Automatic scoring method and system for exploring electromagnetic induction phenomenon experiment

Abstract

The invention discloses an automatic scoring system for an experiment for exploring an electromagnetic induction phenomenon, and belongs to a target detection and image classification technology in the field of computer vision and deep learning. Aiming at high requirements on real-time performance and fine-grained classification in an experiment, the method comprises the following steps: firstly, carrying out experimental equipment positioning and preliminary identification of categories through an improved YOLOv5 model, then cascading MobileNet to carry out fine-grained classification on the states of the experimental equipment, and identifying the categories and states of different experimental equipment in the experiment in real time. And finally, an automatic scoring module is set, and intelligent scoring is performed according to combinational logic of different states of the experimental equipment. According to the method, excellent balance between the detection precision and the processing speed is realized, the state of the experimental equipment can be identified efficiently and accurately, and automatic scoring in the experimental process is realized.

Description

Technical Field

[0001] The present invention relates to the field of target detection and image classification using computer vision technology and deep learning, and is applied to the field of teaching, and in particular to an automatic scoring method and system for an experiment to explore electromagnetic induction phenomena. Background Art

[0002] In middle school teaching activities, physical and chemical experimental operation is an important teaching content. Strengthening experimental teaching and strengthening students' hands-on experimental operation are of great significance to improving scientific literacy. At present, many provinces and cities have included physical and chemical experimental operation as part of the high school entrance examination and as the basis for graduation and further study. Therefore, it is necessary to improve the examination standards, avoid interference from human factors, fully ensure the fairness and justice of the examination, and make each student's experimental operation accurately judged.

[0003] The specific problem is that during the experimental examination, the more traditional method is for the teacher to supervise the students on site and score the students' operations one by one. The teacher has limited energy and there are errors in visual scoring, which leads to very low scoring efficiency. Although there is a unified scoring standard for each experimental operation process, the scoring results will also vary due to the different subjective feelings of the individual teachers, which is unfair to the students. At the same time, there is a lack of unified management of identity confirmation before the exam, and the efficiency of personnel utilization is low.

[0004] In recent years, the development of computer vision technology and artificial intelligence has promoted the widespread application of deep neural networks in the fields of target detection, image classification, semantic segmentation, etc. The key to achieving automatic computer scoring is to accurately detect the state changes of experimental equipment. In high school physics experiments, different states of the same experimental equipment often have many similar features. At the same time, when faced with problems such as equipment occlusion and background interference, there are cases of missed detection and false detection, which lead to scoring errors. In order to solve the above problems, the present invention proposes a method of combining an improved YOLOv5 network with MobileNet, which improves the real-time and accuracy of detection, better adapts to the needs of complex experimental operation scenarios, and thus achieves efficient automatic scoring. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides an automatic scoring system for exploring electromagnetic induction phenomenon experiments, including an intelligent desk device, a user login terminal, a database, an intelligent analysis server and a Web host.

[0006] The smart desk device includes a laboratory table, and the laboratory table has a shooting device for shooting a video of the student's operation;

[0007] The user login terminal is used to input the candidate's identity information and confirm the candidate's identity;

[0008] The intelligent analysis server includes recognition processing of experimental videos;

[0009] The intelligent analysis server includes an intelligent scoring module, which calculates scores for experimental operations;

[0010] The database is connected to the intelligent scoring module;

[0011] The database processes the experimental video and transmits the analysis results to the web host for teachers and students to view.

[0012] Preferably, the camera has dual camera positions, and one of the camera positions records the position of the examinee's limbs and face, while the other camera position records the position and status of the experimental equipment.

[0013] An automatic scoring method for an experiment to explore electromagnetic induction phenomena comprises the following steps:

[0014] Step 1: Use the pre-collected experimental process video to collect and create data sets and train the target detection model;

[0015] Step 2: Use the trained detection model to create a classification data set and train the image classification model;

[0016] Step 3: According to the status of each experimental device and the scoring points in the scoring criteria, an automatic scoring model is constructed;

[0017] Step 4: Cascade the target detection model and the image classification model to finally identify the category, location, and status of each experimental device;

[0018] Step 5: Input the identified experimental results into the automatic scoring model to give the experimental score.

[0019] Preferably, the data set preparation in step 1 comprises the following steps:

[0020] A. The camera resolution is 1080p, and the detection targets are large and medium-sized, with no small targets. Based on the actual position of the camera on the experimental platform, a series of experimental videos are shot to produce the data set. To ensure the integrity and diversity of the data set, each experimental device is collected in different positions, orientations, and backgrounds.

[0021] B. After video acquisition, convert the video into images, delete unclear and non-standard sample images, and screen out qualified sample images; use labellmg to annotate different experimental equipment in the sample images, and the image annotation specification uses the VOC2007 data format.

[0022] Preferably, the target detection model is trained in step 1 as follows:

[0023] Using the improved YOLOV5 model, the model parameters and structure are optimized using the results of the training data set and the validation data set to further improve the recognition accuracy. Through training, a target detection model suitable for experimental equipment recognition is obtained, which is used to preliminarily detect the category of each device. The improvements are as follows:

[0024] A. Use GhostNet lightweight convolution module instead of traditional convolution;

[0025] B. Introduce SimAM parameterless attention, the formula is:

[0026]

[0027] in, and are the mean and variance of other neurons xi on the channel, t is the target neuron, and λ is a hyperparameter for balancing different features;

[0028] C. Replace the loss function with EIOU, the formula is:

[0029]

[0030] IoU is the overlap rate between the predicted detection box and the real detection box; (b, bgt) represents the center point of the predicted box and the real box; ρ is the Euclidean distance between the center points of the two detection boxes, c represents the diagonal length of the minimum bounding rectangle that contains both the predicted box and the real box, w and h are the width and height of the predicted box, wgt and hgt are the width and height of the real box, and cw and ch are the width and height of the minimum bounding rectangle of the predicted box and the real box.

[0031] Preferably, the data set preparation in step 2 comprises the following steps:

[0032] A. Use the object detection model to capture a series of sub-images of the experimental equipment in different states, and select high-quality images from the captured images to create a classification dataset.

[0033] Preferably, the image classification model is trained in step 2 as follows:

[0034] State recognition requires classification of different states. The MobileNet classification model is trained through classification data sets to identify different states of the device.

[0035] Preferably, the status of the scoring points in step 3 is determined automatically by the system, including the step of tidying up the instruments after the experiment.

[0036] Preferably, in step 3, the preset scoring standard is formulated in accordance with the experimental operation regulations issued by the Education Department;

[0037] That is, it is detected that the circuit diagram is connected correctly and a score is obtained. During the experiment, the following states are detected: (1) The coil and the magnet are relatively stationary, and the ammeter pointer does not deflect, so a score is obtained; (2) The coil moves left and right while cutting the magnetic flux lines, and the ammeter pointer deflects, so a score is obtained; (3) The coil moves up and down while cutting the magnetic flux lines parallel to the coil, and the ammeter pointer does not deflect, so a score is obtained; (4) The coil moves obliquely while cutting the magnetic flux lines, and the ammeter pointer deflects, so a score is obtained. The conclusion is drawn: when part of the conductor in a closed loop moves while cutting the magnetic flux lines in a magnetic field, an induced current will be generated in the conductor.

[0038] Preferably, in step 4, the bounding box information of the experimental equipment output by the target detection model is used as the input of the image classification model, and the final output of the image classification model includes the category and status information of the experimental equipment.

[0039] The present invention provides an automatic scoring method and system for an experiment to explore the electromagnetic induction phenomenon. After a student logs into a candidate system, a camera is used to collect the student's operation video on an experimental operation platform. The video is transmitted to a deep learning server through a local area network. The type and state of the instrument during the student's operation are detected through a trained deep learning model. According to the state change, the score items of the experimental operation are evaluated with reference to the operation specifications prescribed by the Ministry of Education.

[0040] The present invention provides an automatic scoring method and system for exploring electromagnetic induction phenomenon experiments, which can be applied to the teaching and examination scenarios of middle school physics, chemistry and biology experiments, and solve the problems of teacher visual scoring errors and different scoring standards in teaching scenarios. The present invention can also realize the full recording of experimental operations for later retrospective review. It can also be used to assist daily experimental teaching, and improve the efficiency and level of experimental guidance, which has a positive effect on promoting teaching effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0042] Figure 1 A schematic diagram of a flow chart of an automatic scoring method for an experiment to explore electromagnetic induction phenomenon according to the present invention;

[0043] Figure 2 This is a schematic diagram of an automatic scoring method for an experiment to explore electromagnetic induction phenomenon and a system experimental effect;

[0044] Figure 3 This is a schematic diagram of an automatic scoring method and system scoring process for an experiment to explore electromagnetic induction phenomenon according to the present invention;

[0045] Figure 4The present invention is a schematic diagram of an automatic scoring method and system improvement model for an experiment to explore electromagnetic induction phenomenon.

[0046] Figure 5 The present invention is a schematic diagram of the structure of an automatic scoring system for an experiment to explore electromagnetic induction phenomenon. DETAILED DESCRIPTION

[0047] The technical scheme of the present invention will be clearly and completely described below in conjunction with specific embodiments, but those skilled in the art should understand that the embodiments described below are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The experimental instrument used in the present invention comprises a switch, a conductor, a coil, a horseshoe magnet and a sensitive ammeter.

[0049] The present invention discloses an automatic scoring method for an experiment to explore electromagnetic induction phenomenon, comprising the following steps:

[0050] Step 1: Adjust the angle of the camera and the front-view camera to face the center of the experimental operation area, collect the student operation video through the camera, and transmit the video to the deep learning server through the local area network.

[0051] Step 2: Use the pre-collected video dataset and train the model to build an object detection model.

[0052] Step 3: Use the trained detection model to preliminarily detect the location and category of each device in the operation video.

[0053] Step 4: Based on the identified location and category, train a classification model to identify different states of the device.

[0054] Step 5: Cascade the target detection network and the classification network to finally identify the category and status of each experimental device.

[0055] Step 6: The examination operation specifications stipulated by the Ministry of Education are combined with the automatic scoring module to evaluate students' examination operation.

[0056] Example

[0057] See also Figure 1 , Figure 3 , Figure 5 , an automatic scoring method and system for exploring electromagnetic induction phenomenon experiments, specifically including the following steps:

[0058] Step 1: Shoot the experimental video and collect the experimental data. Record a video containing the galvanometer, magnetic coil, magnet and switch, then process the recorded video into frames, and then filter the captured images, such as deleting blurry or unclear images.

[0059] Step 2: Data labeling: Label the selected images using the labelimg software.

[0060] Step 3: Train the detection model. The required data set is divided into training set, validation set, and test set according to the ratio of 8:2:2. The detection model can identify and detect the category and location of each device in the experiment in real time. It is necessary to classify the status of each device based on which device is detected. Therefore, the different status of each experimental device is classified in a targeted manner.

[0061] Step 4: Get the classification data set and train the classification model. After the detection model is trained, use the target detection model to capture the sub-image of the experimental device from the original image. Get the sub-image in the recorded video by deleting the blurred and repeated images and resizing them to (224, 224). Select 2000-3000 high-quality images for each state of each type of experimental equipment to create a data set. Organize the data set used to classify each device state.

[0062] Step 5: Cascade the network model to obtain the location category and status of the experimental equipment.

[0063] See also Figure 2 , Figure 2 This is an automatic scoring method for exploring electromagnetic induction phenomena and a schematic diagram of the system's effects. The original video stream is input into the network model for multi-target detection, and the category and location of each physical experimental equipment are detected in real time. Each experimental equipment is classified. For example, the state of the magnet during the experiment can be divided into cut (cut state) and uncut (uncut state), the state of the magnetic coil during the experiment can also be divided into cut (cut state) and uncut (uncut state), and the state of the ammeter during the experiment can be divided into deflect (deflection state), zero_nc (0 scale unwired state), and zero_wire (0 scale wired state). To evaluate the performance of the selected classification model, the total classification accuracy ACC is used as an indicator.

[0064] See also Figure 4As a preferred implementation, based on YOLOv5, an improved YOLOv5 model uses lightweight GhostNet convolution instead of ordinary convolution, and introduces a parameter-free attention module in the neck, with Precision, Recall, mAP@0.5, and mAP@0.5:0.95 as the main evaluation indicators. Compared with the original model, the number of parameters and the amount of calculation are reduced, the model is more lightweight, and the detection accuracy is also improved accordingly.

[0065] The present invention discloses an automatic scoring system for experiments to explore electromagnetic induction phenomena, which belongs to the target detection and image classification technology in the field of computer vision and deep learning. In view of the high requirements for real-time and fine-grained classification in the experiment, the present invention first uses an improved YOLOv5 model to perform preliminary identification of experimental equipment positioning and categories, and then cascades MobileNet to perform fine-grained classification of the status of the experimental equipment, and recognizes the categories and states of different experimental equipment in the experiment in real time. Finally, by setting an automatic scoring module, intelligent scoring is performed according to the combination logic of different states of the experimental equipment. This method achieves an excellent balance between detection accuracy and processing speed, can efficiently and accurately complete the identification of the status of the experimental equipment, and realizes automatic scoring of the experimental process.

[0066] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An automatic scoring system for exploring electromagnetic induction phenomena experiments, characterized in that: Including smart desk equipment, user login terminal, database, smart analysis server and Web host, The smart desk device includes a laboratory table, and the laboratory table has a shooting device for shooting a video of the student's operation; The user login terminal is used to input the candidate's identity information and confirm the candidate's identity; The intelligent analysis server includes recognition processing of experimental videos; The intelligent analysis server includes an intelligent scoring module, which calculates scores for experimental operations; The database is connected to the intelligent scoring module; The database processes the experimental video and transmits the analysis results to the web host for teachers and students to view.

2. The automatic scoring system for exploring electromagnetic induction phenomena according to claim 1 is characterized in that: The camera has two camera positions, and one of the camera positions records the position of the examinee's limbs and face, while the other camera position records the position and status of the experimental equipment.

3. An automatic scoring method for an experiment to explore electromagnetic induction phenomena, comprising the following steps: Step 1: Use the pre-collected experimental process video to collect and create data sets and train the target detection model; Step 2: Use the trained detection model to create a classification data set and train the image classification model; Step 3: According to the status of each experimental device and the scoring points in the scoring criteria, an automatic scoring model is constructed; Step 4: Cascade the target detection model and the image classification model to finally identify the category, location, and status of each experimental device; Step 5: Input the identified experimental results into the automatic scoring model to give the experimental score.

4. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: The data set preparation in step 1 includes the following steps: A. The camera resolution is 1080p, the detection targets are large and medium, and there are no small targets; according to the actual position of the camera on the experimental platform, a series of experimental videos are shot to create a data set; B. After video acquisition, convert the video into images, delete unclear and non-standard sample images, and screen out qualified sample images; use labellmg to annotate different experimental equipment in the sample images, and the image annotation specification uses the VOC2007 data format.

5. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: The target detection model is trained in step 1 as follows: Using the improved YOLOV5 model, the model parameters and structure are optimized using the results of the training and validation data sets to further improve the recognition accuracy. Through training, a target detection model suitable for experimental equipment recognition is obtained, which is used to preliminarily detect the categories of each device. A. Use GhostNet lightweight convolution module instead of traditional convolution; B. Introduce SimAM parameter-free attention, the formula is: in, and are the mean and variance of other neurons xi on the channel, t is the target neuron, and λ is a hyperparameter for balancing different features; C. Replace the loss function with EIOU, the formula is: IoU is the overlap rate between the predicted detection box and the real detection box; (b, bgt) represents the center point of the predicted box and the real box; ρ is the Euclidean distance between the center points of the two detection boxes, c represents the diagonal length of the minimum bounding rectangle that contains both the predicted box and the real box, w and h are the width and height of the predicted box, wgt and hgt are the width and height of the real box, and cw and ch are the width and height of the minimum bounding rectangle of the predicted box and the real box.

6. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: The data set preparation in step 2 includes the following steps: Use the target detection model to capture a series of sub-images of the experimental equipment in different states, and select high-quality images from the captured images to create a classification dataset.

7. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: The image classification model is trained in step 2 as follows: State recognition requires classification of different states. The MobileNet classification model is trained through classification data sets to identify different states of the device.

8. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: The status of the scoring points in step 3 is automatically determined by the system, including the step of tidying up the instruments after the experiment.

9. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: In step 3, the preset scoring criteria are formulated in accordance with the experimental operation regulations issued by the Education Department; That is, the circuit diagram connection is detected to be correct, and the following states are detected during the experiment: (1) the coil and the magnet are relatively stationary, and the ammeter pointer does not deflect, and the points are scored; (2) the coil cuts the magnetic flux lines left and right, and the ammeter pointer deflects, and the points are scored; (3) The coil moves up and down parallel to the cutting magnetic flux lines, and the ammeter pointer does not deflect, so a score is scored; (4) The coil moves diagonally to the cutting magnetic flux lines, and the ammeter pointer deflects, so a score is scored; The conclusion is: when part of the conductor in a closed loop moves in a magnetic field to cut magnetic lines of force, an induced current will be generated in the conductor.

10. The automatic scoring method for exploring the electromagnetic induction phenomenon experiment according to claim 3 is characterized in that: In step 4, the bounding box information of the experimental device output by the target detection model is used as the input of the image classification model, and the output of the final image classification model includes the category and status information of the experimental device.

Images