Quantitative evaluation method, model and system for juggling training

Abstract

The present application belongs to the technical field of chef training, and discloses a method, a model and a system for quantitatively evaluating spoon-flipping training, comprising the following steps: fixing a collecting device behind a pot and starting spoon-flipping training, wherein the collecting device can collect multiple types of action data in multiple directions; obtaining multiple training data collected by the collecting device, wherein one type of action in one direction corresponds to one training data; calculating the fitting degree between each training data and a corresponding full-score curve, wherein the full-score curve is obtained by a variational autoencoder according to action data of a target person flipping a spoon and expert evaluation indexes; and obtaining an evaluation result according to all fitting degrees. The method can monitor each action in the practice process of a student, quantitatively evaluate the action quality, help the student find errors and deficiencies in the practice process, and give reasonable suggestions to improve teaching efficiency and training effect.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of chef training, and particularly relates to a method, a model and a system for quantitatively evaluating wok tossing training. BACKGROUND

[0002] In chef professional skill training, wok tossing is a key basic skill, which directly affects the uniformity of food heating, taste and cooking efficiency. Traditional wok tossing training mainly relies on repeated practice by students, and the action quality is judged by experienced teachers through visual observation, including indicators such as material throwing height, landing uniformity, action continuity and stability. However, this method has many subjective factors, and different teachers may have different evaluation standards, resulting in inconsistent feedback for students. Moreover, teachers need to focus on observation for a long time, which is easy to cause fatigue, affecting the accuracy of evaluation, especially in the case of a large number of students, the workload will be very large and the efficiency is low.

[0003] At present, some technologies attempt to assist training through simple sensors or timing devices, such as counting the number of wok tossing times, but their functions are single and cannot perform multi-dimensional analysis of action quality. In addition, although existing motion capture devices can collect action data, they are high in cost and complex in operation, and are not optimized for wok tossing scenarios, making it difficult to popularize and apply in the field of chef training.

[0004] Therefore, there is an urgent need for a low-cost, high-precision and chef industry-adapted wok tossing training quantitative evaluation technology to improve teaching efficiency and training effect. SUMMARY

[0005] The purpose of the present application is to provide a method, a model and a system for quantitatively evaluating wok tossing training, which can improve teaching efficiency and training effect.

[0006] The first aspect of the present application discloses a method for quantitatively evaluating wok tossing training, comprising:

[0007] Fixing the collection device behind the wok and starting the wok tossing training, the collection device can collect multiple types of action data in multiple directions;

[0008] Obtaining multiple training data collected by the collection device, wherein one type of action in one direction corresponds to one training data;

[0009] Calculating the fitting degree between each training data and the corresponding full score curve, the full score curve is obtained by a variational autoencoder according to the action data of the collection target person and the expert evaluation index when wok tossing;

[0010] Obtaining an evaluation result according to all fitting degrees.

[0011] In some embodiments, the evaluation result is obtained according to all the fitting degrees, and the evaluation result comprises:

[0012] The training scores are obtained by weighting and accumulating all the fitting degrees.

[0013] In some embodiments, the evaluation result is obtained according to all the fitting degrees, and the evaluation result further comprises:

[0014] The action language features are obtained according to each of the fitting degrees, and the training suggestions are obtained according to the action language features.

[0015] In some embodiments, the collection device is a gyroscope, and the gyroscope collects angular velocity, acceleration and angular acceleration in three-dimensional directions.

[0016] In some embodiments, the expression of the full score curve is Y = Asin (Bx + Ct + D) + E, wherein A, B, C, D and E are parameters obtained by using a variational autoencoder, x is a data set composed of action data, and t is a time sequence corresponding to the action data.

[0017] In some embodiments, after the training data is obtained, the training data is further filtered according to a selected time period, and a line chart is generated and displayed according to the filtered training data.

[0018] The second aspect of the present application discloses a spoon training quantitative evaluation model, comprising:

[0019] A training data module is configured to obtain a plurality of training data collected by a collection device, wherein one type of action in one direction corresponds to one of the training data, and the collection device can collect a plurality of types of action data in multiple directions.

[0020] A fitting degree calculation module is configured to calculate the fitting degree between each of the training data and a corresponding full score curve, and the full score curve is obtained by using a variational autoencoder according to action data and expert evaluation indexes when a target person is collecting spoons.

[0021] An evaluation result module is configured to obtain an evaluation result according to all the fitting degrees.

[0022] In some embodiments, the collection device is a gyroscope, and the gyroscope collects angular velocity, acceleration and angular acceleration in three-dimensional directions.

[0023] In some embodiments, a full score curve construction module is further included, and the full score curve construction module is configured to obtain the full score curve by using a variational autoencoder according to action data and expert evaluation indexes when a target person is collecting spoons.

[0024] The third aspect of the present application discloses a spoon training quantitative evaluation system, comprising:

[0025] The acquisition device is used for acquiring multiple types of action data in multiple directions during the juggling training;

[0026] The cloud platform is used for receiving and storing the action data acquired by the acquisition device;

[0027] The juggling training quantitative evaluation model as any one of the above is used for analyzing the action data generated by the juggling training, and giving an evaluation result;

[0028] The display model is used for displaying the evaluation result.

[0029] The juggling training quantitative evaluation system disclosed by the embodiment of the present application has the beneficial effects that the acquisition device is used for acquiring multiple types of action data in multiple directions, and then the pre-obtained full score curve is used as a quantitative standard to calculate the fitting degree of each action data, and the evaluation is performed through the fitting degree to obtain the evaluation result. The juggling training quantitative evaluation system can monitor each action in the practice process of the student, quantitatively evaluate the action quality, help the student find the errors and deficiencies in the practice process, and give reasonable suggestions to improve the teaching efficiency and the training effect. BRIEF DESCRIPTION OF DRAWINGS

[0030] The drawings of the present application show the specific examples of the technical solutions of the present application, and constitute a part of the specification together with the specific embodiments, and are used to explain the technical solutions, principles and effects of the present application.

[0031] Unless specifically stated or defined otherwise, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0032] Figure 1 is an architectural diagram of a juggling training quantitative evaluation system disclosed by an embodiment of the present application;

[0033] Figure 2 is a flowchart of a juggling training quantitative evaluation method disclosed by an embodiment of the present application;

[0034] Figure 3 is a perspective view of a printing connecting piece of an embodiment of the present application;

[0035] Figure 4 is a schematic diagram of an evaluation result of an embodiment of the present application. DETAILED DESCRIPTION

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. In the case of conflict between the present specification and indications of meaning of terms used in the art to which this application pertains, the present specification will control. All technical and scientific terms used herein are intended to have the same meaning as commonly understood by one of ordinary skill in the art to which the present application pertains. Unless specifically set forth herein, the terms "first", "second" and the like do not entail a limitation, but do not preclude different respective quantities and / or sequences in the real scene in which the technical solutions of the present application are implemented. The term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0037] It should be noted that when an element is considered to be "fixed" to another element, it can be directly fixed to the other element or a middle element can exist; when an element is considered to be "connected" to another element, it can be directly connected to the other element or a middle element can exist; when an element is considered to be "mounted" to another element, it can be directly mounted to the other element or a middle element can exist. When an element is considered to be "provided" to another element, it can be directly provided to the other element or a middle element can exist.

[0038] Unless otherwise specified or defined, "the", "said" used herein refers to the technical features or technical contents mentioned or described before the corresponding position, which can be the same as or similar to the technical features or technical contents mentioned. In addition, the terms "include" and "have" used herein and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units not listed, or optionally includes other steps or units inherent to the process, method, product or device.

[0039] Currently, there is no quantitative evaluation technology for the wok training characteristics of the chef industry, and students cannot improve the action details through quantitative data, and the training process lacks scientific and systematic support. The present application accurately records the key data of the wok action through the gyroscope to reflect the throwing angle, frequency, force and other parameters of the wok action, and then calculates the fitting degree between the recorded key data and the full score curve, and obtains the quantitative evaluation result by analyzing the fitting degree. It can help students find the mistakes and deficiencies in the practice process and give reasonable suggestions, reduce the workload of teachers, improve work efficiency and improve training quality.

[0040] Firstly, the embodiment provides a wok training quantitative evaluation system, which comprises a wok, a gyroscope, a data processing unit and a display unit. Figure 1As shown, it comprises: a collection device, a cloud platform, a wok training quantitative evaluation model and a display model. The collection device is a gyroscope fixed on the wok through a printing connector, used to collect various types of action data in multiple directions during wok training. In this embodiment, the gyroscope is used to collect action data such as angular velocity, acceleration and angular acceleration in three directions (x, y, z). In some examples, the collection device also includes a force sensor to collect the force of the wok action. The cloud platform is a storage platform provided by a cloud server to receive and store the action data collected by the collection device. The service provider of the cloud platform is not limited. The collection device is preferably a smart gyroscope, so that the data storage platform provided by the smart gyroscope manufacturer can be used as the cloud platform, saving system deployment costs and improving system deployment efficiency. It should be noted that the cloud platform can also not be set up, and the training data can be directly deployed locally after being collected.

[0041] The wok training quantitative evaluation model is deployed on an electronic terminal such as a PC, a notebook, a local server, a cloud computing server, etc. The wok training quantitative evaluation model can access the cloud platform to obtain the action data generated during wok training saved by the cloud platform, analyze the action data, and give an evaluation result.

[0042] The wok training quantitative evaluation model mainly includes the following modules: a training data module, a fitting degree calculation module and an evaluation result module. The training data module can obtain multiple training data collected by the collection device, wherein one type of action in one direction corresponds to one training data. Specifically, in this embodiment, the training data module is connected with the cloud platform to obtain the training data from the cloud platform.

[0043] The fitting degree calculation module calculates the fitting degree between each training data and the corresponding full score curve when running, wherein the full score curve is obtained by using a variational autoencoder according to the action data of the target person during wok training and the expert evaluation index. Each type of action in each direction corresponds to a full score curve. The above target person is a person who has mastered the wok action, such as a qualified chef, a teacher, etc. The action data of the target person during wok training is collected, and then the expert scores the wok action to obtain the expert evaluation index. The action data set is combined to obtain the expert evaluation index, and the training data set is input into the variational autoencoder to learn various parameters in the full score curve representation, thereby obtaining the full score curve. In this embodiment, the expression of the full score curve is: Y = Asin(Bx + Ct + D) + E, where A, B, C, D and E are parameters to be trained, i.e. parameters that need to be learned by using the variational autoencoder, x is a data set composed of wok data of the target person, and t is the corresponding time sequence.

[0044] Specifically, the spoon training quantitative evaluation model further comprises a full score curve construction module, which is used to find a suitable full score curve by collecting spoon action data and expert evaluation indicators. That is, a deep learning model is constructed by using a variational autoencoder to infer the full score curve. In this embodiment, a variational autoencoder (VAE) in a PyTorch deep learning model is used to fully extract data features and summarize data rules. After the data set of spoon actions is converted into a matrix, it is sent to the trained encoder network. The VAE learns model parameters by maximizing the likelihood probability of the data, and uses variational inference to approximate the learning of the latent distribution. A set of rigorous and scientific quantitative standards are established to meet the training goals of the chef industry by collecting a large amount of spoon action data and expert evaluation indicators to find a suitable full score curve.

[0045] After obtaining the full score curve, the angular velocity, acceleration, and angular acceleration of the nine training data collected in the three-dimensional direction (x, y, z) are converted into a curve form according to the time axis, and are fitted with the full score curve model: Asin(Bx+Ct+D)+E (such as using the function scipy.optimize.curve_fit() for fitting), and the fitting degree is calculated. The fitting degree is a decimal in the range of 0-1.

[0046] When the evaluation result module runs, the evaluation result is obtained according to all the fitting degrees. In this embodiment, the fitting degrees are weighted and accumulated in the evaluation result module to calculate the training score. The specific calculation formula is: S=[(af1+bf2+……if9) / 9]*100, wherein f n represents the fitting degree of each training data and the full score curve, and the value range is 0-1; a, b, …, i are the importance weights of each training data, which can be learned by a deep learning model or set according to expert opinions.

[0047] After the spoon training quantitative evaluation model gives the evaluation result, the evaluation result is displayed on the display model. In this embodiment, the display model is constructed by using Vue to make the front end and Django to make the network framework. The Vue front end and the Django network framework are written in a front-end and back-end separation mode. The goal of Vue is to achieve responsive data binding and combined view components through as simple a programming interface as possible, while Django provides a powerful tool and convention to enable developers to quickly build websites with complete functionality and easy maintenance, and obtain the display model.

[0048] In summary, the embodiment provides a kind of based on smart gyroscope, Django network framework and Pytorch deep model training quantitative evaluation system of training of turning over pot, the system includes hardware system, with software system two parts, hardware system includes the smart gyroscope for detecting the data of turning over pot and the printing connector for fixing gyroscope and turning over pot, software system includes cloud platform, quantitative evaluation model of training of turning over pot and display model;Wherein cloud platform is used to collect and store the training data generated when training of turning over pot;Quantitative evaluation model of training of turning over pot analyzes training data and obtains analysis result, then display analysis result on display model.By using the deep learning model of Pytorch to process various action data generated when turning over pot, and combining the operation evaluation given by expert to quantitatively evaluate the result and then evaluate the pros and cons of the training data of turning over pot.

[0049] The embodiment also provides a kind of quantitative evaluation method of training of turning over pot, as shown in Figure 2 The method comprises the following steps:

[0050] Step S100: after fixing acquisition device on turning over pot, start training of turning over pot, wherein acquisition device can collect multiple types of action data in multiple directions;

[0051] The acquisition device in the embodiment is smart gyroscope, which can collect 9 action data of angular velocity, acceleration and angular acceleration in x, y and z directions, wherein each action data is time series data corresponding to a time range.

[0052] The acquisition device is fixed on the turning over pot by the printing connector. As shown in Figure 3 The printing connector includes a fixing part 10 and a clamp 20, the fixing part 10 is provided with a first mounting hole 11, the printing connector and the handle of the turning over pot can be fixed by using fasteners through the first mounting hole 11, the clamp 20 is provided with a clearance gap 21, the edge of the turning over pot is buckled on the printing connector through the clearance gap 21, and the clamp 20 is provided with a second mounting hole 22, the turning over pot and the printing connector can be fixed by using fasteners through the second mounting hole 22. In the embodiment, the printing connector is made of 3D printing, which is light and solid.

[0053] The acquisition device is fixed on the turning over pot to start training of turning over pot, the action data collected by the gyroscope is stored to the cloud platform by using the cloud platform provided by the smart gyroscope, the data storage is safe and convenient to use.

[0054] Step S200: obtain multiple training data collected by acquisition device, wherein one type of action in one direction corresponds to one training data;

[0055] After the spoon training is completed, a plurality of training data collected by the collection device is obtained from the cloud platform. In this embodiment, the training data is 9 in total, and one training data corresponds to one type of action (angular velocity, acceleration, and angular acceleration) in one direction (x, y, z). Obviously, when obtaining the training data, the training data can also be filtered, screened, and denoised. For example, in order to improve the data quality, the training data is screened according to the selected time period.

[0056] In some embodiments, after obtaining the training data, the training data can also be screened according to the selected time period, and a line chart is generated according to the screened training data and displayed to the trainee through the display model.

[0057] Step S300: Calculate the fitting degree between each training data and the corresponding full score curve, wherein the full score curve is obtained by the variational autoencoder according to the action data of the target person when spooning and the expert evaluation index;

[0058] In this embodiment, the variational autoencoder (VAE) in the PyTorch deep learning model is used. First, the action data of the target person when spooning is collected to construct a training data set, and then the expert scores the spooning action to obtain the expert evaluation index. Then, the training data set with the expert evaluation index of full score is combined and input into the variational autoencoder to learn various parameters represented by the full score curve, so as to obtain the full score curve. In this embodiment, the expression of the full score curve is Y = Asin(Bx+Ct+D)+E, wherein A, B, C, D, and E are parameters to be trained, i.e., parameters that need to be learned by using the variational autoencoder, x is a data set composed of the spooning data of the target person, and t is the corresponding time sequence.

[0059] After obtaining the full score curve, the angular velocity, acceleration, and angular acceleration of 9 training data collected in three directions (x, y, z) are respectively converted into curve form according to the time axis, and are fitted with the corresponding full score curve Asin(Bx+Ct+D)+E (such as using the function scipy.optimize.curve_fit() for fitting). The fitting degree is calculated, and the value of the fitting degree is a decimal in the range of 0-1. Each type of training data in each direction corresponds to one full score curve. It should be noted that the 9 full score curves do not mean that the parameters of all full score curves are different.

[0060] Step S400: Obtain the evaluation result according to all the fitting degrees.

[0061] All the fitting degrees are weighted and accumulated to calculate the training score. The specific calculation formula is S = [(af1+bf2+…if9) / 9]*100, wherein f nThe fitting degree represents the fitting degree of each training data to the perfect curve, and the value range is 0-1; a, b, …, i are the importance weights of each training data. The obtained results are shown in Table 1, wherein the scores are converted into corresponding grades, and the errors and deficiencies in the practice process are described. Figure 4

[0062] The embodiment also obtains corresponding action language features according to each fitting degree, and obtains training suggestions according to the action language features. For example: judging the differences of the 9 fitting degrees, and retrieving the corresponding action language features according to the differences, and giving training suggestions according to the action language features. Specifically, the action language features refer to cooking action problems described in natural language, such as “insufficient wrist turning” or “excessive forward and backward movement of the pot body”, etc. First, a mapping table or rule base is established to correspond the differences of the fitting degrees of the indicators (such as low fitting degree or fitting degree in a certain threshold interval) to specific action language features, and then the differences of the fitting degrees of the 9 indicators in three directions (x, y, z) and three types (angular velocity, acceleration, angular acceleration) are retrieved in the mapping table or rule base to obtain the corresponding action language features. For example: low z-axis angular acceleration fitting degree corresponds to “insufficient height of spoon overturning”, low x-axis acceleration fitting degree corresponds to “excessive left and right shaking”, etc. All the action language features are combined to obtain training suggestions.

[0063] After obtaining the evaluation results, the evaluation results are displayed to realize digitalization and visualization of practice, overcome the confusions encountered in the practice process of the students, improve the practice efficiency, help the students find errors and deficiencies in the practice process, and give reasonable suggestions, reduce the workload of teachers, improve the work efficiency, and improve the training quality. Specifically, a website with complete functions and easy to maintain is built by using Vue to make the front end and Django to make the network framework to display the evaluation results.

[0064] When the spoon overturning training is quantitatively evaluated, the operation process is as follows:

[0065] 1. Start the gyroscope, open the data monitoring software on the PC side, connect the gyroscope for recording data at the same time, fix the gyroscope behind the pot, and start training;

[0066] 2. The gyroscope synchronously transmits each group of data to the cloud platform through the monitoring software;

[0067] 3. The user logs in to the data processing system, logs in to his own account and password, selects the device number that needs to be scored, and sets the time period;

[0068] 4. The line chart of the data in this period is displayed, the score is calculated, and the score, evaluation and suggestion are displayed.

[0069] ​In summary, the embodiment utilizes a sensor monitoring method and deep learning technology to monitor various parameters in the practice process of the student, realizes the digitization and imaging of the practice, overcomes the confusion encountered in the practice process of the student, improves the practice efficiency, helps the student to find the errors and deficiencies in the practice process, and gives reasonable suggestions.

[0070] The above embodiments are intended to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, objects and effects of the present application, so as to make the public more thoroughly and comprehensively understand the disclosed content of the present application, and not to limit the protection scope of the present application.

[0071] The above embodiments are not exhaustive enumeration based on the present application, and there can be a plurality of other embodiments not listed. Any replacement and improvement made without violating the concept of the present application is within the protection scope of the present application.

Claims

1. A method for quantitatively evaluating a spoon training, characterized by, The method comprises the following steps: fixing the collection device on the wok and starting the wok training, wherein the collection device can collect multiple types of motion data in three-dimensional directions, and the multiple types of motion data include angular velocity, acceleration and angular acceleration collected in three-dimensional directions; obtaining multiple training data collected by the collection device, wherein one type of motion in one direction corresponds to one training data, and the training data includes angular velocity, acceleration and angular acceleration collected in three-dimensional directions; calculating the fitting degree between each training data and a full score curve, wherein the full score curve is obtained by using a variational autoencoder according to the motion data of the target person when wok training and expert evaluation indexes; obtaining an evaluation result according to all the fitting degrees, comprising: weighting and accumulating all the fitting degrees to obtain a training score; the expression of the full score curve is Y=Asin(Bx+Ct+D)+E, wherein A, B, C, D and E are parameters obtained by using a variational autoencoder, x is a data set composed of motion data, and t is a time sequence corresponding to the motion data.

2. The quantitative evaluation method of the spatula training according to claim 1, wherein, obtaining an evaluation result according to all the fitting degrees, further comprising: obtaining a corresponding motion language feature according to each fitting degree, and obtaining a training suggestion according to the motion language feature, wherein the motion language feature refers to a cooking motion problem described in natural language.

3. The quantitative evaluation method of the spatula training according to claim 1, wherein, After obtaining the training data, the training data is further filtered according to a selected time period, and a line chart is generated and displayed according to the filtered training data.

4. A quantitative evaluation device for training in tossing food, characterized in that, The method comprises the following steps: a training data module is configured to obtain multiple training data collected by a collection device, wherein one type of motion in one direction corresponds to one training data, the collection device can collect multiple types of motion data in three-dimensional directions, and the multiple types of motion data include angular velocity, acceleration and angular acceleration collected in three-dimensional directions, and the training data includes angular velocity, acceleration and angular acceleration collected in three-dimensional directions; a fitting degree calculation module is configured to calculate the fitting degree between each training data and a full score curve, wherein the full score curve is obtained by using a variational autoencoder according to the motion data of the target person when wok training and expert evaluation indexes; an evaluation result module is configured to obtain an evaluation result according to all the fitting degrees, comprising: weighting and accumulating all the fitting degrees to obtain a training score; the expression of the full score curve is Y=Asin(Bx+Ct+D)+E, wherein A, B, C, D and E are parameters obtained by using a variational autoencoder, x is a data set composed of motion data, and t is a time sequence corresponding to the motion data.

5. A spoon training quantitative evaluation system, characterized in that, The method comprises the following steps: a collection device is configured to collect multiple types of motion data in three-dimensional directions during wok training, and the multiple types of motion data include angular velocity, acceleration and angular acceleration collected in three-dimensional directions; a cloud platform is configured to receive and store the motion data collected by the collection device; a wok training quantitative evaluation device according to claim 4 is configured to analyze the motion data generated by wok training and give an evaluation result; a display model is configured to display the evaluation result.

Images