College physics experiment scoring system and method based on YOLOv8

Abstract

This invention discloses a YOLOv8-based scoring system and method for university physics experiments, relating to the field of computer vision technology. The system employs an image preprocessing module to standardize input experimental images or video frames; a target detection and segmentation module based on the YOLOv8-seg model to achieve multi-target detection and instance segmentation of experimental equipment, capable of recognizing various experimental equipment and adapting to complex experimental scenarios; a state judgment module based on an independent classification model to detect the working status of key experimental equipment; a data calculation module responsible for calculating core experimental data; an intelligent scoring module with a built-in multi-stage weighted scoring mechanism to calculate scores at each stage and the final total score based on preset scoring rules; and a web interaction module to enable uploading experimental images and displaying scoring results. This achieves automation and standardization in physics experiment evaluation, adapting to different types of physics experiment evaluation needs.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, specifically to a scoring system and method for university physics experiments based on YOLOv8. Background Technology

[0002] With the accelerated advancement of educational informatization, cutting-edge technologies such as computer vision and deep learning are gradually permeating all aspects of education, bringing profound changes to traditional teaching models. In the crucial aspect of physics experiment teaching, the evaluation of experimental operations is not only an important means of examining students' learning outcomes, but also a key link in improving teaching quality and promoting teaching feedback. However, traditional manual evaluation methods are limited by the subjectivity of evaluators, differences in experience, and low evaluation efficiency, making it difficult to meet the needs of large-scale, standardized evaluation. Therefore, developing a system and method that can automatically and standardizedly evaluate physics experiment operations has become an important issue that urgently needs to be addressed in the current field of educational technology.

[0003] Existing automated assessment systems for physics experiments are mostly based on traditional image processing techniques such as edge detection and template matching. While these techniques can identify experimental equipment under specific conditions, their application scenarios are highly limited, making it difficult to cope with the complex environments of actual physics experiments. Specifically, traditional techniques have strict requirements on the uniformity of the experimental background, the stability of lighting conditions, and the standardization of equipment placement. However, in real physics experiment scenarios, there is a variety of experimental equipment, different placement postures, and variable lighting conditions. Furthermore, the experimental background is complex and often accompanied by equipment occlusion. These factors all prevent traditional techniques from effectively identifying experimental equipment, thus affecting the accuracy of the assessment. In addition, traditional assessment systems often focus on single-dimensional assessments, such as the presence of equipment or whether the operation steps have been completed, while neglecting multi-dimensional assessment needs such as the correctness of experimental operations and the accuracy of experimental data. This makes it difficult to meet the comprehensive and standardized requirements of physics experiment assessment. Therefore, traditional techniques cannot adapt to the automated assessment needs in complex physics experiment scenarios, becoming a key factor restricting the improvement of the quality of physics experiment teaching. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a university physics experiment scoring system and method based on YOLOv8. It includes an image preprocessing module, a target detection and segmentation module, a state judgment module, a data calculation module, an intelligent scoring module, and a Web interaction module. The image preprocessing module standardizes the input experimental images or video frames; the target detection and segmentation module uses the YOLOv8-seg model to achieve multi-target detection and instance segmentation of experimental equipment, enabling the identification of various experimental equipment and adapting to complex experimental scenarios; the state judgment module uses an independent classification model to detect the working status of key experimental equipment; the data calculation module is responsible for calculating core experimental data; the intelligent scoring module has a built-in multi-stage weighted scoring mechanism, calculating the scores for each stage and the final total score according to preset scoring rules; and the Web interaction module enables the uploading of experimental images and the display of scoring results, achieving automation and standardization of physics experiment evaluation and adapting to different types of physics experiment evaluation needs.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, a university physics experiment scoring system based on YOLOv8, the system comprising: Image preprocessing module: Receives experimental images or video frames, performs affine transformation on them for standardization, unifies the image size to a preset pixel specification, and outputs the standardized image; The target detection and segmentation module receives the standardized image, performs multi-target detection and instance segmentation of the experimental equipment through YOLOv8-seg model inference, and outputs the category, bounding box coordinates, and pixel-level segmentation mask of each piece of equipment. Status determination module: Receives the category and bounding box coordinates of the experimental equipment, performs status detection on the preset key experimental equipment, and outputs the working status results of the key experimental equipment; Data Calculation Module: Receives the type of experimental equipment, bounding box coordinates, and pixel-level segmentation mask; extracts data of the experimental object and reference equipment; calculates the conversion coefficient between pixels and physical units; completes the calculation of core experimental data; and outputs the core experimental data. Intelligent scoring module: Receives the category of experimental equipment, the working status results of key experimental equipment, and core experimental data; calculates the scores for each stage and the final total score according to preset scoring rules; and outputs the scores for each stage and the final total score. Web Interaction Module: Enables the uploading and display of experimental images or video frames and scoring results. It receives experimental images or video frames uploaded by users and transmits them to the image preprocessing module. It also receives the category of experimental equipment, the scores of each stage and the final total score of the experiment, integrates them and outputs the results for display.

[0006] Furthermore, the image preprocessing module receives externally input experimental image or video frame data. First, it performs format normalization on the received raw data. For the experimental video frame data, it performs single-frame extraction and effective frame filtering. For experimental images of different formats, it performs unified conversion of color space and pixel format to form standardized basic image data to be processed. Then, it determines the size parameters of the target image based on preset pixel specifications, constructs an affine transformation matrix based on these parameters, and performs affine transformation processing on the basic image data. Coordinate mapping is used to achieve the correspondence between the original image pixels and the target image pixels. Simultaneously, an interpolation algorithm is used to fill pixel gaps generated during the transformation process, eliminating pixel distortion caused by scaling and angle correction. After completing the affine transformation, the image undergoes final pixel dimension calibration to ensure that the size parameters of the processed image completely match the preset pixel specifications, ultimately outputting a standardized image with uniform specifications and consistent pixel features.

[0007] Furthermore, after receiving the standardized image transmitted by the image preprocessing module, the target detection and segmentation module converts the image data into tensor data adapted to the YOLOv8-seg model input format, loads the pre-trained model weights, and configures the basic inference parameters. Then, the model inference process is initiated. The model's feature extraction network performs multi-scale feature extraction on the standardized image, generating image feature maps at different levels. The detection head performs multi-target detection of the experimental equipment based on the feature maps, and outputs the category information and bounding box coordinates of the experimental equipment through anchor box matching and classification regression calculation. The segmentation head performs upsampling and feature fusion processing on the feature maps to generate mask feature maps that match the size of the input image. After pixel-level classification, pixel-level segmentation masks for each experimental equipment are obtained. After inference, non-maximum suppression processing is performed on the output results of the detection head. Duplicate or low-confidence detection boxes are removed according to the intersection-union ratio (IUGR) calculation formula. The mask feature maps output by the segmentation head are then subjected to coordinate mapping and binarization processing. Finally, the categories of the experimental equipment, accurate bounding box coordinates, and pixel-level segmentation masks corresponding to each equipment are integrated and output.

[0008] Furthermore, the target detection and segmentation module outputs the category information and bounding box coordinates of the experimental equipment through anchor box matching and classification regression calculation. The formula for calculating the bounding box coordinates is: , ,in, The coordinates of the center of the experimental equipment bounding box are: This refers to the offset of the bounding box center coordinates predicted by the model. For sigmoid activation function, The center coordinates of the preset anchor frame, The width and height of the experimental equipment bounding box, The bounding box width and height offsets predicted by the model. Set the width and height of the preset anchor frame.

[0009] Furthermore, the target detection and segmentation module eliminates duplicate or low-confidence detection boxes based on the intersection-union ratio (IU) calculation formula, which is: ,in, For the first The and the first The intersection-union ratio of the bounding boxes of the experimental equipment. , The first The, the The area of ​​the pixel region corresponding to the bounding box of each experimental device.

[0010] Furthermore, after receiving the experimental equipment category and bounding box coordinates transmitted by the target detection and segmentation module, the state judgment module filters and matches the received data, locates the category information and bounding box coordinates corresponding to the preset experimental equipment, and verifies the effective range of the bounding box coordinates. Then, based on the verified bounding box coordinates, a targeted local cropping operation is performed on the original experimental image to extract the image region containing only the key experimental equipment. Then, the local image is subjected to preprocessing operations such as size normalization adjustment and pixel value standardization, and the processed image data is converted into an input format suitable for the classification model. After completing the image preprocessing, the pre-trained classification model weights are loaded and the corresponding inference parameters are configured. The preprocessed local image of the experimental equipment is input into the classification model for inference. The preliminary state result is obtained through the feature extraction and classification judgment process of the model. Then, the result is post-processed and verified to remove abnormal judgment data, and finally the working state result of the preset key experimental equipment is determined.

[0011] Furthermore, after receiving the experimental equipment category, bounding box coordinates, and pixel-level segmentation mask transmitted by the target detection and segmentation module, the data calculation module first performs correlation analysis and matching positioning on various types of data. Based on the preset equipment type classification, it extracts the feature data corresponding to the experimental object equipment and the bounding box coordinates and pixel-level segmentation mask data corresponding to the reference equipment, accurately obtaining the pixel size information from the pixel data of the reference equipment. Subsequently, the extracted experimental object data is sorted and regulated according to the preset coordinate dimensions to eliminate the calculation deviation caused by data disorder. Combining the known actual physical size of the reference equipment with the extracted pixel size, the conversion coefficient between pixels and physical units is calculated through proportional conversion. Based on this conversion coefficient, the pixel-level distance, position, and other data of the experimental object are converted into corresponding data in actual physical units. Then, combined with the preset experimental basic parameters, the derivation and calculation of the experimental core data are completed according to the professional calculation rules of physical experiments. Finally, the rationality of the calculated experimental core data is verified, and abnormal data caused by pixel extraction and data conversion are eliminated. Valid experimental core data is determined and output in a targeted manner, providing accurate data basis for the core data measurement and evaluation of the intelligent scoring module.

[0012] Furthermore, the intelligent scoring module calculates the scores for each stage and the final total score based on preset scoring rules, and outputs the scores for each stage and the final total score in the following specific steps: The first stage of equipment selection evaluation: Based on the pre-set necessary experimental equipment, the score for this stage is calculated according to the proportion of the number of necessary equipment detected to the total number of necessary equipment. The second stage of installation correctness assessment: The score for this stage is calculated based on the working status results of the key experimental equipment output by the status judgment module. The key experimental equipment receives full marks if it is in the preset normal working state; otherwise, it receives zero marks. The third stage is the core data measurement and evaluation: the score for this stage is calculated based on the error range between the core experimental data output by the data calculation module and the preset standard value, and the corresponding score is configured according to the preset error level interval; The final total score of the experiment is calculated by weighted summation according to the pre-set weighting coefficients of each evaluation stage.

[0013] Furthermore, the intelligent scoring module calculates the final total score of the experiment by weighted summation according to the preset weight coefficients for each evaluation stage. The calculation formula is as follows: ,in, It is the final total score for the physics experiment assessment; This is the actual score during the equipment selection and evaluation phase. These are the preset weighting coefficients in the equipment selection and evaluation phase. This is the actual score during the installation correctness assessment phase. These are the preset weighting coefficients for the installation correctness assessment phase. It is the actual score in the core data measurement and evaluation phase. These are the preset weighting coefficients for the core data measurement and evaluation phase.

[0014] On the other hand, a YOLOv8-based scoring method for university physics experiments, applicable to the YOLOv8-based scoring system for university physics experiments as described in any one of claims 1-9, is characterized in that the method comprises: S1: Load the pre-trained YOLOv8-seg model and classification model, configure the preset parameters for model inference, and set the preset weights and preset scoring rules for multi-stage weighted scoring in the intelligent scoring module; S2: Upload experimental images or video frames via the Web interaction module and transmit them to the image preprocessing module; S3: The image preprocessing module performs affine transformation on the experimental images or video frames, unifies the size to a preset pixel specification, outputs the standardized image, and transmits it to the target detection and segmentation module; S4: The object detection and segmentation module performs model inference on the standardized image and outputs the category of the experimental equipment, bounding box coordinates, and pixel-level segmentation mask, which are then transmitted to the state judgment module, data calculation module, and intelligent scoring module, respectively. S5: The intelligent scoring module first calculates the equipment selection evaluation score based on the category of experimental equipment; S6: The intelligent scoring module calculates the final total score of the experiment based on the evaluation scores at each stage and the corresponding preset weights; S7: The intelligent scoring module transmits the scores of each stage and the final total score of the experiment to the Web interaction module. The target detection and segmentation module transmits the category of the experimental equipment to the Web interaction module. The Web interaction module integrates these data and outputs the results for display.

[0015] Compared with existing technologies, this YOLOv8-based scoring system and method for university physics experiments has the following advantages: This invention utilizes an image preprocessing module to perform image standardization through affine transformation, eliminating interference from differences in shooting angle and size on model inference and significantly improving the stability and accuracy of subsequent perception stages. The YOLOv8-seg-based target detection and segmentation module can achieve accurate detection and pixel-level instance segmentation of various physical experimental equipment under complex experimental scenarios, lighting variations, and equipment occlusion conditions. The state judgment module and data calculation module respectively realize intelligent determination of the working status of key equipment and automated quantitative calculation of core experimental data, filling the gap in existing technologies that cannot accurately assess the standardization of equipment installation and the accuracy of experimental data. The intelligent scoring module employs multi-stage weighted scoring. The scoring mechanism quantifies the evaluation based on three dimensions: equipment selection, correct installation, and core data measurement. This standardizes, objectsifies, and makes reproducible the assessment of physics experiments, resolving the issue of inconsistent evaluation standards. The web interaction module enables the uploading of experimental images and the visualization of scoring results, constructing an end-to-end automated scoring process. This significantly improves the efficiency of physics experiment assessment and adapts to new teaching scenarios such as online teaching and large-scale assessments. Furthermore, the system can be quickly adapted to different types of physics experiments through parameter adjustments, demonstrating strong versatility and scalability. This effectively reduces teachers' workload, optimizes the quality of experimental teaching, and truly meets the digital and intelligent development needs of physics experiment teaching in universities.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art based on the following examination or study, or may be taught from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a block diagram of a university physics experiment scoring system based on YOLOv8.

[0019] Figure 2 This is a flowchart of a scoring method for university physics experiments based on YOLOv8.

[0020] Figure 3 This is a flowchart showing the scores at each stage and the final total score of the college physics experiment scoring method based on YOLOv8. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] This invention provides a scoring system and method for university physics experiments based on YOLOv8, such as... Figure 1 As shown, the system includes an image preprocessing module, an object detection and segmentation module, a state judgment module, a data calculation module, an intelligent scoring module, and a web interaction module. The image preprocessing module standardizes the input experimental images or video frames; the object detection and segmentation module uses the YOLOv8-seg model to perform multi-target detection and instance segmentation of experimental equipment, enabling the identification of various experimental equipment and adapting to complex experimental scenarios; the state judgment module uses an independent classification model to detect the working status of key experimental equipment; the data calculation module is responsible for calculating the core experimental data; the intelligent scoring module has a built-in multi-stage weighted scoring mechanism, calculating the scores of each stage and the final total score of the experiment according to preset scoring rules; and the web interaction module enables the uploading of experimental images and the display of scoring results, realizing the automation and standardization of physical experiment evaluation and adapting to different types of physical experiment evaluation needs.

[0023] The image preprocessing module receives raw experimental images or video frames transmitted from the Web interaction module. It supports parsing mainstream image and video formats. For video frame data, it uses a frame extraction algorithm to extract single frames at a fixed frame rate and filters valid frames by calculating pixel variance (removing invalid frames such as blurry, black screen, or frames without experimental equipment, and retaining valid frames with a pixel variance greater than a preset threshold). For image data of different formats, it performs a unified conversion between color space and pixel format, converting grayscale images and RGBA images to RGB three-channel color images, and converting 16-bit depth pixel images to 8-bit depth pixel images, forming a standardized basic image data to be processed. The target pixel specifications for model inference are pre-set, and based on the size of the original basic image and the target pixel specifications, a model is constructed that includes translation, rotation, and scaling. The affine transformation matrix is ​​obtained by solving the correspondence between feature points of the original image and the target image, realizing the coordinate mapping between pixels of the original image and pixels of the target image. During the affine transformation, a bilinear interpolation algorithm is used to fill the pixel gaps generated by the transformation. This algorithm obtains the value of the missing pixel by calculating the weighted average of four adjacent pixels, effectively eliminating pixel blurring and distortion caused by image scaling and angle correction. The pixel size of the image after affine transformation is checked. If there is a pixel size deviation (such as a size deviation caused by rounding), calibration is completed by cropping / padding edge pixels to ensure that the size parameters of the processed image are completely matched with the preset pixel specifications. Finally, a standardized RGB image with uniform specifications and consistent pixel features is output and transmitted to the target detection and segmentation module.

[0024] The object detection and segmentation module, based on the YOLOv8-seg instance segmentation model, performs multi-object detection and pixel-level segmentation of experimental equipment. It outputs the equipment's category, precise bounding box coordinates, and pixel-level segmentation mask. The module receives a standardized 640×640 RGB image from the image preprocessing module and converts it into tensor data adapted to the YOLOv8-seg model. This conversion includes normalizing image pixel values ​​to the 0-1 range and converting HWC pixel format to CHW format, ultimately generating a 3×640×640 tensor. It then loads the weights of a YOLOv8-seg model pre-trained on a dataset of labeled university physics experimental equipment and configures the model's basic inference parameters: inference confidence threshold set to 0.5 and non-maximum suppression threshold set to 0.5. In this embodiment, the training dataset for the YOLOv8-seg model is a self-made dataset of physics experimental equipment from universities. It includes labeled images of various experimental equipment such as mechanics (weights, spring balances, inclined planes), optics (convex lenses, concave lenses, optical benches), and electromagnetism (power supplies, resistors, ammeters). The labeling information includes equipment category, bounding box coordinates, and pixel-level segmentation mask. The tensor data is input into the YOLOv8-seg model, and multi-scale feature extraction is completed through the backbone network of the model: the backbone network consists of a C2f module and SPPF. The input image is downsampled by 8x, 16x, and 32x in sequence to generate three different levels of feature maps, which correspond to the detection of small-sized, medium-sized, and large-sized experimental equipment, respectively, so as to achieve accurate feature extraction of experimental equipment of different specifications.

[0025] The model detection head performs multi-target detection of experimental equipment based on multi-scale feature maps. It outputs the category information and bounding box coordinates of the experimental equipment through anchor box matching and classification regression calculation. The bounding box coordinates are calculated using a preset formula. , In the formula, The coordinates of the center of the experimental equipment bounding box are: This refers to the offset of the bounding box center coordinates predicted by the model. For sigmoid activation function, The center coordinates of the preset anchor frame, The width and height of the experimental equipment bounding box, The bounding box width and height offsets predicted by the model. The width and height of the preset anchor boxes are determined. The model segmentation head performs transposed convolution upsampling and feature fusion processing on the multi-scale feature maps, restoring the downsampled feature maps to the same 640×640 size as the input image, generating mask feature maps. Pixel-level binary classification is performed on the mask feature maps, with a binarization threshold of 0.5. Pixel values ​​greater than 0.5 are classified as equipment areas, otherwise as background areas, ultimately obtaining pixel-level segmentation masks for each experimental device. Non-maximum suppression processing is applied to the output results of the detection head, and duplicate or low-confidence detection boxes are removed according to the intersection-over-union (IoU) calculation formula. The IoU calculation formula is: ,in, For the first The and the first The intersection-union ratio of the bounding boxes of the experimental equipment. , The first The, the The area of ​​the pixel region corresponding to the bounding box of each experimental device, if the two detection boxes If the value is greater than the preset threshold of 0.5, the detection box with low confidence is removed; the mask feature map output by the segmentation head is optimized by coordinate mapping and binarization to eliminate pixel noise at the mask edge, and finally the category of experimental equipment (including confidence), accurate bounding box coordinates and pixel-level segmentation mask corresponding to each equipment are integrated and output, and transmitted to the state judgment module, data calculation module and intelligent scoring module respectively.

[0026] The state judgment module receives the experimental equipment categories and bounding box coordinates transmitted by the target detection and segmentation module. Based on a pre-set list of key experimental equipment (such as power on / off status, ammeter wiring method, and convex lens placement angle), it filters and matches the received data, locating the category information and bounding box coordinates corresponding to the key experimental equipment, and validating the effective range of the bounding box coordinates. Based on the verified bounding box coordinates, it performs targeted local cropping on the original experimental image. To avoid cropping the edge features of the equipment, it crops outward by 5 pixels from the bounding box, extracting only the image region containing the key experimental equipment. This local image is then preprocessed with size normalization and pixel value standardization, converting the processed image data into 3×224×224 tensor data suitable for the classification model, and loading the pre-trained lightweight classification model. For weighting, this implementation method preferably uses the MobileNet lightweight classification model, which balances inference speed and classification accuracy. Alternatively, ResNet18 or other classification models can be used. This model is trained for different operating states of key experimental equipment (e.g., power on / off, ammeter positive / reverse connection). Inference parameters are configured, and preprocessed local image tensor data of the key equipment are input into the classification model. Feature extraction is performed through convolutional and pooling layers, and classification is determined through fully connected layers to obtain preliminary state results. These preliminary results are then post-processed for verification. A majority voting method is used to remove abnormal data. For example, in multi-frame detection results of the same equipment, if the confidence percentage of a certain state is less than 50%, it is considered abnormal data and removed. Finally, the operating state result (normal / abnormal) of the preset key experimental equipment is determined, and the result is transmitted to the intelligent scoring module.

[0027] The data calculation module performs physical unit conversion and core experimental data calculation based on the pixel data of the experimental equipment, outputting accurate core experimental data to provide a basis for the core data measurement and evaluation of the intelligent scoring module. The data calculation module receives the experimental equipment category, bounding box coordinates, and pixel-level segmentation mask transmitted by the target detection and segmentation module, performs correlation analysis and matching positioning on various types of data (corresponding the equipment category, bounding box, and segmentation mask one by one), and extracts the feature data corresponding to the experimental object equipment and the bounding box coordinates and pixel-level segmentation mask data corresponding to the reference object equipment according to the preset equipment type classification. It accurately obtains the pixel size information (such as the pixel length of the ruler and the pixel scale spacing of the measuring cup) from the segmentation mask of the reference object equipment.

[0028] By combining the known physical dimensions of the reference equipment (e.g., how many pixels correspond to 1cm on a ruler, how many pixel areas correspond to 1mL in a measuring cup) with the extracted pixel size information, a conversion factor between pixels and physical units is calculated using a proportional conversion method. Based on the calculated conversion factor, the pixel-level distance, position, area, angle, and other data of the experimental equipment are converted into corresponding data in actual physical units. Combined with preset experimental basic parameters, the derivation and calculation of the core experimental data are completed according to the professional calculation rules of physical experiments. The rationality of the calculated core experimental data is verified, and abnormal data is eliminated using the 3σ principle. At the same time, a second verification is performed based on common sense in physical experiments. Finally, the valid core experimental data is determined and transmitted to the intelligent scoring module, providing accurate data basis for core data measurement and evaluation.

[0029] The intelligent scoring module receives experimental equipment categories from the target detection and segmentation module, key experimental equipment working status results from the status judgment module, and core experimental data from the data calculation module. It then structures and organizes all received data to form a basic scoring dataset.

[0030] For multi-stage scoring calculation: This implementation method divides the physics experiment scoring into three core stages, each scored independently, with a maximum score of 100 points for each stage. Figure 3 As shown, the specific steps for calculating the scores at each stage and the final total score, and outputting the scores at each stage and the final total score, are as follows: Phase 1: Equipment Selection Evaluation: Based on the pre-set list of necessary experimental equipment, the number of necessary equipment detected by the target detection and segmentation module is counted, and the score for this phase is calculated proportionally; if non-necessary equipment is detected, no points are deducted or added.

[0031] Phase 2: Installation Correctness Assessment: The score for this phase is calculated based on the working status results of the key experimental equipment output by the status judgment module. If the working status of all preset key experimental equipment is normal, the full score for this phase is 100 points; if the working status of any key experimental equipment is abnormal, the score for this phase is 0 points, thus achieving a rigid assessment of the correctness of the installation and debugging of experimental equipment.

[0032] Phase 3: Core Data Measurement and Evaluation: The score for this phase is calculated based on the error range between the core experimental data output by the data calculation module and the preset standard value. Multiple error level ranges and corresponding scores are pre-configured, and the error range and score can be adjusted according to the experimental accuracy requirements.

[0033] Final total score calculation: The final total score is calculated by weighted summation according to the pre-set weighting coefficients for each evaluation stage. The calculation formula is as follows: ,in, It is the final total score for the physics experiment assessment; This is the actual score during the equipment selection and evaluation phase. These are the preset weighting coefficients in the equipment selection and evaluation phase. This is the actual score during the installation correctness assessment phase. These are the preset weighting coefficients for the installation correctness assessment phase. It is the actual score in the core data measurement and evaluation phase. These are the preset weighting coefficients for the core data measurement and evaluation phase. In this implementation method, the preferred weighting coefficients are: =0.2、 =0.3、 =0.5, highlighting the evaluation weight of core experimental data measurement, which can be flexibly adjusted according to the type of experiment.

[0034] Scoring results output: The scores for the three stages of equipment selection, installation correctness, and core data measurement, as well as the final total score of the experiment, are structured and packaged, and transmitted to the Web interactive module for display.

[0035] The web interaction module is built based on a B / S architecture, featuring a web front-end interface with three main functional areas: a data upload area, a results display area, and a parameter setting area. The data upload area supports local uploading and online preview of experimental images / video frames, displaying upload progress and file format verification results. The results display area uses a visual approach to show the experimental equipment test results (including equipment category, location markings, and segmentation masks) and scoring results (scores at each stage, final total score, and scoring details). The parameter setting area allows users to flexibly adjust parameters such as preset pixel specifications, scoring weight coefficients, and error level ranges without modifying the underlying code.

[0036] The web backend adopts an interface-based development approach, providing data upload interfaces, model inference interfaces, and scoring result query interfaces. The interface communication protocol uses HTTP / HTTPS to ensure data transmission security. After receiving experimental image / video frames uploaded by users, the backend verifies the file format, discards invalid files, and transmits valid files to the image preprocessing module. At the same time, it receives experimental equipment category data transmitted by the target detection and segmentation module, and scores for each stage and the final experimental score transmitted by the intelligent scoring module, and integrates and processes the two types of data.

[0037] The integrated test results and scoring results are sent back to the web front end for visualization in the results display area: the experimental images are labeled with the location of the equipment and overlaid with segmentation masks to intuitively display the detected experimental equipment; the scores of each stage and the final total score are displayed in the form of bar charts and digital panels, while providing a scoring detail query function to display information such as missed / excessive detection of equipment, status judgment results of key equipment, and comparison of the measured values ​​and standard values ​​of core data, so that users can easily view the scoring basis.

[0038] like Figure 2 As shown, this invention provides a scoring method for university physics experiments based on YOLOv8, and its specific implementation steps are as follows: S1: Load the pre-trained YOLOv8-seg instance segmentation model and key equipment status classification model, and configure the preset parameters for model inference, including preset pixel size 640×640, inference confidence threshold 0.5, non-maximum suppression threshold 0.5, and classification model confidence threshold 0.6; at the same time, in the Web parameter setting area of ​​the intelligent scoring module, set the preset weight coefficients and preset scoring rules for multi-stage weighted scoring according to the experiment type; S2: Users upload experimental image or video frame files through the front-end data upload area of ​​the Web interaction module. After the Web backend completes file format verification and valid frame screening, it transmits the processed raw experimental data to the image preprocessing module.

[0039] S3: The image preprocessing module performs format normalization, affine transformation, and pixel dimension calibration on the received experimental images or video frames, unifies the image size to the preset 640×640 pixel specification, eliminates pixel distortion problems, outputs the standardized RGB image, and transmits it to the target detection and segmentation module.

[0040] S4: The object detection and segmentation module converts standardized images into tensor data, loads pre-trained model weights to complete inference, and outputs the category of experimental equipment, bounding box coordinates and pixel-level segmentation mask through multi-scale feature extraction, anchor box matching, segmentation mask generation and post-processing. These are then transmitted to the state judgment module, data calculation module and intelligent scoring module, respectively.

[0041] S5: The intelligent scoring module first counts the number of necessary equipment to be tested based on the type of received experimental equipment, and calculates the equipment selection evaluation score; at the same time, the status judgment module completes the status detection of key experimental equipment, outputs the working status result and transmits it to the intelligent scoring module, which calculates the installation correctness evaluation score; the data calculation module completes the calculation of pixel-to-physical unit conversion coefficient and experimental core data, outputs the valid experimental core data and transmits it to the intelligent scoring module, which calculates the core data measurement evaluation score based on the error range.

[0042] S6: The intelligent scoring module uses a weighted summation formula based on the preset weighting coefficients for each evaluation stage. Calculate the final total score for the experiment; S7: The intelligent scoring module transmits the scores of each stage and the final total score of the experiment to the Web interaction module, and the target detection and segmentation module transmits the category data of the experimental equipment to the Web interaction module. The Web interaction module integrates and processes the two types of data, and displays the detection results and scoring results of the experimental equipment in a visual manner in the front-end result display area. Users can view the scoring details and basis online.

[0043] This implementation method achieves multi-target detection and pixel-level segmentation of experimental equipment using the YOLOv8-seg model, resulting in high detection accuracy and fast inference speed. It can accurately identify different types and specifications of university physics experimental equipment. Through image preprocessing and pixel-to-physical unit conversion algorithms, the measurement accuracy of core experimental data is effectively improved, solving the error problem of manual measurement. Through multi-stage weighted scoring rules, the standardization, intelligence, and automation of physics experiment scoring are achieved, eliminating the subjectivity of manual scoring. The Web interaction module enables convenient uploading of experimental data and visualization of scoring results, making it simple to operate and easy for university teachers and students to use.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A grading system for university physics experiments based on YOLOv8, characterized in that... The system includes the following components: Image preprocessing module: Receives experimental images or video frames, performs affine transformation on them for standardization, unifies the image size to a preset pixel specification, and outputs the standardized image; The target detection and segmentation module receives the standardized image, performs multi-target detection and instance segmentation of the experimental equipment through YOLOv8-seg model inference, and outputs the category, bounding box coordinates, and pixel-level segmentation mask of each piece of equipment. Status determination module: Receives the category and bounding box coordinates of the experimental equipment, performs status detection on the preset key experimental equipment, and outputs the working status results of the key experimental equipment; Data Calculation Module: Receives the type of experimental equipment, bounding box coordinates, and pixel-level segmentation mask; extracts data of the experimental object and reference equipment; calculates the conversion coefficient between pixels and physical units; completes the calculation of core experimental data; and outputs the core experimental data. Intelligent scoring module: Receives the category of experimental equipment, the working status results of key experimental equipment, and core experimental data; calculates the scores for each stage and the final total score according to preset scoring rules; and outputs the scores for each stage and the final total score. Web Interaction Module: Enables the uploading and display of experimental images or video frames and scoring results. It receives experimental images or video frames uploaded by users and transmits them to the image preprocessing module. It also receives the category of experimental equipment, the scores of each stage and the final total score of the experiment, integrates them and outputs the results for display.

2. The university physics experiment scoring system based on YOLOv8 according to claim 1, characterized in that... The image preprocessing module receives experimental images or video frame data from external sources. First, it normalizes the received raw data. For video frame data, it extracts individual frames and filters valid frames. For experimental images of different formats, it performs a unified conversion of color space and pixel format, forming standardized basic image data to be processed. Then, it determines the size parameters of the target image based on preset pixel specifications. An affine transformation matrix is ​​constructed based on these parameters, and an affine transformation is performed on the basic image data. Coordinate mapping is used to establish the correspondence between pixels in the original image and pixels in the target image. Simultaneously, an interpolation algorithm is used to fill pixel gaps generated during the transformation, eliminating pixel distortion caused by scaling and angle correction. After the affine transformation, the image undergoes final pixel dimension calibration to ensure that the size parameters of the processed image perfectly match the preset pixel specifications, ultimately outputting a standardized image with uniform specifications and consistent pixel features.

3. The university physics experiment scoring system based on YOLOv8 according to claim 1, characterized in that... The target detection and segmentation module receives the standardized image transmitted by the image preprocessing module, converts the image data into tensor data adapted to the YOLOv8-seg model input format, loads the pre-trained model weights, and configures the basic inference parameters. Then, it initiates the model inference process, extracting multi-scale features from the standardized image through the model's feature extraction network to generate image feature maps at different levels. The detection head performs multi-target detection of the experimental equipment based on the feature maps, outputting the category information and bounding box coordinates of the experimental equipment through anchor box matching and classification regression calculation. The segmentation head performs upsampling and feature fusion processing on the feature maps to generate mask feature maps matching the input image size. After pixel-level classification, pixel-level segmentation masks for each experimental equipment are obtained. After inference, non-maximum suppression processing is performed on the detection head output results, and duplicate or low-confidence detection boxes are removed according to the intersection-union ratio (IUGR) calculation formula. The mask feature maps output by the segmentation head are then subjected to coordinate mapping and binarization processing. Finally, the categories of the experimental equipment, accurate bounding box coordinates, and pixel-level segmentation masks corresponding to each equipment are integrated and output.

4. The university physics experiment scoring system based on YOLOv8 according to claim 3, characterized in that... The target detection and segmentation module outputs the category information and bounding box coordinates of the experimental equipment through anchor box matching and classification regression calculation. The formula for calculating the bounding box coordinates is: , ,in, The coordinates of the center of the experimental equipment bounding box are: This refers to the offset of the bounding box center coordinates predicted by the model. For sigmoid activation function, The center coordinates of the preset anchor frame, The width and height of the experimental equipment bounding box, The bounding box width and height offsets predicted by the model. Set the width and height of the preset anchor frame.

5. The university physics experiment scoring system based on YOLOv8 according to claim 3, characterized in that... The target detection and segmentation module eliminates duplicate or low-confidence detection boxes based on the intersection-union ratio (IU) calculation formula, which is: ,in, For the first The and the first The intersection-union ratio of the bounding boxes of the experimental equipment. , The first The, the The area of ​​the pixel region corresponding to the bounding box of each experimental device.

6. The university physics experiment scoring system based on YOLOv8 according to claim 1, characterized in that... After receiving the experimental equipment category and bounding box coordinates transmitted by the target detection and segmentation module, the state judgment module filters and matches the received data, locates the category information and bounding box coordinates corresponding to the preset experimental equipment, and verifies the effective range of the bounding box coordinates. Then, based on the verified bounding box coordinates, it performs targeted local cropping operations on the original experimental image, extracting only the image region containing the key experimental equipment. Then, it performs preprocessing operations such as size normalization adjustment and pixel value standardization on the local image, converting the processed image data into an input format suitable for the classification model. After completing the image preprocessing, it loads the pre-trained classification model weights and configures the corresponding inference parameters, inputs the preprocessed local image of the experimental equipment into the classification model for inference, obtains the preliminary state result through the model's feature extraction and classification judgment process, and then performs post-processing verification on the result, removes abnormal judgment data, and finally determines the working state result of the preset key experimental equipment.

7. The university physics experiment scoring system based on YOLOv8 according to claim 1, characterized in that... After receiving the experimental equipment category, bounding box coordinates, and pixel-level segmentation mask transmitted by the target detection and segmentation module, the data calculation module first performs correlation analysis and matching positioning on various types of data. Based on the preset equipment type classification, it extracts the feature data corresponding to the experimental object equipment and the bounding box coordinates and pixel-level segmentation mask data corresponding to the reference equipment. It accurately obtains the pixel size information from the pixel data of the reference equipment. Then, it sorts and organizes the extracted experimental object data according to the preset coordinate dimensions to eliminate the calculation deviation caused by the disorder of the data. Combining the known actual physical size of the reference equipment with the extracted pixel size, it calculates the conversion coefficient between pixels and physical units through proportional conversion. Based on the conversion coefficient, it converts the pixel-level distance, position, and other data of the experimental object into corresponding data in actual physical units. Then, combined with the preset experimental basic parameters, it completes the derivation calculation of the experimental core data according to the professional calculation rules of physical experiments. Finally, it verifies the rationality of the calculated experimental core data, removes abnormal data caused by pixel extraction and data conversion, determines the valid experimental core data, and outputs it in a targeted manner, providing accurate data basis for the core data measurement and evaluation of the intelligent scoring module.

8. The university physics experiment scoring system based on YOLOv8 according to claim 1, characterized in that... The intelligent scoring module calculates the scores for each stage and the final total score based on preset scoring rules, and outputs the scores for each stage and the final total score in the experiment. The specific steps are as follows: The first stage of equipment selection evaluation: Based on the pre-set necessary experimental equipment, the score for this stage is calculated according to the proportion of the number of necessary equipment detected to the total number of necessary equipment. The second stage of installation correctness assessment: The score for this stage is calculated based on the working status results of the key experimental equipment output by the status judgment module. The key experimental equipment receives full marks if it is in the preset normal working state; otherwise, it receives zero marks. The third stage is the core data measurement and evaluation: the score for this stage is calculated based on the error range between the core experimental data output by the data calculation module and the preset standard value, and the corresponding score is configured according to the preset error level interval; The final total score of the experiment is calculated by weighted summation according to the pre-set weighting coefficients of each evaluation stage.

9. The university physics experiment scoring system based on YOLOv8 according to claim 8, characterized in that... The intelligent scoring module calculates the final total score of the experiment by weighted summation according to the preset weight coefficients for each evaluation stage. The calculation formula is as follows: ,in, It is the final total score for the physics experiment assessment; This is the actual score during the equipment selection and evaluation phase. These are the preset weighting coefficients in the equipment selection and evaluation phase. This is the actual score during the installation correctness assessment phase. These are the preset weighting coefficients for the installation correctness assessment phase. It is the actual score in the core data measurement and evaluation phase. These are the preset weighting coefficients for the core data measurement and evaluation phase.

10. A YOLOv8-based scoring method for university physics experiments, applicable to the YOLOv8-based scoring system for university physics experiments as described in any one of claims 1-9, characterized in that... The method includes: S1: Load the pre-trained YOLOv8-seg model and classification model, configure the preset parameters for model inference, and set the preset weights and preset scoring rules for multi-stage weighted scoring in the intelligent scoring module; S2: Upload experimental images or video frames via the Web interaction module and transmit them to the image preprocessing module; S3: The image preprocessing module performs affine transformation on the experimental images or video frames, unifies the size to a preset pixel specification, outputs the standardized image, and transmits it to the target detection and segmentation module; S4: The object detection and segmentation module performs model inference on the standardized image and outputs the category of the experimental equipment, bounding box coordinates, and pixel-level segmentation mask, which are then transmitted to the state judgment module, data calculation module, and intelligent scoring module, respectively. S5: The intelligent scoring module first calculates the equipment selection evaluation score based on the category of experimental equipment; S6: The intelligent scoring module calculates the final total score of the experiment based on the evaluation scores at each stage and the corresponding preset weights; S7: The intelligent scoring module transmits the scores of each stage and the final total score of the experiment to the Web interaction module. The target detection and segmentation module transmits the category of the experimental equipment to the Web interaction module. The Web interaction module integrates these data and outputs the results for display.

Images