An experimental scoring system for exploring factors related to internal pressure of liquid based on machine vision

Abstract

The application relates to an experimental scoring system for exploring factors related to internal pressure of liquid based on machine vision, and belongs to the fields of computer vision and intelligent education. The application can realize fast and accurate scoring of the operation of an examinee in an experiment for exploring factors related to internal pressure of liquid by a computer. The system is divided into a visual detection module, a scoring decision module and an interface visualization module. Firstly, visual detection features are determined according to experiment requirements to form a scoring scheme. Secondly, images corresponding to each scoring feature are collected for labeling to make an image set, and a visual intelligent detection model Dila-Yolo v3 is constructed and trained to form the visual detection module. Then, an intelligent scoring decision mechanism is established according to the scoring scheme and model test results. Finally, an intelligent scoring interface is designed, a visual detection algorithm and the scoring decision mechanism are integrated to form an intelligent scoring system which is applied to actual experiment examination to score students.

Description

TECHNICAL FIELD

[0001] The application relates to a machine vision-based experimental scoring system for exploring factors related to internal pressure of liquid, and belongs to the fields of computer vision and intelligent education. BACKGROUND

[0002] Intelligent experiment education is a development trend of future education, and an experiment for exploring factors related to internal pressure of liquid is a common experiment in a middle school examination, which can help students deeply understand the meaning of internal pressure of liquid in textbooks; however, the judgment of correctness of traditional experiment operation adopts a manual scoring mode, and the manual scoring is difficult to guarantee objectivity and fairness.

[0003] For example, Chinese invention patent CN201310186978.7 discloses a liquid pressure gauge, and the specification discloses that colored liquid (such as red ink) is filled into a pressure sensing box and a transparent rubber tube through a liquid inlet hole, a soft plug is covered, a syringe needle is used to pierce the soft plug to increase or decrease the colored liquid, the liquid level in the tube can be accurately adjusted to be just at the zero scale, the pressure sensing box is placed into the liquid to be measured, the rubber film is pressed to compress the colored liquid in the transparent rubber tube, and direct reading can be achieved; the above patent can prove the defects of the prior art.

[0004] Therefore, the application is improved, and a machine vision-based experimental scoring system for exploring factors related to internal pressure of liquid is provided. SUMMARY

[0005] (1) The technical problem to be solved by the application is that the judgment of correctness of traditional experiment operation adopts a manual scoring mode, and the manual scoring is difficult to guarantee objectivity and fairness.

[0006] (2) Technical scheme

[0007] In order to achieve the above application purpose, the application provides a machine vision-based experimental scoring system for exploring factors related to internal pressure of liquid, which comprises the following steps:

[0008] Step 1: dividing the experiment for exploring factors related to internal pressure of liquid into five steps, determining score points and finding corresponding visual judgment features, and forming an intelligent scoring scheme;

[0009] Step 2: collecting images corresponding to each key feature, and preparing a training set and a test set;

[0010] Step 3: taking a Yolo v3 model as a basis to construct a visual intelligent detection model Dila-Yolo v3 model;

[0011] Step 4: training and testing the visual detection model Dila-Yolo v3 using an experimental image training set;

[0012] Step five: According to the scoring scheme and model test results, an intelligent scoring decision mechanism is established;

[0013] Step six: Design an intelligent scoring interface, integrate visual detection algorithms and scoring decision mechanisms to form an intelligent scoring system;

[0014] Step seven: Select the experimental operation video of the examinee, identify the key scoring features through the visual detection module of the intelligent scoring system, and visualize the target information through the scoring interface module;

[0015] Step eight: According to the detection target information fed back by the visual detection module of the intelligent scoring system, the scoring decision module is used to judge the score and save all scoring details to the specified file.

[0016] Among them, the specific method of forming an intelligent scoring scheme in step one includes:

[0017] S1, the U-shaped tube pressure gauge metal box is placed in a certain depth of a deep glass barrel containing water with the downward orientation, and the depth is recorded;

[0018] S2, record the orientation and depth information of the metal box when changing the orientation of the metal box;

[0019] S3, change the depth of the metal box in the water and keep it unchanged, repeat S2;

[0020] S4, use salt water, the depth of the metal box is the same as the previous time in the water, repeat step S2;

[0021] S5, arrange the experimental instruments.

[0022] Among them, the method of preparing the training set and the test set in step two includes:

[0023] According to the scoring scheme, the key features that need to be intelligently detected in this experiment are summarized, including: "front view of deep glass barrel", "deep glass barrel view", "liquid surface in deep glass barrel", "metal box with upward orientation", "metal box with side orientation", "metal box with downward orientation";

[0024] Collect pictures containing key features to make sample sets, each feature contains, use LabelImg tool to label corresponding features and name classification, after labeling, LabelImg will convert the labeled pictures into.txt files containing corresponding picture names, target category names, image sizes, and target box position information, generating training and test data sets.

[0025] Among them, the method of constructing a visual intelligent detection model Dila-Yolo v3 model in step three includes:

[0026] Step A, based on Yolo v3, aiming at the problem of semantic information loss of small target detected by Yolo v3 model, a parallel hollow residual branch is designed to fuse shallow features;

[0027] Step B, improve the anchor box size matching strategy, speed up the prediction regression of small scale target.

[0028] Step A includes:

[0029] In the feature extraction network-Darknet53, the 36th layer is designed as an empty residual branch to replace the original up-sampling channel. The features output by the empty residual branch are fused with the output of the 36th layer to form a 52*52 small-scale detection channel.

[0030] After down-sampling at the end of the empty residual branch, the features in the middle layer of the feature extraction network are fused to form a 26*26 medium-scale channel.

[0031] Step B includes:

[0032] Use K-means algorithm to re-cluster anchor boxes, and use intersection over union as distance loss indicator;

[0033] The product of the length-width ratio of the predicted box and the real box is added to the original distance loss function as the scale influence term, which changes adaptively with the anchor box scale to balance the influence of different scale targets on the final prediction accuracy in the training process, as shown in the following formula:

[0034]

[0035] In the formula, w b , h b are the length and height of the predicted box, w c , h c are the length and height of the real box, and σ is the Sigmoid transformation, which maps the value in the parentheses to 0-1. In the formula, is the scale information term, and λ is the weight coefficient.

[0036] The method for training the visual intelligent detection model Dila-Yolo v3 model in step four includes:

[0037] The sample batch size of the training is set to 8, and there are 300 iteration cycles. The initial learning rate of the training is set to 0.001, and the learning rate adjustment adopts the epoch-decay strategy, which decreases with the iteration cycle.

[0038] The method for establishing the scoring decision mechanism in step five includes:

[0039] In addition to arranging the experimental instruments, each step of the experiment requires the examinee to input the water depth value and the U-shaped tube liquid level difference, that is, the experiment needs to be strictly carried out according to the experimental step sequence;

[0040] In the algorithm implementation process, all experimental steps start detection at the same time, and a start signal and an end signal are set for each experimental step, and when the start signal and the end signal of a step occur, the score of the step is obtained.

[0041] When the start signal of a step occurs, if the start signal or the end signal of the previous step does not occur, the score of the previous step is set to 0, and only the subsequent step is detected when the step ends.

[0042] The method for designing the intelligent scoring interface in the sixth step comprises:

[0043] The python tkinter framework is used to design the intelligent scoring interface to realize the visual scoring process.

[0044] The interface functions include video path selection detection function, real-time visualization of target detection function and score information display function.

[0045] (Three) beneficial effects

[0046] The scoring system for the experiment of exploring the related factors of the internal pressure of liquid based on machine vision has the beneficial effects that:

[0047] The scoring system for the experiment of exploring the related factors of the internal pressure of liquid based on machine vision has the beneficial effects that: BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0049] Figure 1 is a whole flow chart of an experimental scoring system for exploring liquid internal pressure related factors based on machine vision of the present application;

[0050] Figure 2 is a Dila-Yolo v3 visual detection model structure diagram improved by the present application;

[0051] Figure 3 is a partial result display of target detection on an experimental scene by the present application;

[0052] Figure 4 is an intelligent scoring interface display designed by the present application. DETAILED DESCRIPTION

[0053] The specific embodiments of the present application will be further described in detail below in combination with the accompanying drawings and examples. The following examples are only used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0054] Embodiment:

[0055] As shown in Figure 1 , the present embodiment proposes an experimental scoring system for exploring liquid internal pressure related factors based on machine vision, including the following steps:

[0056] Step one: divide the experiment for exploring liquid internal pressure related factors into five steps, determine the scoring points and find out the corresponding visual judgment features, and form an intelligent scoring scheme;

[0057] The specific method for forming the intelligent scoring scheme includes:

[0058] S1, place the U-shaped tube pressure gauge metal box with the downward orientation into a deep glass barrel containing water at a certain depth, and record;

[0059] S2, record the metal box orientation and depth information when changing the three orientations of the metal box;

[0060] S3, change the depth of the metal box in the water and keep the depth unchanged, and repeat S2;

[0061] S4, use salt water, place the metal box at the same depth as in the water, and repeat step S2; S5, arrange the experimental instruments;

[0062] The scoring scheme of the above method is shown in the following table:

[0063]

[0064]

[0065]

[0066] Step two: Collect images corresponding to each key feature, prepare training set and test set, the specific method includes:

[0067] According to the scoring scheme, the key features that need to be intelligently detected in this experiment are summarized, including: "front view of deep glass barrel", "panting view of deep glass barrel", "liquid surface in deep glass barrel", "metal box facing upward", "metal box facing sideways", "metal box facing downward". Collect pictures containing key features to make sample set;

[0068] Use LabelImg tool to label corresponding features and name classification. After labeling, LabelImg will convert the labeled pictures into.txt files containing corresponding picture names, target category names, image sizes, and target box position information, forming a training data set.

[0069] Step three: Based on Yolo v3 model, build a visual intelligent detection model Dila-Yolo v3 model, the overall structure is as shown in Figure 2 The specific method includes:

[0070] Step A, based on Yolo v3, to solve the problem of semantic information loss of small target detection in Yolo v3 model, design parallel hollow residual branch to fuse shallow features;

[0071] Step B, improve anchor box size matching strategy to speed up small scale target prediction regression;

[0072] Step A includes:

[0073] In the feature extraction network-Darknet53, the 36th layer is designed with a hollow residual branch to replace the original up and down sampling channel. The features output by the hollow residual branch are fused with the output of the 36th layer to form a 52*52 small scale detection channel;

[0074] After down-sampling at the end of the hollow residual branch, the features in the middle layer of the feature extraction network are fused to form a 26*26 medium scale channel;

[0075] Step B includes:

[0076] Use K-means algorithm to re-cluster anchor boxes in the data set, and use intersection over union as distance loss indicator;

[0077] The product of the length-width ratio of the prediction box and the real box is added to the original distance loss function as a scale influence term, which is self-adaptive to the anchor box scale to balance the influence of different scale targets on the final prediction accuracy in the training process, as shown in the following formula:

[0078]

[0079] In the formula, w b , h b are the length and height of the prediction box, w c , h c are the length and height of the real box, and σ is the Sigmoid transformation, which maps the value in the parentheses to 0-1, where is the scale information term, and λ is the weight coefficient.

[0080] Step four: use the experimental image training set to train and test the visual detection model Dila-Yolo v3, and the detection image is as shown in Figure 3 , and the specific method comprises:

[0081] The sample batch size of the training is set to 8, and the training is performed for a total of 300 iteration cycles (epochs), and the weight is updated once every 10 iterations;

[0082] The initial learning rate of the training is set to 0.001, and in order to prevent overfitting of the model during training, the learning rate is adjusted using the epoch-decay strategy, which decreases with the iteration cycle;

[0083] The recognition accuracy P, the recognition recall rate R, and the average precision mean mAP under the IOU threshold parameter (output box and real box intersection over union) of 0.5 are used as the model performance evaluation indexes.

[0084] Step five: according to the scoring scheme and the model test results, an intelligent scoring decision mechanism is established, and the specific method comprises:

[0085] The liquid internal pressure related factor experiment needs to input the water depth value and the U-shaped tube liquid level difference of the examinee in each step except for arranging the experimental instruments, that is, the experiment needs to be strictly carried out according to the experimental step sequence;

[0086] In the algorithm implementation process, all experimental steps start detection at the same time, a start signal and an end signal are set for each experimental step, and when the start signal and the end signal of a step occur, the score of the step is obtained;

[0087] When the start signal of a step occurs, if the start signal or the end signal of the previous step does not occur, the score of the previous step is set to 0, and only the subsequent step is detected at the end of the step, as shown in the following table:

[0088]

[0089]

[0090] Step six: design an intelligent scoring interface, integrate visual detection algorithms, and scoring decision mechanisms to form an intelligent scoring system, including the following specific methods:

[0091] An intelligent scoring interface is designed using the python tkinter framework to visualize the scoring process.

[0092] As shown in Figure 4 The interface functions include: video path selection function for detection, real-time visualization function for detection targets, score information display function, etc.

[0093] Integrate visual detection and scoring logic algorithms to form an intelligent scoring system.

[0094] Step seven: select the examinee's experimental operation video, identify key scoring features through the visual detection module of the intelligent scoring system, and visualize target information through the scoring interface module.

[0095] Step eight: based on the detection target information fed back by the visual detection module of the intelligent scoring system, judge the score through the scoring decision module and save all scoring details to a specified file.

[0096] The present application constructs an experimental scoring system based on machine vision for exploring the related factors of internal pressure of liquid, which provides experimental operation information for the scoring decision module of the system through improved visual detection algorithms, and can realize fast and accurate intelligent scoring.

[0097] The above embodiments are only used to illustrate the present application, and are not intended to limit the present application. Although the present application has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that various combinations, modifications or equivalent replacements of the technical solutions of the present application do not deviate from the spirit and scope of the present application, and should be covered in the scope of the claims of the present application.

Claims

1. A machine vision-based experimental scoring system for exploring factors related to internal pressure of liquid, comprising the following steps: Step one: divide the experiment for exploring factors related to internal pressure of liquid into five steps, determine the scoring points and find the corresponding visual judgment features, and form an intelligent scoring scheme; the specific method for forming the intelligent scoring scheme in step one includes: S1, place the U-shaped tube pressure gauge metal box in a certain depth of the deep glass barrel containing water with the downward orientation, and record; S2, record the orientation and depth information of the metal box when changing the orientation of the metal box; S3, change the depth of the metal box in the water and keep the depth unchanged, and repeat S2; S4, use salt water, the depth of the metal box is the same as a certain time in the water, and repeat step S2; S5, arrange the experimental instruments; Step two: collect images corresponding to each key feature, prepare training set and test set; Step three: based on Yolo v3 model, build a visual intelligent detection model Dila-Yolo v3 model; Step four: use the experimental image training set to train and test the visual detection model Dila-Yolo v3; Step five: according to the scoring scheme and the model test result, establish an intelligent scoring decision mechanism; Step six: design an intelligent scoring interface, integrate the visual detection algorithm and the scoring decision mechanism, and form an intelligent scoring system; Step seven: select the examinee's experimental operation video, identify the key scoring features through the visual detection module of the intelligent scoring system, and visualize the target information through the scoring interface module; Step eight: according to the detection target information fed back by the visual detection module of the intelligent scoring system, judge the score through the scoring decision module and save all the scoring details to the specified file; The method for building the visual intelligent detection model Dila-Yolo v3 model in step three includes: Step A: based on Yolo v3, design a parallel hollow residual branch to fuse shallow features in order to solve the problem of semantic information loss of small target detection in Yolo v3 model; Step B: improve the anchor box size matching strategy to speed up the prediction regression of small scale targets; Step A includes: At the 36th layer of the feature extraction network-Darknet53, a hollow residual branch is designed to replace the original up and down sampling channel, the features output by the hollow residual branch are fused with the output of the 36th layer, forming a 52*52 small scale detection channel; After down-sampling at the end of the hollow residual branch, the middle layer features of the feature extraction network are fused to form a 26*26 medium scale channel; Step B includes: Use K-means algorithm to re-cluster the anchor box of the data set, and use the intersection over union as the distance loss index; The product of the length-width ratio of the predicted box and the real box is added to the original distance loss function as a scale influence item, which changes adaptively with the anchor box scale, to balance the influence of different scale targets on the final prediction accuracy in the training process, as shown in the following formula: where , are the length and height of the prediction box, respectively, , are the length and height of the ground truth box, respectively, is a Sigmoid transformation mapping the value in the brackets to 0-1, where is the scale information term, is the weight coefficient.

2. The machine vision-based experimental scoring system for exploring factors related to internal pressure of a liquid according to claim 1, wherein, The method for preparing the training set and test set in step two includes: The key features to be intelligently detected in this experiment are summarized according to the scoring scheme, including: "front view of deep glass barrel", "top view of deep glass barrel", "liquid surface in deep glass barrel", "metal box facing upward", "metal box facing sideways", and "metal box facing downward"; Collect pictures containing key features to make a sample set, each containing a certain number of features, use LabelImg tool to label the corresponding features and name the categories, after labeling, LabelImg will convert the labeled pictures into.txt files containing the corresponding picture name, target category name, image size, and target box position information, generating training and testing data sets.

3. The machine vision-based experimental scoring system for exploring factors related to internal pressure of a liquid according to claim 1, wherein, The method for training the visual intelligent detection model Dila-Yolo v3 model in step four includes: The sample batch size for training is set to 8, with a total of 300 iteration cycles, and the initial learning rate for training is set to 0.

001. The learning rate adjustment uses the epoch-decay strategy, which decreases with the iteration cycle.

4. The machine vision-based experimental scoring system for exploring factors related to internal pressure of a liquid according to claim 1, wherein, The method for establishing a scoring decision mechanism in step five includes: The liquid internal pressure related factor experiment requires the examinee to input the water depth value and the U-shaped tube liquid level difference at each step, except for organizing the experimental instruments, i.e., the experiment needs to be strictly followed in the order of experimental steps; In the algorithm implementation process, all experimental steps start detection at the same time, each experimental step is set with a start signal and an end signal, and when the start signal and the end signal of a certain step occur, the step score is obtained; When the start signal of a certain step occurs, if the start signal or the end signal of the previous step has not occurred, the score of the previous step is set to 0, and only the subsequent step is detected at the end of the step.

5. The machine vision-based experimental scoring system for exploring factors related to internal pressure of a liquid according to claim 1, wherein, The method for designing an intelligent scoring interface in step six includes: Designing an intelligent scoring interface using the python tkinter framework to visualize the scoring process; The interface functions include: video path selection function for detection, real-time visualization function for detection targets, and score information display function.

Images