Physical and chemical practical operation AI auxiliary paper marking method and system based on multi-modal information event detection

Abstract

The invention relates to the field of physical and chemical practical operation examination paper marking, in particular to a physical and chemical practical operation AI auxiliary paper marking method and system based on multi-modal information event detection, and solves the technical problems that an existing paper marking system is low in efficiency and poor in accuracy. According to the system, user roles are divided into examination center personnel, project group leader, task question leader, paper marking personnel and arbitration personnel, different function permissions are provided for different roles, so that the paper marking efficiency is improved, meanwhile, an AI + manual scoring mode is adopted, a multi-modal information event detection method is adopted in AI scoring, and the scoring efficiency is improved. According to the method, the event can be accurately detected, the function and application description of the event can be effectively processed, meanwhile, multi-modal information is fully utilized, a more accurate event detection result is provided, and scoring is more accurate and efficient.

Description

Technical Field

[0001] The present invention relates to the field of physics and chemistry practical examination marking, and specifically to a physics and chemistry practical examination AI-assisted marking method and system based on multimodal information event detection. Background Art

[0002] The physics, chemistry and biology experimental operation test is a way to assess students' experimental operation ability in physics, chemistry and biology. In traditional physics and chemistry practical examinations, teachers need to spend a lot of time and energy to observe and judge students' experimental operations one by one. Especially in large-scale examinations or centralized examinations organized by schools, the workload of marking is huge. With the development of information technology, many marking systems have appeared on the market, but there are some common shortcomings: 1. The existing system does not set different permissions for different personnel according to the nature of the practical examination. Since it is impossible to assign different permissions to roles, there will be unclear division of labor in the marking process, which will lead to low marking efficiency and poor accuracy; 2. In the process of AI automatic grading in the marking system, event recognition is the key, but existing image event detection methods generally rely on target detection, that is, identifying specific target objects in the image and inferring the occurrence of events. However, these methods have some problems and limitations. First, after detecting the target object, traditional target detection methods usually require many rules and conditions to be set manually to further determine whether an event will occur, especially in the fields of physics, chemistry, and biology experimental operation examinations. Secondly, traditional methods mainly focus on the spatial position and temporal relationship of the target, and lack effective processing methods for the function and use description of the target object, so inaccurate or incomplete situations may occur when inferring events. Summary of the invention

[0003] The present invention mainly solves the technical problems of low efficiency and poor accuracy of existing marking systems, and provides a physical and chemical practical AI-assisted marking method and system based on multimodal information event detection.

[0004] The present invention provides a method for physics and chemistry practical AI-assisted examination paper marking based on multimodal information event detection, comprising the following steps: S10: Data collection and preprocessing Multiple cameras installed in the examination room can be used to collect videos of candidates' experimental operations from different angles; Extract key frames from the experimental operation video and perform image preprocessing on the key frames; S20: Multimodal feature extraction and model training The pre-processed key frame images are processed using a target detection algorithm to determine the target object, so as to extract different modal features in the experimental process, and then the different modal features are represented by multimodal information; the target object is fused with multimodal information and processed to form a multidimensional tensor; the processed multidimensional tensor is used as basic data to train a deep learning model of a convolutional neural network. The trained deep learning model of the convolutional neural network can classify the operation events in the new experimental operation video, and determine whether it is a correct operation or an incorrect operation, as well as the specific operation type; S30: Scoring rules and automatic scoring According to the specific requirements of the experiment and the syllabus, formulate detailed scoring rules; the scoring rules include the following scoring items: correct use of experimental equipment, completeness or accuracy of experimental steps, rationality of experimental data, and correctness of experimental results; assign different weights to each scoring item according to its importance; Automatic scoring is performed based on the features extracted from the experimental operation video and the formulated scoring rules; for each scoring item, a score is given based on the correctness and conformity of its features, and the final score is the weighted sum of the scores of each scoring item.

[0005] The further step S20 comprises the following steps: S201: Object detection and functional usage description extraction The pre-processed key frame images are processed using a target detection algorithm to detect multiple target objects in them; each target object is accompanied by a text description related to its function and purpose; these text descriptions are descriptions of the properties, behaviors or uses of the target object; these descriptions are spliced ​​together to form an overall text description, which will be used for subsequent processing (equivalent to providing language information about the characteristics of the target object); S202: Processing of text description The generated text description is sent to a pre-trained language model, and the language model converts the text description into a multi-dimensional tensor 1, wherein the multi-dimensional tensor 1 can capture semantic information and context information in the text description and provide feature representation for subsequent event detection; S203: Image processing using multi-layer residual convolutional neural network The same key frame image in S201 is processed by a multi-layer residual convolutional neural network and converted into a multi-dimensional tensor 2; the multi-dimensional tensor 2 contains the visual information captured in the key frame image, including the shape, position and appearance information of the target object (equivalent to the visual information of the target object characteristics); S204: Fusion of multidimensional tensors The multidimensional tensor 1 and the multidimensional tensor 2 are fused to generate a multidimensional tensor 3 including visual information and text information; the multidimensional tensor 3 includes comprehensive features of the visual information and text information; S205: Use the Transformer structure to process the fused tensor The multidimensional tensor 3 is processed by the encoder part of the multi-layer Transformer structure to extract features and model the relationship between multimodal information, and the multidimensional tensor 4 is obtained after processing; S206: Event classification and determination The multidimensional tensor 4 is used for event classification. Through the deep learning model of the trained convolutional neural network, the multidimensional tensor 4 is mapped to different event categories to determine the operation events occurring in the image. S207: Steps S201 to S206 are performed on all key frames, thereby determining the operation events occurring in all key frame images.

[0006] Furthermore, in step S202, the language model adopts a natural language processing model BERT or GPT based on deep learning.

[0007] Furthermore, the preprocessing includes video decoding, image cropping, filtering and denoising, and contrast enhancement.

[0008] The present invention discloses a physics and chemistry practical AI-assisted marking system based on multimodal information event detection. The system has multiple user roles, including examination center personnel, project team leaders, task leaders, marking personnel, and arbitration personnel. Different user roles have different permissions after logging into the system and realize different functions. Test center staff After logging into the system, the test center staff can set up the basic configuration module and marking project module in the system. The basic configuration module includes the marking method, preview examination paper, and arbitration condition settings. After the preview examination paper is enabled, the project leader can create a preview task; the arbitration condition setting can determine whether the arbitrator can view the previous review records of the examiner when arbitrating the examination paper; The marking project module includes a marking monitoring unit, a marking result unit, and a statistical analysis unit, and can create projects; the marking monitoring unit displays the progress distribution of each project, the arbitration ratio of the marked papers, the progress of each task, and the task progress list. In the marking monitoring unit, you can also view the records of the marked questions by the marking staff / arbitrators, the experimental questions and scoring rules of the questions, and the marking results; the marking result unit is used to view the marking results; the statistical analysis unit can view the marking task statistics, personnel statistics, and question task statistics of each project; Project Leader The project leader logs into the system and enters the marking project module, where he can see the projects he is responsible for. After selecting a project to enter, he can submit the project when it is completed. The project leader sets up the personnel management unit, pre-reading management unit, marking management unit, and marking monitoring unit under the project he is responsible for. The personnel management unit is used to select marking personnel and arbitration personnel for the project. The project leader can set properties for the personnel, including setting the personnel's type, subject, and question type, and can export the selected personnel's account and delete the personnel. The pre-reading management unit can create pre-reading tasks. If the pre-reading task in the basic configuration module is closed, the pre-reading task cannot be created. After the pre-reading task is created, a pre-reading person can be automatically assigned to each pre-reading task, or the pre-reading person can be manually adjusted and selected, and the pre-reading record of the person can be viewed. The marking management unit can create marking tasks. After creating the marking task, marking rules can be set, and automatic assignment, release, and task length can be specified. Personnel, and details can be viewed. If the project leader sets the task length to mark the papers, the task length also has marking tasks. The examination monitoring unit can view the progress distribution of the project, the arbitration ratio of the examination papers that have been examined, the progress of each task, and the progress list of each task; Task length After the task leader logs into the system, he can view the participating marking projects in the marking project module. After selecting a project, he can see my task unit, marking monitoring unit, and statistical analysis unit. My task unit includes the task leader task subunit and the marking task subunit. In the task leader task subunit, he can view the task assignment details and submit tasks. Through the task assignment details, he can view the task information, progress, and problem papers of the team members; the marking task subunit displays the list of marking tasks assigned to me, and I can mark the papers; the marking monitoring unit can view the task progress of each team member in the group, and the statistical analysis unit can count the marking task statistics of each team member in the group; Examination staff After logging into the system, examiners can see the projects they are responsible for in the examination project module. After selecting a project, they can see the examination task unit, examination unit, and examination result unit management. The examination task unit manages the examination tasks of examiners. The marking unit (including the aforementioned marking task subunit) adopts AI automatic marking + manual marking for the collected test videos. The system automatically displays the AI ​​score, and the manual scores are entered in the scoring box according to the video. It supports video playback, frame capture, video capture, speed playback, synchronous / asynchronous video processing, and can choose to recognize or not recognize the AI ​​score. When the video quality is unclear or the video is short, the test paper can be set as a question paper; the AI ​​automatic marking adopts the physical and chemical practical AI assisted marking method based on multimodal information event detection; the marking result unit is used to view the marking records; Arbitrators After the arbitrator logs into the system, he can see the marking projects he participated in. After entering the marking project, he can see the arbitration task unit, arbitration unit, and arbitration result unit. The arbitration task unit can view my arbitration tasks. The arbitration unit is used to process arbitration tasks and conduct arbitration scoring. The arbitration result unit can view the arbitration results of the processed projects.

[0009] The pre-reading task refers to the selection of a small number of test papers (i.e., test videos of a few candidates) for marking after the test is over, in order to assess the time required for marking and provide a reference for the subsequent formal marking. For marking monitoring, marking results, and statistical analysis units, the test center personnel have the greatest authority and can view the marking monitoring, marking results, and statistical analysis units at all levels below. The project leader can view the marking monitoring, marking results, and statistical analysis units under this project, and the task leader has the authority below the project leader.

[0010] Furthermore, when the test center personnel set the marking method of the basic configuration module, the marking method includes single marking, double marking, and triple marking. When setting these three marking methods, the return marking can be turned on or off.

[0011] Furthermore, when the system is used, the examination center staff first performs basic configuration, creates a marking project, and designates a project leader. Then, after the project leader logs into the system, he designates the marking staff and arbitration staff for the project in the personnel management unit, sets marking rules in the marking management unit, creates marking tasks, designates task lengths, automatically assigns marking staff to tasks, and publishes tasks. After the tasks are published, the marking staff logs into the system to mark the papers. After the marking is completed, the system puts the test questions with score errors exceeding the critical value into the arbitration task. Finally, the arbitration staff logs into the system to process the test papers that need to be arbitrated.

[0012] This system divides user roles into examination center personnel, project leaders, task leaders, examiners, and arbitration personnel, and provides different functional permissions for different roles to improve the efficiency of marking. At the same time, it adopts AI + manual scoring method. Multimodal information event detection method is used in AI scoring. This method can not only accurately detect events, but also effectively process the function and use description of events. At the same time, it makes full use of multimodal information to provide more accurate event detection results, and scoring is more accurate and efficient.

[0013] Compared with the prior art, the invention has the following beneficial effects: (1) This system makes full use of multimodal information, including images and text descriptions, in its AI scoring. Through the fusion of multidimensional tensors, the relationship between multimodal information can be better understood and modeled, thereby improving the accuracy of event detection.

[0014] (2) The system adopts a multi-role hierarchical authorization method with clear division of labor, which improves the efficiency of marking and the reliability of marking results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of system functional structure.

[0016] Figure 2 Flowchart of AI automatic scoring method.

[0017] Figure 3 Flowchart of multimodal information fusion event detection method.

[0018] Figure 4 Specific implementation process of multimodal information fusion event detection method. DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings. Figure 1 As shown, this system divides user roles into examination center personnel, project leaders, task leaders, examiners, and arbitration personnel according to the actual application scenarios of physical and chemical experiment operation examinations. Different roles log into the system to realize different functions, thereby maximizing the availability of the system.

[0020] Test center staff: Test center staff are the initiators of the marking project. They set marking rules, create projects, designate project leaders, and assign manual marking tasks to project leaders. They also monitor the marking process and view statistical analysis data of marking results.

[0021] Test center staff can access the basic configuration module and the marking project module and set the above modules; the basic configuration module includes the settings of marking method, pre-reading test paper, and arbitration conditions. The marking methods include single marking, double marking, and triple marking. When setting these three marking methods, you can turn on or off the review. After the pre-reading test paper is turned on, the project leader can create a pre-reading task; the arbitration condition setting can realize whether the arbitrator can view the previous review records of the examiner when arbitrating the test paper.

[0022] The marking project module can create projects, including the marking monitoring unit, marking result unit, and statistical analysis unit. The marking monitoring unit displays the progress distribution of each project, the arbitration ratio of the marked papers, the progress of each task, and the task progress list. In the marking monitoring unit, you can also view the records of all marked papers / arbitrators who have marked the papers, the experimental questions and scoring rules of the papers, and the marking results. The marking result unit is for viewing the marking results. The statistical analysis unit can view the statistics of each marking task, personnel statistics, and test task statistics.

[0023] Project Leader: The project leader manages the manual marking part of the marking project, selects markers, creates / manages pre-marking tasks, creates / manages marking tasks, monitors the marking process, and views statistical analysis data of marking results.

[0024] Log in to the system and enter the marking project module. In the marking project module, you can see the project you are responsible for. After selecting the project to enter, you can submit the project when it is completed. It includes personnel management unit, pre-reading management unit, marking management unit, and marking monitoring unit. The personnel management unit is used to select examiners and arbitrators for the project. You can set properties for the personnel, including the personnel type, subject, and question types they are good at. You can also export and select personnel accounts and delete personnel. The preview management unit can create preview tasks. If the preview is turned off in the basic configuration module, you cannot create preview tasks. After the preview task is created, you can automatically assign a preview person to each preview task, or you can manually adjust and select the appointment person. You can view the preview record of the person. The marking management unit can create marking tasks. After creating the marking task, enter the marking management, where you can set marking rules, automatically assign, publish, specify the task length, assign personnel, and view details.

[0025] The marking monitoring unit can view the project progress distribution, the arbitration ratio of marked papers, the progress of each task, and the progress list of each task for the project leader.

[0026] Task leader: The task leader is the person in charge of each marking task. He is mainly responsible for monitoring the marking progress and quality of the team members, processing the question papers submitted by the team members, and submitting the task to the project leader after the task is completed. If the project leader sets the task leader to mark the papers, the task leader also has the marking task.

[0027] After entering the system, you can view the participating marking projects in the marking project module. After selecting a project, you can enter the My Task Unit, the Marking Monitoring Unit, and the Statistical Analysis Unit. My Task Unit includes the Question Length Task Subunit and the Marking Task Subunit. In the Question Length Task Subunit, you can view the task assignment details and submit the task. Through the task assignment details, you can view the task information, progress, and problem papers of the team members. In the Marking Task Subunit, the list of marking tasks assigned to me is displayed, and you can mark the papers.

[0028] Markers: Markers are the main members of the marking project and are responsible for marking test questions. Each marker can have multiple marking tasks.

[0029] In the marking project module, you can see the project you are responsible for. After selecting a project, you can enter the module, which includes the marking task unit, marking unit, and marking result unit management. The marking task unit manages the marking tasks of the marking personnel. The marking unit and the marking task sub-unit adopt AI automatic marking + manual method to ensure the accuracy and fairness of the examination. The system automatically displays the AI ​​score, and the manual scores are entered in the scoring box according to the video. It supports video playback, frame capture, video capture, double-speed playback, synchronous / asynchronous video processing, and can choose to approve or reject the AI ​​score. When the video quality is unclear or the video is short, this test paper can be set as a question paper.

[0030] like Figure 2 As shown, the AI ​​automatic marking method is as follows: S10: Data collection and preprocessing Multiple cameras installed in the examination room can be used to collect videos of candidates' experimental operations from different angles to ensure the clarity and stability of the videos for subsequent analysis.

[0031] Extract key frames from the video to reduce the amount of data and highlight important experimental operation steps. Extract key frames using methods such as time interval and image change detection.

[0032] Image preprocessing is performed on key frames, including video decoding, image cropping, filtering and denoising, contrast enhancement and other operations, to improve the quality and clarity of the image in preparation for subsequent feature extraction and target detection.

[0033] S20: Multimodal feature extraction and model training The pre-processed key frame images are processed using a target detection algorithm to determine the target object, identify various experimental equipment on the experimental table, such as beakers, test tubes, alcohol lamps, etc., and record the position and status of the equipment (such as whether it is used, whether it is placed correctly, etc.) to extract different modal features in the experimental process, and then represent the different modal features through multimodal information; the target object is fused with multimodal information, and a multidimensional tensor is formed after processing; the multidimensional tensor formed after processing is used as basic data to train the deep learning model of the convolutional neural network. The trained deep learning model of the convolutional neural network (CNN) can classify the operation events in the new experimental operation video, determine whether it is a correct operation or an incorrect operation, and the specific operation type; the features of the different modalities can be fused in the tensor space to provide rich information for more accurate judgment of the process and results of the experiment. Taking the chemical experiment as an example, the action features of the student operating the instrument (such as hand movement trajectory, instrument use posture, etc.) can be extracted from the experimental operation video tensor, and the physical and chemical features of the chemical substance reaction process (such as temperature change rate, pH change, etc.) can be extracted by analyzing the sensor data tensor.

[0034] S30: Scoring rules and automatic scoring According to the specific requirements of the experiment and the syllabus, formulate detailed scoring rules; the scoring rules include the following scoring items: correct use of experimental equipment, completeness or accuracy of experimental steps, rationality of experimental data, and correctness of experimental results; assign different weights to each scoring item according to its importance; Automatic scoring is performed based on the features extracted from the experimental operation video (equivalent to repeating the multimodal feature extraction steps in S10 and S20) and the established scoring rules; for each scoring item, a score is given based on the correctness and conformity of its features, and the final score is the weighted sum of the scores of each scoring item.

[0035] like Figure 3 , 4 As shown, the event detection method of the present invention integrates multimodal information to enhance detection accuracy. The specific implementation method is as follows: S201: Object detection and functional usage description extraction A frame of RGB image (keyframe image) is processed by the object detection algorithm to detect multiple target objects in it. Each target object is accompanied by a text description related to its function and purpose. These text descriptions can be descriptions of the properties, behaviors or uses of the target object. These descriptions are spliced ​​together to form an overall text description, which will be used for subsequent processing.

[0036] S202: Processing of text description The generated text description is fed into a pre-trained language model, which can be a natural language processing model based on deep learning, such as BERT or GPT. The language model converts the text description into a multidimensional tensor 1. This multidimensional tensor 1 will capture the semantic and contextual information in the text description, providing important feature representation for subsequent event detection.

[0037] S203: Image processing using multi-layer residual CNN The original RGB image (key frame image) is also processed by a multi-layer residual convolutional neural network (CNN) and converted into a multi-dimensional tensor 2. This multi-dimensional tensor 2 will capture the visual information in the image, including the shape, position, and appearance information of the target object.

[0038] S204: Fusion of multidimensional tensors The multidimensional tensor 1 (text description features from the language model, i.e. text information) and the multidimensional tensor 2 (image features from CNN, i.e. visual information) are fused. This fusion can be done in different ways, such as concatenation, weighted averaging, etc., to generate a multidimensional tensor 3. The multidimensional tensor 3 will contain the comprehensive features of the image (visual) and text information.

[0039] S205: Using Transformer structure to process fused tensors The multidimensional tensor 3 is processed by the encoder part of the multi-layer Transformer structure to further extract features and model the relationship between multimodal information, and then a multidimensional tensor 4 is obtained. The Transformer structure can effectively process sequence data and allow interaction between multidimensional tensors.

[0040] S206: Event classification and determination The multidimensional tensor 4 is used for event classification. The multidimensional tensor 4 is mapped to different event categories through the deep learning model of the trained convolutional neural network, so as to determine the operation events occurring in the image. S207: Steps S201 to S206 are performed on all key frames, thereby determining the operation events occurring in all key frame images.

[0041] The marking result unit allows markers to view marking records.

[0042] Arbitrator: The arbitrator is responsible for arbitrating the questions whose score errors exceed the critical value during the process of the examiners' review of the questions in the project. The arbitration score becomes the final score.

[0043] Arbitrators can view the marking projects they participate in by entering the marking project module. After entering the marking project, it includes arbitration task unit, arbitration unit and arbitration result unit.

[0044] The arbitration task unit can view my arbitration tasks. The arbitration unit is used to process arbitration tasks and conduct arbitration scoring. The arbitration result unit can view the arbitration results of the processed projects.

[0045] The whole process of using this system is as follows: first, the test center staff performs basic configuration, creates a marking project, and designates a project leader. Then, after the project leader logs into the system, he designates the markers and arbitration personnel for the project in the personnel management unit, sets marking rules in the marking management unit, creates marking tasks, designates task lengths, automatically assigns markers to tasks, and publishes tasks. After the tasks are published, the markers log into the system to mark the papers. After the marking is completed, the system puts the questions with score errors exceeding the critical value into the arbitration task. Finally, the arbitration personnel log into the system to handle the papers that need to be arbitrated.

Claims

1. A physical and chemical practice AI-assisted examination marking method based on multimodal information event detection, characterized in that: The steps include: S10: Data collection and preprocessing Through multiple cameras installed in the examination room, the experimental operation videos of candidates are collected from different angles; key frames are extracted from the experimental operation videos, and image preprocessing is performed on the key frames; S20: Multimodal feature extraction and model training The pre-processed key frame images are processed using a target detection algorithm to determine the target object in order to extract different modal features during the experiment, and then the different modal features are represented by multimodal information; The target object is fused with multimodal information and processed to form a multidimensional tensor. The processed multidimensional tensor is used as basic data to train a deep learning model of a convolutional neural network. The trained deep learning model of the convolutional neural network can classify operation events in a new experimental operation video, determine whether it is a correct operation or an incorrect operation, and the specific operation type. S30: Scoring rules and automatic scoring According to the specific requirements of the experiment and the syllabus, formulate detailed scoring rules; the scoring rules include the following scoring items: correct use of experimental equipment, completeness or accuracy of experimental steps, rationality of experimental data, and correctness of experimental results; assign different weights to each scoring item according to its importance; Automatic scoring is performed based on the features extracted from the experimental operation video and the formulated scoring rules; for each scoring item, a score is given based on the correctness and conformity of its features, and the final score is the weighted sum of the scores of each scoring item.

2. The method for physics and chemistry practical AI-assisted examination paper marking based on multimodal information event detection as claimed in claim 1, characterized in that: Step S20 specifically includes the following steps: S201: Object detection and functional usage description extraction The pre-processed key frame images are processed using a target detection algorithm to detect multiple target objects therein; each target object is accompanied by a text description related to its function and purpose; these text descriptions are descriptions of the properties, behaviors or purposes of the target object; these descriptions are spliced ​​together to form an overall text description, which will be used for subsequent processing; S202: Processing of text description The generated text description is sent to a pre-trained language model, and the language model converts the text description into a multi-dimensional tensor 1, wherein the multi-dimensional tensor 1 can capture semantic information and context information in the text description and provide feature representation for subsequent event detection; S203: Image processing using multi-layer residual convolutional neural network The same key frame image in S201 is processed by a multi-layer residual convolutional neural network and converted into a multi-dimensional tensor 2; the multi-dimensional tensor 2 contains the visual information captured in the key frame image; S204: Fusion of multidimensional tensors The multidimensional tensor 1 and the multidimensional tensor 2 are fused to generate a multidimensional tensor 3 including visual information and text information; the multidimensional tensor 3 includes comprehensive features of the visual information and text information; S205: Use Transformer structure to process fused tensors The multidimensional tensor 3 is processed by the encoder part of the multi-layer Transformer structure to extract features and model the relationship between multimodal information, and the multidimensional tensor 4 is obtained after processing; S206: Event classification and determination The multidimensional tensor 4 is used for event classification. Through the deep learning model of the trained convolutional neural network, the multidimensional tensor 4 is mapped to different event categories to determine the operation events occurring in the image. S207: Steps S201 to S206 are performed on all key frames, thereby determining the operation events occurring in all key frame images.

3. The method for physics and chemistry practical AI-assisted examination paper marking based on multimodal information event detection as claimed in claim 2, characterized in that: In step S202, the language model adopts a natural language processing model BERT or GPT based on deep learning.

4. The method for physics and chemistry practical AI-assisted examination paper marking based on multimodal information event detection according to any one of claims 1 to 3, characterized in that: The preprocessing includes video decoding, image cropping, filtering and denoising, and contrast enhancement.

5. A physics and chemistry practical AI-assisted examination marking system based on multimodal information event detection, characterized in that: The system has multiple user roles, including test center personnel, project team leaders, task leaders, examiners, and arbitration personnel; different user roles have different permissions after logging into the system and realize different functions; Test center staff After logging into the system, the test center staff can set up the basic configuration module and marking project module in the system. The basic configuration module includes the marking method, preview examination paper, and arbitration condition settings. After the preview examination paper is enabled, the project leader can create a preview task; the arbitration condition setting can determine whether the arbitrator can view the previous review records of the examiner when arbitrating the examination paper; The marking project module includes a marking monitoring unit, a marking result unit, and a statistical analysis unit, and can create projects; the marking monitoring unit displays the progress distribution of each project, the arbitration ratio of the marked papers, the progress of each task, and the task progress list. In the marking monitoring unit, you can also view the records of the marked questions by the marking staff / arbitrators, the experimental questions and scoring rules of the questions, and the marking results; the marking result unit is used to view the marking results; the statistical analysis unit can view the marking task statistics, personnel statistics, and question task statistics of each project; Project Leader The project leader logs into the system and enters the marking project module, where he can see the projects he is responsible for. After selecting a project to enter, he can submit the project when it is completed. The project leader sets up the personnel management unit, pre-reading management unit, marking management unit, and marking monitoring unit under the project he is responsible for. The personnel management unit is used to select marking personnel and arbitration personnel for the project. The project leader can set properties for the personnel, including setting the personnel's type, subject, and question type, and can export the selected personnel's account and delete the personnel. The pre-reading management unit can create pre-reading tasks. If the pre-reading task in the basic configuration module is closed, the pre-reading task cannot be created. After the pre-reading task is created, a pre-reading person can be automatically assigned to each pre-reading task, or the pre-reading person can be manually adjusted and selected, and the pre-reading record of the person can be viewed. The marking management unit can create marking tasks. After creating the marking task, marking rules can be set, and automatic assignment, release, and task length can be specified. Personnel, and details can be viewed. If the project leader sets the task length to mark the papers, the task length also has marking tasks. The examination monitoring unit can view the progress distribution of the project, the arbitration ratio of the examination papers that have been examined, the progress of each task, and the progress list of each task; Task length After the task leader logs into the system, he can view the participating marking projects in the marking project module. After selecting a project, he can see my task unit, marking monitoring unit, and statistical analysis unit. My task unit includes the task leader task subunit and the marking task subunit. In the task leader task subunit, he can view the task assignment details and submit tasks. Through the task assignment details, he can view the task information, progress, and problem papers of the team members; the marking task subunit displays the list of marking tasks assigned to me, and I can mark the papers; the marking monitoring unit can view the task progress of each team member in the group, and the statistical analysis unit can count the marking task statistics of each team member in the group; Examination staff After logging into the system, examiners can see the projects they are responsible for in the examination project module. After selecting a project, they can see the examination task unit, examination unit, and examination result unit management. The examination task unit manages the examination tasks of examiners. The marking unit and the marking task subunit adopt AI automatic marking + manual mode for the collected test videos. The system automatically displays the AI ​​score, and the score is manually entered in the scoring box according to the video. It supports video playback, frame capture, video capture, double-speed playback, synchronous / asynchronous video processing, and can choose to recognize or not recognize the AI ​​score. When the video quality is unclear or the video is short, the test paper can be set as a question paper; the AI ​​automatic marking adopts the physical and chemical practical AI-assisted marking method based on multimodal information event detection as described in any one of claims 1-4; the marking result unit is used to view the marking record; Arbitrators After the arbitrator logs into the system, he can see the marking projects he participated in. After entering the marking project, he can see the arbitration task unit, arbitration unit, and arbitration result unit. The arbitration task unit can view my arbitration tasks. The arbitration unit is used to process arbitration tasks and conduct arbitration scoring. The arbitration result unit can view the arbitration results of the processed projects.

6. The physical and chemical practice AI-assisted examination marking system based on multimodal information event detection as claimed in claim 5, characterized in that: When the test center personnel set the marking method of the basic configuration module, the marking method includes single marking, double marking, and triple marking. When setting these three marking methods, the return marking can be turned on or off.

7. The physical and chemical practice AI-assisted examination marking system based on multimodal information event detection as claimed in claim 5 or 6, characterized in that: When using this system, the examination center staff will first perform basic configuration, create a marking project, and designate a project leader. Then, after the project leader logs in to the system, he will designate the marking staff and arbitration staff for the project in the personnel management unit, set marking rules in the marking management unit, create marking tasks, designate task lengths, automatically assign marking staff to tasks, and publish tasks. After the tasks are published, the marking staff will log in to the system to mark the papers. After the marking is completed, the system will put the test questions with score errors exceeding the critical value into the arbitration task. Finally, the arbitration staff will log in to the system to handle the test papers that need to be arbitrated.

Images