, characterized in that: In the step (4), the constructed student classroom status detection model adopts the improved YOLOv8-SEAM algorithm, which is based on the YOLOv8 detector. In order to improve the recognition accuracy of student classroom status under the conditions of high student density and severe occlusion, an occlusion-aware attention module (SEAM) and a repulsion loss function (Repulsion Loss) are added; The YOLOv8-SEAM consists of four parts: an input end, a backbone structure, a neck and a head. An enhanced image dataset is input at the input end, and operations such as Mosaic data enhancement and adaptive anchor box calculation are used. A backbone structure with C2f (CSPLayer_2Conv) is used for feature extraction to reduce the amount of calculation and accelerate the convergence of the algorithm. The neck structure of the PANet structure is further used to transmit information bidirectionally to enhance the model's detection ability for student targets of different scales. The head part is replaced with a decoupled head structure with a SEAM module. By integrating the occlusion-aware attention mechanism into the detection head design, the network can detect occluders while focusing on the decoupled classification and regression tasks. At the same time, a small target detection layer is added to extract the location information and feature details of smaller targets.

5. The Internet-based teaching task management system according to claim 4 is characterized in that: In step (4), the SEAM module learns multi-scale information in space and channels through three depth-separable convolution operations of different sizes, outputs the features of the depth-separable convolution after point-by-point convolution, and then inputs the average pooling operation after residual connection to integrate the global information of the image, and then further enhances the expression ability of the target through channel expansion and multiplication to deal with the target occlusion problem; The rejection loss function formula in step (4) is as follows: Loss=L Attr +αL RepGT +βL RepBox Where, L Attr Represents the attraction loss, which is used to reduce the distance between the predicted box and the corresponding real box. RepGT Represents the exclusion loss of the predicted box to other real boxes, which is used to reduce the aggregation of predicted boxes of different targets. RepBox Represents the exclusion loss between prediction boxes, which is used to improve the aggregation of prediction boxes of the same target. α and β represent the weight coefficients of the loss. In the formula, P+ represents the set of positive samples taken according to the set intersection-over-union ratio threshold. Here the threshold is set to 0.

5. is the real target box with the maximum intersection-over-union ratio with the detection box p, B P is the predicted box after regression offset from the detection box p, The function formula is as follows: Where γ is the smoothing parameter; In the formula, It represents the largest real target box except the real target box that has been matched with the detection box p. IoG represents the intersection area of ​​the predicted area and the real area divided by the area of ​​the real area. The function can achieve robustness while improving fast convergence. The formula is as follows: In the formula, δ is a smoothing parameter, and its value range is [0,1); Where i, j represent the positive sample numbers of the mutually exclusive subsets divided by the two P+ sets traversed respectively, ε is the minimum value, Ι is the identity function, and IoU is the intersection over union ratio calculation, which is the intersection area of ​​the predicted area and the true area divided by their union area; In step (5), the learning status data of each student is collated, and the various states of the students in the classroom, such as distraction, inefficiency, and concentration, are screened and marked according to the timeline using the time series analysis method; at the same time, they are classified and summarized, and the duration of each state of the students is calculated by counting the duration of distraction, inefficiency, and concentration. Based on this, a report on the concentration status of each student is generated for the push of teaching video slices.

6. The Internet-based teaching task management system according to claim 1, characterized in that: The teaching video slice generation module includes the following units and steps: (1) An audio acquisition unit that collects the teacher's classroom teaching video recorded by a high-definition camera in the classroom, extracts the audio track from the video, and uses audio preprocessing to obtain clear and stable audio; (2) a speech recognition unit, which recognizes the clear and stable audio, converts the audio into text, and establishes a time association between the text and the video; (3) a knowledge point acquisition unit, extracting keywords from the text to obtain classroom teaching knowledge points, using natural language processing technology to perform topic analysis on the text, marking classroom teaching knowledge points in the video, and generating and storing corresponding information of the classroom knowledge points; (4) The teaching video slice resource generation unit selects the video clips of classroom teaching knowledge points related to the students’ distraction or low efficiency according to each student’s concentration status report, and generates teaching video slice resources. Students can focus on learning the video clips of classroom teaching knowledge points related to the students’ distraction or low efficiency based on the teaching resources.

7. The Internet-based teaching task management system according to claim 6, characterized in that: In step (1), the audio track is extracted from the video using audio preprocessing; The audio preprocessing steps include: using noise reduction to process the fan sound, air conditioner sound, table and chair movement sound, etc. in the classroom, echo elimination to eliminate the impact of the sound amplification equipment, and gain adjustment to avoid excessive or low volume affecting the speech recognition effect. The preprocessed audio is divided into small audio files at a certain interval to facilitate subsequent parallel processing and improve speech recognition efficiency. In step (2), the short audio file after the audio preprocessing is input into the speech recognition tool for speech-to-text conversion, and the obtained text results are stored in chronological order and associated with the corresponding video timestamp.

8. The Internet-based teaching task management system according to claim 6, characterized in that: In step (3), the text is analyzed using a keyword extraction algorithm to extract key terms and phrases that can represent classroom teaching knowledge points, and the TF-IDF algorithm is used to calculate the importance of each word in the text, and words with higher importance are selected as keywords; Use natural language processing technology to conduct topic analysis on the text, use topic models such as the LDA algorithm to perform topic analysis, perform cluster analysis on the text, determine the topic category to which it belongs, so as to accurately mark the classroom teaching knowledge points of the video, and store the corresponding information of the generated knowledge points in a structured manner in the database, including the name of the knowledge point, the topic to which it belongs, the time range of occurrence, etc., to facilitate subsequent query and call; In step (4), each student's concentration status report is read, the time period when the student is distracted or inefficient is screened out, and the video clip corresponding to the time period is accurately located for use in the subsequent personalized teaching resource generation module.

9. The Internet-based teaching task management system according to claim 1, characterized in that: The personalized teaching resource generation module includes the following units and steps: (1) A data acquisition unit, which collects the attention status report data of each student generated by the student classroom attention status recognition module and collects the content of web exercises to generate a teaching resource library; (2) The teaching video slice resource push unit analyzes students’ concentration in different subjects and teaching links based on students’ classroom concentration status report data, determines students’ interest in and difficulty in accepting various types of knowledge content, and pushes teaching video slice resources in a targeted manner; (3) The unit for pushing relevant exercise resources selects relevant exercises from the teaching resource library based on students’ weak knowledge points and points of interest, and pushes relevant exercise resources in a targeted manner; (4) The personalized teaching resource package generation unit customizes each student's personalized teaching resource package by integrating the teaching video slice resources when students are distracted or inefficient and the relevant exercise resources in the teaching resource library. The package is classified and organized according to subject, knowledge point and learning stage, and resource usage instructions and learning suggestions are attached to guide students to make rational use of resources for learning.

10. The Internet-based teaching task management system according to claim 1, characterized in that: The education indicator analysis module includes units and steps: The individual indicator analysis unit calculates the concentration status indicators for each student based on the classroom concentration status data, including the percentage of concentration time, average concentration time, etc. It also determines the students' review needs and autonomous learning ability for specific knowledge points through the teaching video viewing records, and determines the mastery of knowledge points based on the number of views of the knowledge point videos; The class overall indicator analysis unit determines the key points and difficulties of teaching based on the average indicators of the class's overall classroom concentration status, including the average proportion of class concentration time, the distribution of distraction frequency, etc., and adjusts the teaching process in a timely manner. At the same time, it understands the overall learning atmosphere and concentration of the class. The teacher can view the knowledge map and adjust the key points and difficulties. Based on these analyses of the class's performance, a corresponding teaching plan is formulated.

Images