Juggling score generation method and system based on neural network

Abstract

The invention relates to the technical field of juggling scoring, in particular to a juggling score generation method and system based on a neural network. Based on action tracking data captured by camera equipment, a molecular dynamic analysis technology is adopted, and the deformation of a sphere and the change of friction force are estimated by applying a force field model; and analyzing the force borne by the ball and the mechanical reaction of the contact point of the athlete, and generating physical response data of the ball. According to the method, technologies such as molecular dynamic analysis, Shannon entropy calculation, deep learning anomaly detection, reverse engineering and time sequence analysis are introduced, so that the microscopic physical interaction of contact between the ball and an athlete can be deeply discussed; the complexity and unpredictability of juggling actions recorded by the capturing equipment are revealed through information entropy calculation, and a new quantitative index is provided for evaluating the diversity and innovation of juggling skills. The integration of the anomaly detection model enables real-time recognition and correction of actions or technical defects deviating from conventional skills.

Description

Technical Field

[0001] The present invention relates to the technical field of juggling scoring, and in particular to a juggling score generation method and system based on a neural network. Background Art

[0002] The field of juggling scoring technology involves the use of neural network algorithms to evaluate and generate juggling scores. This technology combines artificial intelligence, machine learning, and sports science, and aims to automate the scoring process by analyzing various parameters in juggling (such as juggling speed, stability, number of juggling, etc.). The use of neural networks enables this process to learn and simulate the evaluation criteria of professional coaches or judges for juggling skills, thereby providing an objective, accurate, and efficient scoring system. This technology has broad application prospects in sports training, athlete selection, skill assessment, and other fields.

[0003] The juggling score generation method is a technical method that uses neural networks to automatically evaluate juggling skills. By collecting relevant data from juggling activities (for example, capturing the characteristics of juggling movements through video analysis, sensors, etc.), and inputting these data into the trained neural network model, juggling scores or ratings are generated. The purpose is to provide an accurate and objective evaluation standard, reduce the subjectivity and inconsistency of manual scoring, thereby helping coaches better understand the skill level of athletes and provide customized training suggestions for athletes. Through this method, it is intended to improve the efficiency and fairness of juggling skill training and evaluation, while promoting the rapid development and improvement of athlete technology.

[0004] Although the existing technology has realized the evaluation of juggling skills through neural network algorithms and improved the automation and objectivity of scoring, in terms of simulation and analysis of microscopic physical processes, the existing technology has not fully considered the complex physical reactions when the ball contacts the athlete, such as detailed changes in deformation and friction. This limits the ability of the scoring system to deeply understand the physical accuracy and efficiency of juggling skills. In addition, in the quantitative evaluation of movement complexity and unpredictability, there is a lack of in-depth exploration of the diversity and innovation of juggling skills, making it difficult to provide a comprehensive evaluation of skill complexity and innovation. For the real-time identification and correction of abnormal movements or deviations in skills, the existing technology has failed to effectively integrate efficient anomaly detection algorithms, resulting in blind spots in the timely identification and improvement of technical shortcomings. In the optimization of scoring strategies, there is a lack of reverse engineering methods to infer key evaluation factors from ideal skill performance, which affects the accuracy and adaptability of the scoring criteria. For the identification and analysis of juggling rhythm patterns, it is difficult for the existing technology to make in-depth use of time series analysis technology, which limits the improvement of the accuracy of skill level evaluation and the refinement of athlete technical guidance. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method and system for generating juggling scores based on a neural network.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for generating juggling scores based on a neural network, comprising the following steps: S1: Based on the motion tracking data captured by the camera, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change of friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball; S2: Based on the physical response data of the ball, the complexity of the juggling action is evaluated by using the Shannon entropy calculation method, and the information entropy value in the juggling process is calculated to analyze the randomness of the action sequence and generate the entropy value information of the juggling action; S3: Based on the entropy value information of the juggling action, an abnormality detection model based on deep learning is used to analyze the juggling action again, and abnormal actions or deviations that deviate from the normal pattern are identified. The abnormal actions are located and their deviation characteristics are analyzed through neural network and pattern recognition technology, and abnormal action identification and analysis results are generated; S4: Based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network models are used to analyze the feature data of the abnormal action, and key image features and data indicators are inferred from the scoring target of the juggling skill, and image processing and feature extraction are performed to generate a key evaluation factor model; S5: Based on the physical response data of the ball and the entropy value information of the juggling action, a time series analysis method is used to analyze the time interval and duration of each ball contact during the juggling process, a statistical analysis technique is used to extract characteristic parameters of the juggling rhythm, and a Fourier transform method is applied to identify and analyze the juggling rhythm pattern to generate a juggling rhythm characteristic parameter set; S6: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average and standardization processing methods are used to evaluate the juggling skills, and dynamic skill tracking and feedback optimization are performed to generate a comprehensive juggling skill scoring result; S7: Based on the comprehensive scoring results of the juggling skills, use decision tree and rule engine technology to analyze the player's skills and formulate training plans, and track the results through the training plan to generate juggling results and technical improvement feedback results.

[0007] As a further solution of the present invention, the physical response data of the ball includes the elastic coefficient, deformation recovery time and deformation degree of the ball, the entropy value information of the juggling action includes the average entropy value, maximum entropy value and entropy value variation coefficient of the action sequence, the abnormal action identification and analysis results include the frequency of occurrence of abnormal actions, the type of abnormal actions and the degree of influence of abnormal actions on the juggling rhythm, the key evaluation factor model includes the spatial accuracy, time synchronization and movement fluency indicators of the juggling action, the juggling rhythm characteristic parameter set includes the basic frequency of the rhythm, the width of the frequency distribution and the characteristic value of the key rhythm pattern, the comprehensive scoring result of the juggling skill includes the player's skill level, skill improvement potential index and skill balance evaluation, and the juggling results and technical improvement feedback results include skill improvement plans, skill mastery improvement intervals and priority technical improvement points.

[0008] As a further solution of the present invention, based on the motion tracking data captured by the camera device, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change of friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball. The specific steps are: S101: Based on the motion tracking data captured by the camera, molecular dynamic simulation is performed, and numerical simulation operations are performed using LAMMPS software. The behavior of the metal sphere is simulated by defining the EAM force field model, setting the time step to 1fs, adjusting the friction coefficient parameter to 0.3, and the elastic modulus parameter to 200 GPa, to generate contact interaction simulation data; S102: Based on the contact interaction simulation data, finite element analysis is performed, and static structural analysis is performed using ANSYS software. Through tetrahedral meshing operations, the mesh size is set to 0.5 mm, a predefined contact force is applied to the top surface of the sphere, and the bottom surface of the sphere is fixed as a boundary condition to generate mechanical response analysis data; S103: Based on the mechanical response analysis data, data visualization and analysis are performed, and the stress-strain curve is drawn using the Matplotlib library in Python. The Pandas library performs data screening and processing operations, and the X-axis of the chart is set to strain, and the Y-axis is set to stress. Colors are used to identify various contact points to generate physical response data of the sphere.

[0009] As a further solution of the present invention, based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity of the juggling action, and the information entropy value in the juggling process is calculated to analyze the randomness of the action sequence. The specific steps of generating the entropy value information of the juggling action are as follows: S201: Based on the physical response data of the ball, using the pandas library in the Python environment, time serialization processing is performed on the contact time point and force information, including converting the original data into a sequence indexed by a timestamp, defining a unique time point and force value for each ball juggling contact, and generating time serialized ball juggling data; S202: Based on the time-series juggling data, the entropy function in the SciPy library is applied to calculate the information entropy of the juggling action recorded by the capture device, and the Shannon entropy formula is applied by counting the frequency of each unique action state in the sequence. Calculate the information entropy of the entire sequence, where Indicates action The probability of occurrence is used to quantify the irregularity and randomness of the action sequence and generate the information entropy measurement results of the juggling action; S203: Based on the juggling action information entropy measurement result, use the Matplotlib library to graphically display the entropy value change trend, including mapping the time series data and its corresponding entropy value in the same chart, displaying the entropy value changes of the juggling action at various time points through graphics, identifying the change rules of the action sequence in the juggling process, and generating the entropy value information of the juggling action.

[0010] As a further solution of the present invention, based on the entropy information of the juggling action, an abnormality detection model based on deep learning is used to analyze the juggling action again, identify abnormal actions or deviations that deviate from the normal pattern, locate the abnormal action and analyze its deviation characteristics through neural network and pattern recognition technology, and generate abnormal action identification and analysis results in the following specific steps: S301: Based on the entropy information of the juggling action, a deep learning model is constructed using TensorFlow and Keras libraries, including using an LSTM layer to process time series data to capture the temporal dependency of the juggling action, configuring 256 hidden units in the LSTM layer to process the juggling action data, adding a Dropout layer with a ratio of 0.5 to reduce the risk of overfitting, performing action feature dimension conversion through a fully connected layer, identifying potential abnormal action patterns, and generating an abnormal action candidate set; S302: Based on the abnormal action candidate set, the abnormal score threshold method and IsolationForest algorithm are used, and the number of isolation trees is set to 100. The abnormal score of each action is calculated. The abnormal score is calculated based on the degree of deviation between the probability distribution predicted by the model and the probability distribution of the conventional action. By comparing with the preset threshold, the actions with scores higher than the threshold are screened and identified as abnormal actions, and the selected abnormal action set is generated; S303: Based on the selected abnormal action set, a convolutional neural network is used to perform feature analysis. The convolutional neural network model configuration includes a convolution layer and a pooling layer. The convolution layer uses a 3x3 filter with a step size of 1. The pooling layer uses a 2x2 pooling operation to extract the spatial features of the abnormal actions, the speed of the actions, and the changes in the intensity. The position and nature of the abnormal actions are located through the features, and the abnormal action identification and analysis results are generated.

[0011] As a further solution of the present invention, based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network model are used to analyze the feature data of abnormal actions, and key image features and data indicators are inferred from the scoring target of juggling skills, and image processing and feature extraction are performed to generate the key evaluation factor model. The specific steps are: S401: Based on the abnormal action identification and analysis results, a convolutional neural network model is constructed in a Python environment using TensorFlow and Keras libraries, including three convolutional layers Conv2D for image feature extraction. Each convolution layer is followed by a MaxPooling2D layer to reduce the dimension of the feature map. The convolutional layers are configured with 32, 64 and 128 filters respectively, and the filter size is set to 3x3. The ReLU activation function is used to introduce nonlinearity to capture the key features of the abnormal action image and generate a preliminary feature data set. S402: Based on the preliminary feature data set, iteratively optimize the network structure, add two batch normalization layers BatchNormalization, introduce two fully connected layers Dense to learn and classify deep features, the first fully connected layer is configured with 256 nodes and uses ReLU activation function, the second fully connected layer is used as the output layer, the number of nodes corresponds to the number of categories of juggling skill scores, and the softmax activation function is used for classification decision, the evaluation factors of juggling skills are learned and identified from the image features, and an optimized feature learning model is generated; S403: Based on the optimized feature learning model, the model is trained and parameter optimized. During the training process, the batch size is set to 32, the number of iterations is set to 100, the Adam optimizer is used to adjust the learning rate to 0.001, and the loss function uses categorical_crossentropy to perform multi-classification loss calculation. By training and optimizing the model parameters, the model can extract key features from abnormal action images, reflect the scoring indicators of juggling skills, and generate a key evaluation factor model.

[0012] As a further solution of the present invention, based on the physical response data of the ball and the entropy value information of the juggling action, a time series analysis method is used to analyze the time interval and duration of each ball contact during the juggling process, a statistical analysis technique is used to extract characteristic parameters of the juggling rhythm, and a Fourier transform method is applied to identify and analyze the juggling rhythm pattern. The specific steps of generating a juggling rhythm characteristic parameter set are as follows: S501: Based on the physical response data of the ball and the entropy value information of the juggling action, the pandas library in the Python environment is used to perform processing operations on the time point data, and the to_datetime function and the diff method are used to perform format conversion and time interval calculation, so as to provide a timestamp for each contact in the juggling process and generate processed time series data; S502: Based on the processed time series data, the numpy library is used to perform basic statistical analysis of the juggling rhythm, and the mean, standard deviation, minimum and maximum values ​​of the time interval are calculated through the mean, std, min and max functions to provide a quantitative feature description for the juggling rhythm and generate a basic statistical parameter set of the juggling rhythm; S503: Based on the basic statistical parameter set of the juggling rhythm, Fourier transform is applied to analyze the frequency characteristics of the juggling rhythm, and the fft function in the numpy library is used to convert the time series data from the time domain to the frequency domain, and the key rhythm patterns of the juggling action are identified. By acquiring and analyzing the rhythm change rules during the juggling process, a juggling rhythm characteristic parameter set is generated.

[0013] As a further solution of the present invention, based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average and standardization processing methods are applied to evaluate the juggling skills, and dynamic skill tracking and feedback optimization are performed to generate the juggling skills comprehensive scoring results in the following specific steps: S601: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average of multiple indicators is calculated using the numpy and pandas libraries in the Python environment, including assigning a weight to each indicator based on its influence on the juggling skill, the abnormal action identification is assigned a weight according to the influence on the skill performance, and the numpy.average method is used to input each indicator data and its corresponding weight, calculate the weighted average, and generate a weighted average performance score; S602: Based on the weighted average performance score, use the StandardScaler function of the sklearn library to convert it into a distribution with zero mean and unit variance, and pass the weighted average performance score data as an input value, calculate and apply the mean and standard deviation, standardize the scoring data, and generate a standardized scoring result; S603: Based on the standardized scoring results, regularly analyze the juggling skill scores, and guide the continuous optimization of the training plan by comparing the current scores with historical scoring data to generate a comprehensive juggling skill scoring result.

[0014] As a further solution of the present invention, based on the comprehensive scoring results of the juggling skills, the decision tree and rule engine technology are used to analyze the player's skills and formulate a training plan, and the effectiveness is tracked through the training plan to generate juggling scores and technical improvement feedback results. The specific steps are: S701: Based on the comprehensive scoring result of the dribbling skill, a decision tree algorithm is used, the input features include ball possession time, number of dribbling times and error rate, the splitting standard is set to information gain, the data is trained through the fit method, model training and analysis are performed to identify skill weaknesses, and a player skill weakness analysis result is generated; S702: Based on the analysis results of the player's skill weaknesses, a training plan is formulated using rule engine technology, and a Drools rule engine is used to define training rules based on skill weaknesses. The rule engine automatically matches the training plan according to the analysis results to generate a dedicated training plan; S703: Based on the exclusive training program, the effectiveness is tracked through the training plan, and the decision tree analysis is used to compare the skill score differences before and after the training. By tracking the results and formulating a technology improvement plan, the juggling results and technology improvement feedback results are generated.

[0015] As a further solution of the present invention, a juggling score generation system based on a neural network includes a contact mechanics analysis module, an entropy value analysis module, an abnormal action recognition module, a feature extraction module, a rhythm analysis module, a skill scoring module, a training program formulation module, and a feedback optimization module; The contact mechanics analysis module uses molecular dynamic analysis technology based on the motion tracking data captured by the camera device, defines a force field model and uses the molecular dynamics simulation software LAMMPS to perform numerical simulation and adjust the friction coefficient parameters and elastic modulus parameters to generate physical response data of the sphere; The entropy value analysis module uses the Shannon entropy calculation method based on the physical response data of the ball, performs time series processing on the data through the pandas library in the Python environment, and uses the entropy function in the SciPy library to calculate the information entropy to generate the entropy value information of the juggling action; The abnormal action recognition module is based on the entropy information of the juggling action, adopts an abnormal detection model based on deep learning, uses the long short-term memory network and isolation forest algorithm built by TensorFlow and Keras libraries, processes time series data, identifies abnormal actions, and generates abnormal action identification and analysis results; The feature extraction module uses reverse engineering technology and convolutional neural network models based on abnormal action identification and analysis results, uses TensorFlow and Keras libraries to build models, including multi-layer convolution and pooling layers to extract image features, iteratively optimizes the network structure, and generates a key evaluation factor model; The rhythm analysis module uses a time series analysis method based on the physical response data of the ball and the entropy value information of the juggling action, performs basic statistical analysis and Fourier transform through the numpy library in the Python environment, identifies the characteristics of the juggling rhythm, and generates a juggling rhythm characteristic parameter set; The skill scoring module applies weighted average and standardization processing methods based on abnormal action identification and analysis results, key evaluation factor model and juggling rhythm characteristic parameter set, calculates the weighted average of multiple indicators through numpy and pandas libraries and performs standardization processing to generate a comprehensive juggling skill scoring result; The training program formulation module analyzes the player's skills based on the comprehensive scoring results of the juggling skills, adopts a decision tree algorithm, uses the DecisionTreeClassifier class in the Scikit-learn library, performs model training and analysis to identify skill weaknesses, and adopts rule engine technology to formulate a training program to generate a dedicated training program; The feedback optimization module is based on a dedicated training program, tracks the results through the training plan, adopts decision tree analysis, uses the DecisionTreeClassifier class in the Scikit-learn library, compares the skill score differences before and after training, and generates juggling results and technical improvement feedback results.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by introducing technologies such as molecular dynamic analysis, Shannon entropy calculation, anomaly detection of deep learning, reverse engineering and time series analysis, it is not only possible to deeply explore the microscopic physical interaction between the ball and the athlete's contact, but also to reveal the complexity and unpredictability of the juggling action recorded by the capture device through information entropy calculation, which provides a new quantitative index for evaluating the diversity and innovation of juggling skills. The integration of anomaly detection models enables real-time identification and correction of actions or technical defects that deviate from the norm in skills, thereby realizing timely discovery and targeted improvement of athletes' technical shortcomings. Through reverse engineering and convolutional neural network models, key image features and data indicators are inferred from scoring targets, greatly improving the accuracy and adaptability of scoring strategies. In addition, the juggling rhythm pattern is identified and analyzed using time series analysis technology, further refining the accuracy of skill level assessment, and providing athletes with specific technical guidance. Therefore, the present invention not only significantly improves the objectivity, accuracy and efficiency of juggling skill evaluation, but also provides more scientific and personalized training suggestions for athletes' technical improvement, promoting the rapid development and improvement of juggling skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a method for generating juggling scores based on a neural network is proposed for the present invention; Figure 2 A schematic diagram of a detailed flow chart of step S1 in the method for generating juggling scores based on a neural network proposed in the present invention; Figure 3 The present invention provides a detailed flowchart of step S2 in the method for generating juggling scores based on a neural network; Figure 4 The present invention provides a detailed flowchart of step S3 in the method for generating juggling scores based on a neural network; Figure 5 The present invention provides a detailed flowchart of step S4 in the method for generating juggling scores based on a neural network; Figure 6 The present invention provides a detailed flowchart of step S5 in the method for generating juggling scores based on a neural network; Figure 7 The present invention provides a detailed flowchart of step S6 in the method for generating juggling scores based on a neural network; Figure 8 The present invention provides a detailed flowchart of step S7 in the method for generating juggling scores based on a neural network; Fig. 9 The present invention proposes a module diagram of a juggling score generation system based on a neural network. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0020] Embodiment 1 See also Figure 1 In order to achieve the above purpose, the present invention adopts the following technical solution: A method for generating juggling scores based on a neural network comprises the following steps: S1: Based on the motion tracking data captured by the camera, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change of friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball; S2: Based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity of the juggling action. By calculating the information entropy value in the juggling process, the randomness of the action sequence is analyzed and the entropy value information of the juggling action is generated; S3: Based on the entropy information of the juggling action, the deep learning-based anomaly detection model is used to analyze the juggling action again, identify abnormal actions or deviations from the normal pattern, locate the abnormal actions and analyze their deviation characteristics through neural network and pattern recognition technology, and generate abnormal action identification and analysis results; S4: Based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network models are used to analyze the feature data of abnormal actions, infer key image features and data indicators from the scoring targets of juggling skills, perform image processing and feature extraction, and generate a key evaluation factor model; S5: Based on the physical response data of the ball and the entropy value information of the juggling action, the time series analysis method is used to analyze the time interval and duration of each ball touch during the juggling process, the statistical analysis technology is used to extract the characteristic parameters of the juggling rhythm, and the Fourier transform method is applied to identify and analyze the juggling rhythm pattern to generate the juggling rhythm characteristic parameter set; S6: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average and standardization processing methods are used to evaluate the juggling skills, and dynamic skill tracking and feedback optimization are performed to generate a comprehensive juggling skill score result; S7: Based on the comprehensive scoring results of juggling skills, decision tree and rule engine technology are used to analyze player skills and formulate training plans. The results are tracked through training plans to generate juggling scores and technical improvement feedback results.

[0021] The physical response data of the ball include the elastic coefficient, deformation recovery time and deformation degree of the ball. The entropy value information of the juggling action includes the average entropy value, maximum entropy value and entropy variation coefficient of the action sequence. The abnormal action identification and analysis results include the frequency of abnormal actions, the type of abnormal actions and the degree of influence of abnormal actions on the juggling rhythm. The key evaluation factor model includes the spatial accuracy, time synchronization and action fluency indicators of the juggling action. The juggling rhythm characteristic parameter set includes the basic frequency of the rhythm, the width of the frequency distribution and the characteristic value of the key rhythm pattern. The comprehensive scoring results of the juggling skills include the player's skill level, skill improvement potential index and skill balance evaluation. The juggling results and technical improvement feedback results include skill improvement plans, skill mastery improvement ranges and priority technical improvement points.

[0022] In step S1, the motion tracking data captured by the camera device is used to analyze the microscopic physical interaction between the ball and the athlete when they come into contact. This includes the application of a force field model, where the model is configured to simulate the deformation of the ball and the change in friction. By simulating the scene of contact between the ball and the athlete in a specific physical environment, the algorithm calculates the force and corresponding mechanical response of the ball at different contact points based on experimental data. The physical response data of the ball is generated through simulation and analysis. This data provides a microscopic perspective for understanding the behavior of the ball, and also lays the foundation for the subsequent analysis of juggling skills.

[0023] In step S2, based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity and unpredictability of the juggling action recorded by the capture device. This step analyzes the action sequence during the juggling process and quantifies the randomness and irregularity of the action sequence by calculating the information entropy value. Including statistical analysis of the time interval, force and direction factors of each touch of the ball, using the Shannon entropy formula in information theory to quantify the information content of the action sequence, it can not only reveal the patterns and variations in juggling skills, but also generate juggling action entropy information, providing a basis for subsequent abnormal action detection.

[0024] In step S3, based on the entropy information of the juggling action, a deep learning anomaly detection model is used to analyze the juggling action, identify abnormal actions that deviate from the normal pattern, and use neural network and pattern recognition technology to learn the normal pattern of juggling action through training models, and analyze new action sequences to identify abnormal actions. This process refers to the multi-dimensional features of juggling actions, including the speed, rhythm, and accuracy of the action. The neural network model is used to learn and match the features to identify actions that are different from the normal pattern in the training set. Finally, abnormal action identification and analysis results are generated, which not only provide an in-depth understanding of the potential problems in juggling skills, but also provide a basis for athletes and coaches to formulate targeted training plans and technical improvement measures.

[0025] In step S4, the feature data of abnormal actions are deeply analyzed through reverse engineering technology and convolutional neural network models. This stage first involves a detailed interpretation of the abnormal action identification and analysis results, and the convolutional neural network model is used to extract and learn features from the data. The model is trained to identify key image features and data indicators in the juggling skill scoring target, including multi-level image processing and feature extraction operations. The convolution layer is used to automatically extract useful features from the image of the juggling action, and the pooling layer further processes the features to reduce the amount of calculation and retain key information. Finally, the features are converted into juggling skill scores through the fully connected layer. The key features based on the score can be directly inferred from the image of the juggling action. The generated key evaluation factor model not only provides a powerful tool for the objective evaluation of juggling skills, but also provides athletes with a direction to improve their skills.

[0026] In step S5, based on the physical response data of the ball and the entropy value information of the juggling action, the time series analysis method is used to deeply analyze the time interval and duration of each touch of the ball during the juggling process. The characteristic parameters of the juggling rhythm are extracted from the time series data through statistical analysis technology. Then, the Fourier transform method is applied to analyze the time series to identify the basic pattern and potential variation of the juggling rhythm. Fourier transform can convert the time series data from the time domain to the frequency domain, so that not only the rhythm characteristics of the juggling action can be clearly identified, but also its periodic and non-periodic components can be analyzed. The generated juggling rhythm characteristic parameter set provides a quantitative perspective for understanding the rhythm control of the athlete's juggling skills. This parameter can not only reflect the stability of the athlete's juggling rhythm, but also provide guidance for training and optimize the rhythm control in juggling skills.

[0027] In step S6, the juggling skills are comprehensively evaluated by using the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, and the weighted average and standardized processing methods are used. In this process, each data type is assigned a different weight to reflect its importance in the juggling skill evaluation. The abnormal action identification and analysis results will receive a higher weight because it is directly related to the accuracy and technical flaws of the skill. The various types of data are combined by the weighted average method, and then standardized processing is used to eliminate the influence of different scales to ensure the fairness and comparability of the scoring results. The final generated comprehensive juggling skill scoring result is a quantitative summary of the various characteristics of the athlete's juggling skills. The scoring result not only reflects the athlete's current skill level, but also provides a basis for identifying skill progress and regression, supports dynamic skill tracking and feedback optimization, and provides athletes and coaches with clear improvement directions and training priorities.

[0028] In step S7, based on the comprehensive scoring results of the juggling skills, the decision tree and rule engine technology are used to conduct an in-depth analysis of the players' skills, and a dedicated training plan is formulated accordingly. The decision tree is used to identify different levels of skill levels and specific skill defects from the comprehensive scoring results, while the rule engine automatically generates targeted training suggestions and plans based on this analysis result. In this process, the rule engine proposes a training plan to improve technical flaws and enhance skill performance based on the preset logic and rules, referring to the athlete's needs and goals, as well as the details of his skill score. Through the execution of the training plan and tracking the results, the system can dynamically adjust the suggestions to better adapt to the athlete's progress speed and training response, and finally generate juggling results and technical improvement feedback results. This result not only provides athletes with tailor-made training plans, but also helps athletes improve their skills systematically through continuous tracking and feedback.

[0029] See also Figure 2 Based on the motion tracking data captured by the camera, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change of friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball. The specific steps are: S101: Based on the motion tracking data captured by the camera, molecular dynamic simulation is performed, and numerical simulation operations are performed using LAMMPS software. The behavior of the metal sphere is simulated by defining the EAM force field model, setting the time step to 1fs, adjusting the friction coefficient parameter to 0.3, and the elastic modulus parameter to 200 GPa, to generate contact interaction simulation data; S102: Based on the contact interaction simulation data, finite element analysis is performed, and static structural analysis is performed using ANSYS software. Through tetrahedral meshing operations, the mesh size is set to 0.5 mm, a predefined contact force is applied to the top surface of the sphere, and the bottom surface of the sphere is fixed as a boundary condition to generate mechanical response analysis data; S103: Based on the mechanical response analysis data, data visualization and analysis are performed. The stress-strain curve is drawn using the Matplotlib library in Python. The Pandas library performs data screening and processing operations. The X-axis of the chart is set to strain, the Y-axis is set to stress, and colors are used to identify various contact points to generate the physical response data of the sphere.

[0030] In substep S101, the behavior of the metal sphere under contact mechanics experimental conditions is explored in detail through molecular dynamics simulation. The numerical simulation operation is performed using the Large Scale Atomic / Molecular Parallel Simulator (LAMMPS). First, the embedded atom method (EAM) force field model is defined to simulate the microscopic behavior of the metal sphere. During the simulation, the time step is set to 1 femtosecond (fs) to ensure the accuracy of the simulation, the friction coefficient parameter is adjusted to 0.3 to reflect the friction characteristics of the material, and the elastic modulus parameter is set to 200 gigapascals (GPa) to simulate the elastic response of the metal. The fine adjustment of the parameters is based on the physical properties of the material and the experimental conditions, aiming to generate contact interaction simulation data, which accurately reflects the behavior of the metal sphere under specific mechanical conditions and provides a basis for further mechanical response analysis.

[0031] In substep S102, the mechanical response of the metal sphere is analyzed in depth by finite element analysis. Static structural analysis is performed using ANSYS software. First, a tetrahedral meshing operation is performed, and the mesh size is set to 0.5 millimeters (mm) to obtain sufficient analysis accuracy. During the simulation, a predefined contact force is applied to the top surface of the sphere, and the bottom surface of the sphere is fixed as a boundary condition. Finally, mechanical response analysis data is generated, which provides detailed stress and deformation information of the sphere when it is subjected to force, providing an important perspective for understanding the behavior of the material under experimental conditions.

[0032] In sub-step S103, the physical response of the sphere is presented through data visualization and analysis. The Matplotlib library in the Python programming language is used to draw the stress-strain curve, and the Pandas library performs data screening and processing operations. In the chart, the X-axis is set to strain and the Y-axis is stress. Different colors are used to represent the data of various contact points to intuitively display the stress response of the material under different strain conditions. It not only intuitively displays the mechanical properties of the material, but also facilitates the analysis and comparison of the stress-strain relationship under different contact points. The generated sphere physical response data provides an intuitive way to understand and analyze the behavior of the metal sphere in the contact mechanics experiment in the form of charts, providing important information for the research and application of the mechanical properties of the material.

[0033] See also Figure 3 Based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity of the juggling action. By calculating the information entropy value in the juggling process and analyzing the randomness of the action sequence, the specific steps for generating the entropy value information of the juggling action are as follows: S201: Based on the physical response data of the ball, the pandas library in the Python environment is used to perform time serialization processing on the contact time point and force information, including converting the original data into a sequence indexed by timestamps, defining a unique time point and force value for each ball juggling contact, and generating time serialized ball juggling data; S202: Based on the time-series juggling data, the entropy function in the SciPy library is used to calculate the information entropy of the juggling action recorded by the capture device. By counting the frequency of each unique action state in the sequence, the Shannon entropy formula is applied. Calculate the information entropy of the entire sequence, where Indicates action The probability of occurrence is used to quantify the irregularity and randomness of the action sequence and generate the information entropy measurement results of the juggling action; S203: Based on the information entropy measurement results of the juggling action, the Matplotlib library is used to graphically display the entropy value change trend, including mapping the time series data and its corresponding entropy value in the same chart, displaying the entropy value changes of the juggling action at various time points through graphics, identifying the change rules of the action sequence in the juggling process, and generating the entropy value information of the juggling action.

[0034] In sub-step S201, the physical response data of the ball is processed in a time series through the pandas library in the Python environment. This includes converting the motion tracking data captured by the original camera device into a sequence indexed by timestamps, ensuring that each juggling contact event is assigned a unique time point and corresponding force value. The contact time point and force information are read, and the time series function of pandas is used to convert the occurrence time of each contact event into a timestamp index. At the same time, the corresponding force value is recorded for each event. The time series juggling data finally generated provides a basis for analyzing the complexity and unpredictability of juggling skills.

[0035] In sub-step S202, based on the time-serialized juggling data, the entropy function in the SciPy library is used to calculate the information entropy of the juggling action recorded by the capture device. This step uses the Shannon entropy formula to quantify the information entropy of the entire action sequence by counting the frequency of each unique action state in the sequence. It includes identifying and counting the frequency of each unique action in the time-serialized data, applying the information entropy calculation formula, where the probability of each action occurring is used as input, and the calculated information entropy value reflects the irregularity and randomness of the juggling action recorded by the capture device. Through this process, not only can the complexity of the juggling action be quantified, but it also provides a method for evaluating the changes in juggling skills. The entropy value results of the juggling action recorded by the capture device are helpful to deeply understand the unpredictability of the juggling action.

[0036] In sub-step S203, based on the measurement results of the information entropy of the juggling action, the Matplotlib library is used to graphically display the entropy change trend. This includes mapping the time series data and its corresponding entropy value to the same chart, and visually displaying the entropy value changes of the juggling action recorded by the capture device at different time points through visual means. Such graphical processing is not only convenient for identifying the change rules of the action sequence during the juggling process, but also makes the analysis of the juggling skills more intuitive and easy to understand. By setting the X-axis of the chart to the time series and the Y-axis to the entropy value, and using different colors or marks to distinguish different contact points, the generated juggling action entropy value information chart becomes a powerful tool for evaluating and displaying the changes in juggling skills, and provides visual feedback for training and skill improvement.

[0037] See also Figure 4 Based on the entropy information of the juggling action, the deep learning-based anomaly detection model is used to analyze the juggling action again, identify abnormal actions or deviations from the normal pattern, and locate the abnormal actions and analyze their deviation characteristics through neural network processing and pattern recognition technology. The specific steps of generating abnormal action identification and analysis results are as follows: S301: Based on the entropy information of juggling actions, a deep learning model is constructed using TensorFlow and Keras libraries, including using the LSTM layer to process time series data and capture the temporal dependency of juggling actions. The LSTM layer is configured with 256 hidden units to process juggling action data. A Dropout layer is added with a ratio of 0.5 to reduce the risk of overfitting. The action feature dimension is transformed through a fully connected layer to identify potential abnormal action patterns and generate abnormal action candidate sets. S302: Based on the abnormal action candidate set, the abnormal score threshold method and IsolationForest algorithm are used, and the number of isolation trees is set to 100. The abnormal score of each action is calculated. The abnormal score calculation is based on the deviation degree between the probability distribution predicted by the model and the probability distribution of the conventional action. By comparing with the preset threshold, the actions with scores higher than the threshold are screened and identified as abnormal actions, and the selected abnormal action set is generated; S303: Based on the selected abnormal action set, a convolutional neural network is used to perform feature analysis. The convolutional neural network model configuration includes a convolution layer and a pooling layer. The convolution layer uses a 3x3 filter with a step size of 1. The pooling layer uses a 2x2 pooling operation to extract the spatial features of the abnormal actions, the speed of the actions, and the changes in the intensity. The position and nature of the abnormal actions are located through the features, and the abnormal action identification and analysis results are generated.

[0038] In sub-step S301, a deep learning model is built using TensorFlow and Keras libraries, especially using a long short-term memory network (LSTM) layer to process the juggling action data. The LSTM layer is configured with 256 hidden units to capture the temporal dependencies of the juggling action. This configuration allows the model to understand and remember the long-term dependencies in the action sequence, which is particularly critical for the time series data of the juggling action. In order to reduce the risk of overfitting, a Dropout layer is added with a ratio set to 0.5, which can randomly ignore a part of the network connections during the training process and enhance the generalization ability of the model. The dimension of the action features is transformed by the fully connected layer. This step aims to transform the complex time series features captured by the LSTM layer into a form more suitable for action recognition, thereby identifying potential abnormal action patterns. This process generates a candidate set of abnormal actions, which provides a basis for further abnormal action recognition, and also provides a preliminary screening of potential abnormal patterns in the action sequence.

[0039] In sub-step S302, based on the abnormal action candidate set, the IsolationForest algorithm and the abnormal score threshold method are used to identify abnormal actions. The number of isolation trees is set to 100. Multiple isolation trees are constructed to isolate each action and calculate the abnormal score of each action. The abnormal score is calculated based on the degree of deviation between the probability distribution predicted by the model and the probability distribution of conventional actions. This approach allows the degree of abnormality of each action to be accurately quantified. By comparing with a preset threshold, actions with scores higher than the threshold are screened and identified as abnormal actions. This step generates a selected abnormal action set, which provides a precise target for in-depth analysis of the characteristics and properties of abnormal actions, and also provides a specific basis for training and improving juggling skills.

[0040] In sub-step S303, based on the selected abnormal action set, a convolutional neural network (CNN) is used for feature analysis to further extract and analyze the spatial features of abnormal actions. The convolution layer uses a 3x3 filter and a step size of 1 to capture subtle changes in the action, and the pooling layer uses a 2x2 pooling operation to reduce the feature dimension while retaining important information. It helps to extract key spatial features of abnormal actions, such as changes in speed and strength of the action, locate the specific location and nature of the abnormal action, and generate abnormal action identification and analysis results.

[0041] See also Figure 5 Based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network model are used to analyze the characteristic data of abnormal actions, and the key image features and data indicators are inferred from the scoring target of juggling skills. Image processing and feature extraction are performed to generate the key evaluation factor model. The specific steps are as follows: S401: Based on the abnormal action identification and analysis results, a convolutional neural network model is constructed in the Python environment using TensorFlow and Keras libraries, including three convolutional layers Conv2D for image feature extraction. Each convolution layer is followed by a MaxPooling2D layer to reduce the dimension of the feature map. The convolutional layers are configured with 32, 64 and 128 filters respectively, and the filter size is set to 3x3. The ReLU activation function is used to introduce nonlinearity to capture the key features of abnormal action images and generate a preliminary feature dataset. S402: Based on the preliminary feature data set, iteratively optimize the network structure, add two batch normalization layers, introduce two fully connected layers Dense to learn and classify deep features, the first fully connected layer is configured with 256 nodes and uses the ReLU activation function, the second fully connected layer is used as the output layer, the number of nodes corresponds to the number of categories of the juggling skill score, and the softmax activation function is used for classification decision, learn and identify the evaluation factors of juggling skills from the image features, and generate an optimized feature learning model; S403: Based on the optimized feature learning model, the model is trained and parameter optimized. During the training process, the batch size is set to 32, the number of iterations is set to 100, the Adam optimizer is used to adjust the learning rate to 0.001, and the loss function uses categorical_crossentropy to perform multi-classification loss calculations. By training and optimizing the model parameters, the model can extract key features from abnormal action images, reflect the scoring indicators of juggling skills, and generate a key evaluation factor model.

[0042] In sub-step S401, a convolutional neural network model is built in Python environment through TensorFlow and Keras libraries, focusing on image feature extraction. The model design includes three layers of convolutional layers Conv2D, each followed by a MaxPooling2D layer to reduce the dimension of the feature map and optimize computational efficiency. The convolutional layers are configured with 32, 64 and 128 filters, respectively, with a filter size of 3x3, and the ReLU activation function is used to introduce nonlinearity, with the purpose of capturing key features in abnormal action images. This series of configurations enables the network to effectively extract the spatial features of the juggling action images and generate a preliminary feature dataset. The dataset generated in this step provides a basis for in-depth analysis and identification of juggling skill evaluation factors, and provides rich information for subsequent feature learning and classification by refining and optimizing the feature representation of the image.

[0043] In sub-step S402, based on the preliminary feature data set, the model structure is improved through iterative optimization. Two batch normalization layers BatchNormalization are introduced to improve the training stability of the model, while two fully connected layers Dense are used to deeply learn and classify deep features. The first fully connected layer is configured with 256 nodes and uses the ReLU activation function, aiming to further process and transform feature data and enhance the nonlinear expression ability of the model. The second fully connected layer serves as the output layer. The number of nodes corresponds to the number of categories of juggling skill scores, and the softmax activation function is used for the final classification decision. This step optimizes the feature learning model so that it can more accurately learn and identify the evaluation factors of juggling skills from image features, and generates an optimized feature learning model, which provides an effective tool for accurately evaluating juggling skills.

[0044] In sub-step S403, model training and parameter optimization are performed based on the optimized feature learning model. The batch size is set to 32, the number of iterations is set to 100, and the learning rate is adjusted to 0.001 using the Adam optimizer to balance the training speed and convergence performance. It is suitable for multi-classification problems and ensures the accuracy of loss calculation during model training. This training process optimizes the model parameters, enabling the model to effectively extract key features from abnormal action images and accurately reflect the scoring indicators of juggling skills, generating a key evaluation factor model. See also Figure 6 Based on the physical response data of the ball and the entropy value information of the juggling action, the time series analysis method is used to analyze the time interval and duration of each ball touch during the juggling process. The characteristic parameters of the juggling rhythm are extracted using statistical analysis technology. The Fourier transform method is applied to identify and analyze the juggling rhythm pattern. The specific steps for generating the juggling rhythm characteristic parameter set are as follows: S501: Based on the physical response data of the ball and the entropy value information of the juggling action, the pandas library in the Python environment is used to perform processing operations on the time point data, and the to_datetime function and the diff method are used to perform format conversion and time interval calculation, so as to provide a timestamp for each contact in the juggling process and generate processed time series data; S502: Based on the processed time series data, the numpy library is used to perform basic statistical analysis of the juggling rhythm, and the mean, standard deviation, minimum and maximum values ​​of the time interval are calculated through the mean, std, min and max functions, so as to provide a quantitative feature description for the juggling rhythm and generate a basic statistical parameter set of the juggling rhythm; S503: Based on the basic statistical parameter set of juggling rhythm, Fourier transform is applied to analyze the frequency characteristics of juggling rhythm. The fft function in the numpy library is used to convert the time series data from the time domain to the frequency domain, and the key rhythm patterns of the juggling action are identified. By acquiring and analyzing the rhythm change rules during the juggling process, a characteristic parameter set of the juggling rhythm is generated.

[0045] In sub-step S501, the time point data of the physical response data of the ball and the entropy value information of the juggling action are processed through the pandas library in the Python environment. This includes converting the original time data format into a unified timestamp format and calculating the time interval between each contact during the juggling process. The specific operation uses the to_datetime function of the pandas library to convert the original time point into a timestamp, and then uses the diff method to calculate the time interval between two adjacent contacts. Such processing provides an accurate timestamp for each contact in the juggling process and generates processed time series data. The completion of this step provides a basis for further juggling rhythm analysis, ensures the accuracy and availability of time data, and enables subsequent rhythm analysis and feature extraction to be based on accurate time information.

[0046] In sub-step S502, based on the processed time series data, the numpy library is used to perform basic statistical analysis of the juggling rhythm. This step includes calculating the mean value, standard deviation, minimum value and maximum value of the time interval. Such analysis provides a quantitative feature description for the juggling rhythm and generates a set of basic statistical parameters for the juggling rhythm. Through statistical parameters, the key features of the rhythm stability and variation range of the juggling process can be quantitatively described, providing important data support for further rhythm characteristic analysis and training optimization.

[0047] In sub-step S503, based on the basic statistical parameter set of juggling rhythm, Fourier transform is applied to analyze the frequency characteristics of juggling rhythm. This step uses the fft function in the numpy library to convert time series data from the time domain to the frequency domain to identify the key rhythm patterns of juggling action. Through Fourier transform, the rhythm change law in the juggling process can be obtained and analyzed, including the main frequency components of the juggling action and the rhythm change characteristics. The analysis from the frequency domain perspective generates a juggling rhythm characteristic parameter set, which provides a new perspective for understanding the rhythm characteristics of juggling skills. By identifying the main and secondary rhythm patterns in the juggling action, this step helps to design targeted training plans and optimize the rhythm control and stability of juggling skills.

[0048] See also Figure 7 Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average and standardization processing methods are used to evaluate the juggling skills, and dynamic skill tracking and feedback optimization are performed. The specific steps for generating the comprehensive juggling skill score results are as follows: S601: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average of multiple indicators is calculated using the numpy and pandas libraries in the Python environment, including assigning a weight to each indicator based on its impact on the juggling skill, and the abnormal action identification is assigned a weight according to the impact on the skill performance. Through the numpy.average method, the data of each indicator and its corresponding weight are input to calculate the weighted average and generate a weighted average performance score; S602: Based on the weighted average performance score, the weighted average performance score is standardized using the StandardScaler function of the sklearn library. The weighted average performance score is converted into a distribution with a mean of 0 and a unit variance, and the weighted average performance score data is passed in as an input value. The mean and standard deviation are calculated and applied to standardize the score data to generate a standardized score result. S603: Based on the standardized scoring results, regularly analyze the juggling skill scores, and guide the continuous optimization of the training plan by comparing the current scores with the historical scoring data to generate a comprehensive juggling skill scoring result.

[0049] In sub-step S601, the abnormal action identification and analysis results, key evaluation factor model and juggling rhythm characteristic parameter set are comprehensively evaluated through the numpy and pandas libraries in the Python environment. Each indicator is assigned a corresponding weight according to its influence on the juggling skill, including the abnormal action identification, whose weight reflects the influence of the action on the skill performance. Using the average method of numpy, the data of each indicator and its corresponding weight are input to calculate the weighted average performance score. This calculation process generates a weighted average performance score that comprehensively reflects all aspects of juggling skills, providing a quantitative basis for in-depth analysis of juggling skills, and also provides a standardized indicator for evaluating and comparing the juggling skill levels of different athletes.

[0050] In sub-step S602, based on the weighted average performance score, the StandardScaler function of the sklearn library is used to standardize the score, so that the score data is converted into a distribution with zero mean and unit variance. By passing in the weighted average performance score data as the input value, StandardScaler calculates and applies the corresponding mean and standard deviation for standardization, ensuring that the comparison and analysis of the scores are carried out under the same standard. The standardized scoring results generated by this process provide a unified standard for juggling skill evaluation that removes the influence of dimensions and individual differences, making the comparison of skill levels more fair and objective.

[0051] In sub-step S603, based on the standardized scoring results, regular analysis of juggling skill scores is performed. By regularly comparing the current score with the historical score data, the changing trend of the athlete's skill level can be monitored, thereby guiding the continuous optimization of the training plan. It not only helps to identify the improvement areas of juggling skills, but also motivates athletes to conduct targeted training in specific areas. The final generated comprehensive juggling skill score provides athletes and coaches with clear skill improvement goals and training feedback, promoting the continuous improvement of training effects and the overall improvement of skill levels.

[0052] See also Figure 8 Based on the comprehensive scoring results of juggling skills, the decision tree and rule engine technology are used to analyze the players' skills and formulate training plans. The results are tracked through the training plan. The specific steps to generate juggling scores and technical improvement feedback results are as follows: S701: Based on the comprehensive scoring results of the dribbling skills, a decision tree algorithm is used. The input features include the ball possession time, the number of dribbling times and the error rate. The splitting standard is set to information gain. The data is trained through the fit method. Model training and analysis are performed to identify skill weaknesses, and the player skill weakness analysis results are generated. S702: Based on the analysis results of players' skill weaknesses, a training plan is formulated using rule engine technology. The Drools rule engine is used to define training rules based on skill weaknesses. The rule engine automatically matches the training plan according to the analysis results to generate a dedicated training plan. S703: Based on the exclusive training program, the training plan is used to track the results. Decision tree analysis is used to compare the skill scores before and after training. By tracking the results and formulating a technical improvement plan, feedback results on juggling scores and technical improvements are generated.

[0053] In sub-step S701, the decision tree algorithm is used to analyze the comprehensive scoring results of the juggling skills, with ball possession time, number of juggling times and error rate as input features. Information gain is used as the splitting criterion to select the best splitting attribute, and the data set is trained through the fit method. This process aims to identify the weaknesses of the players in juggling skills. After the training is completed, the decision tree model can accurately identify the weaknesses of the players' skills based on the input features, generate the analysis results of the players' skills weaknesses and provide clear improvement directions for coaches and players, helping them understand the areas that need to be focused on and improved.

[0054] In sub-step S702, based on the analysis results of the players' skill weaknesses, a dedicated training plan is developed using rule engine technology. The Drools rule engine is used to define training rules based on skill weaknesses. The rules automatically match the corresponding training plan based on the analysis results. The rule engine will recommend training tasks such as fast ball control and completing a specified number of dribbles in a short time. In this way, a dedicated training plan for each player's specific skill weaknesses is generated, making the training more targeted and efficient.

[0055] In sub-step S703, based on the exclusive training program, the training plan is used to track the results, and the decision tree analysis is used again to compare the differences in skill scores before and after training. This process is designed to evaluate the effectiveness of the training program. By comparing the differences in skill scores before and after training, the player's progress in specific skills can be intuitively seen. In addition, through effect tracking, the training plan can be adjusted and optimized in a timely manner to ensure the maximization of training effects. The final generated juggling results and technical improvement feedback results provide players with specific feedback on skill improvement and guidance and suggestions for future training, which promotes the continuous progress and improvement of players' juggling skills.

[0056] See also Fig. 9 , a juggling score generation system based on neural network, the system includes contact mechanics analysis module, entropy analysis module, abnormal action recognition module, feature extraction module, rhythm analysis module, skill scoring module, training program formulation module, feedback optimization module; The contact mechanics analysis module uses molecular dynamic analysis technology based on the motion tracking data captured by the camera device. By defining the force field model and using the molecular dynamics simulation software LAMMPS, it performs numerical simulation and adjusts the friction coefficient parameters and elastic modulus parameters to generate the physical response data of the sphere. The entropy analysis module uses the Shannon entropy calculation method based on the physical response data of the ball, uses the pandas library in the Python environment to perform time series processing on the data, and uses the entropy function in the SciPy library to calculate the information entropy to generate the entropy value information of the juggling action; The abnormal action recognition module is based on the entropy information of the juggling action, adopts an abnormal detection model based on deep learning, uses the long short-term memory network and isolation forest algorithm built by TensorFlow and Keras libraries, processes time series data, identifies abnormal actions, and generates abnormal action identification and analysis results; The feature extraction module uses reverse engineering technology and convolutional neural network models based on abnormal action identification and analysis results, and uses TensorFlow and Keras libraries to build models, including multi-layer convolution and pooling layers to extract image features, iteratively optimize the network structure, and generate a key evaluation factor model; The rhythm analysis module uses the time series analysis method based on the physical response data of the ball and the entropy value information of the juggling action, and uses the numpy library in the Python environment to perform basic statistical analysis and Fourier transform to identify the characteristics of the juggling rhythm and generate a set of juggling rhythm characteristic parameters; The skill scoring module uses weighted average and standardization processing methods based on abnormal action identification and analysis results, key evaluation factor model and juggling rhythm characteristic parameter set. It calculates the weighted average of multiple indicators through numpy and pandas libraries and performs standardization to generate a comprehensive juggling skill score result. The training program formulation module analyzes the players' skills based on the comprehensive scoring results of the juggling skills. It adopts the decision tree algorithm and the DecisionTreeClassifier class in the Scikit-learn library to perform model training and analysis to identify skill weaknesses. It also uses the rule engine technology to formulate training programs and generate exclusive training programs. The feedback optimization module is based on a dedicated training program. It tracks the results through the training plan, adopts decision tree analysis, and uses the DecisionTreeClassifier class in the Scikit-learn library to compare the skill scores before and after training, and generate feedback results on juggling results and technical improvements.

[0057] The contact mechanics analysis module provides a microscopic perspective for understanding the complex physical interactions between the ball and the athlete through molecular dynamic analysis technology and numerical simulation. This not only enhances the understanding of the mechanical response during the juggling process, but also provides a scientific basis for improving ball design and athlete training methods, thereby helping to reduce the risk of sports injuries and improve the safety of training.

[0058] The entropy analysis module and abnormal action recognition module can accurately identify irregularities and abnormal actions in juggling skills through sophisticated data analysis and deep learning technology. This recognition not only helps athletes to find and correct technical defects in a timely manner, but also improves the stability and predictability of sports performance, thereby achieving better results in competitions.

[0059] The feature extraction module and rhythm analysis module extract key evaluation factors and rhythm features through advanced algorithms, which makes skill evaluation more accurate and comprehensive. Athletes can adjust their training focus based on the analysis results and improve specific skills more effectively, thereby improving their overall juggling skills.

[0060] The weighted average and normalized processing methods of the skill scoring module provide a quantitative score for athletes' skills, which not only helps athletes objectively evaluate their skill levels, but also promotes fair skill comparisons between athletes and stimulates healthy competition among athletes.

[0061] The training program development module and feedback optimization module use intelligent analysis and rule engine technology to develop exclusive training plans for athletes and adjust them according to the training results. This dynamic adjustment mechanism not only improves the pertinence and efficiency of training, but also provides timely feedback on training results, helping athletes to continuously improve their techniques and achieve long-term skill improvement.

[0062] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for generating juggling scores based on a neural network, characterized in that: The following steps are involved: Based on the motion tracking data captured by the camera equipment, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change of friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball; Based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity of the juggling action, and the information entropy value in the juggling process is calculated to analyze the randomness of the action sequence and generate the entropy value information of the juggling action; Based on the entropy value information of the juggling action, an abnormality detection model based on deep learning is used to analyze the juggling action again, and abnormal actions or deviations that deviate from the normal pattern are identified. Through neural network and pattern recognition technology, abnormal actions are located and their deviation characteristics are analyzed to generate abnormal action identification and analysis results; Based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network models are used to analyze the feature data of abnormal actions, and key image features and data indicators are inferred from the scoring targets of juggling skills, and image processing and feature extraction are performed to generate a key evaluation factor model; Based on the physical response data of the ball and the entropy value information of the juggling action, a time series analysis method is used to analyze the time interval and duration of each ball contact during the juggling process, a statistical analysis technique is used to extract characteristic parameters of the juggling rhythm, and a Fourier transform method is applied to identify and analyze the juggling rhythm pattern to generate a juggling rhythm characteristic parameter set; Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the juggling skills are evaluated by applying weighted average and standardization processing methods, and dynamic skill tracking and feedback optimization are performed to generate a comprehensive juggling skill scoring result; Based on the comprehensive scoring results of the juggling skills, decision tree and rule engine technology are used to analyze player skills and formulate training plans. The results are tracked through the training plan to generate juggling scores and technical improvement feedback results.

2. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: The physical response data of the ball includes the elastic coefficient, deformation recovery time and deformation degree of the ball, the entropy value information of the juggling action includes the average entropy value, maximum entropy value and entropy value variation coefficient of the action sequence, the abnormal action identification and analysis results include the frequency of occurrence of abnormal actions, the type of abnormal actions and the degree of influence of abnormal actions on the juggling rhythm, the key evaluation factor model includes the spatial accuracy, time synchronization and action fluency indicators of the juggling action, the juggling rhythm characteristic parameter set includes the basic frequency of the rhythm, the width of the frequency distribution and the characteristic value of the key rhythm pattern, the comprehensive scoring result of the juggling skill includes the player's skill level, skill improvement potential index and skill balance evaluation, and the juggling performance and technical improvement feedback results include skill improvement plans, skill mastery improvement intervals and priority technical improvement points.

3. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the motion tracking data captured by the camera, the molecular dynamic analysis technology is used to estimate the deformation of the ball and the change in friction by applying the force field model, analyze the force on the ball and the mechanical response of the athlete's contact point, and generate the physical response data of the ball. The specific steps are as follows: Based on the motion tracking data captured by the camera, molecular dynamic simulation was performed, and numerical simulation operations were performed using LAMMPS software. The behavior of the metal sphere was simulated by defining the EAM force field model, setting the time step to 1fs, adjusting the friction coefficient parameter to 0.3, and the elastic modulus parameter to 200 GPa, to generate contact interaction simulation data; Based on the contact interaction simulation data, finite element analysis is performed, and static structural analysis is performed using ANSYS software. Through tetrahedral meshing operations, the mesh size is set to 0.5 mm, a predefined contact force is applied to the top surface of the sphere, and the bottom surface of the sphere is fixed as a boundary condition to generate mechanical response analysis data; Based on the mechanical response analysis data, data visualization and analysis are performed. The stress-strain curve is drawn using the Matplotlib library in Python. The Pandas library performs data screening and processing operations. The X-axis of the chart is set to strain, the Y-axis is set to stress, and colors are used to identify various contact points to generate physical response data of the sphere.

4. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the physical response data of the ball, the Shannon entropy calculation method is used to evaluate the complexity of the juggling action. By calculating the information entropy value in the juggling process and analyzing the randomness of the action sequence, the specific steps of generating the entropy value information of the juggling action are as follows: Based on the physical response data of the ball, the contact time point and force information are processed in a time series manner using the pandas library in the Python environment, including converting the original data into a sequence indexed by a timestamp, defining a unique time point and force value for each ball juggling contact, and generating time series juggling data; Based on the time-series juggling data, the entropy function in the SciPy library is used to calculate the information entropy of the juggling action recorded by the capture device. By counting the frequency of each unique action state in the sequence, the Shannon entropy formula is applied. Calculate the information entropy of the entire sequence, where Indicates action The probability of occurrence is used to quantify the irregularity and randomness of the action sequence and generate the information entropy measurement results of the juggling action; Based on the juggling action information entropy measurement results, the Matplotlib library is used to graphically display the entropy value change trend, including mapping the time series data and its corresponding entropy value in the same chart, displaying the entropy value changes of the juggling action at various time points through graphics, identifying the change rules of the action sequence in the juggling process, and generating the entropy value information of the juggling action.

5. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the entropy value information of the juggling action, a deep learning-based anomaly detection model is used to analyze the juggling action again, identify abnormal actions or deviations from the normal pattern, locate the abnormal actions and analyze their deviation characteristics through neural network and pattern recognition technology, and generate abnormal action identification and analysis results. The specific steps are as follows: Based on the entropy information of the juggling action, a deep learning model is constructed using TensorFlow and Keras libraries, including using an LSTM layer to process time series data to capture the temporal dependency of the juggling action, configuring 256 hidden units in the LSTM layer to process the juggling action data, adding a Dropout layer with a ratio of 0.5 to reduce the risk of overfitting, performing action feature dimension conversion through a fully connected layer, identifying potential abnormal action patterns, and generating an abnormal action candidate set; Based on the abnormal action candidate set, the abnormal score threshold method and IsolationForest algorithm are used, and the number of isolation trees is set to 100. The abnormal score of each action is calculated. The abnormal score is calculated based on the degree of deviation between the probability distribution predicted by the model and the probability distribution of conventional actions. By comparing with a preset threshold, actions with scores higher than the threshold are screened and identified as abnormal actions, and a selected abnormal action set is generated; Based on the selected abnormal action set, a convolutional neural network is used to perform feature analysis. The convolutional neural network model configuration includes a convolution layer and a pooling layer. The convolution layer uses a 3x3 filter with a step size of 1. The pooling layer uses a 2x2 pooling operation to extract the spatial features of the abnormal actions, the speed of the actions, and the changes in the strength of the actions. The position and nature of the abnormal actions are located through the features, and the abnormal action identification and analysis results are generated.

6. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the abnormal action identification and analysis results, reverse engineering technology and convolutional neural network models are used to analyze the feature data of abnormal actions, infer key image features and data indicators from the scoring targets of juggling skills, perform image processing and feature extraction, and generate the key evaluation factor model. The specific steps are as follows: Based on the abnormal action identification and analysis results, a convolutional neural network model is constructed in a Python environment using TensorFlow and Keras libraries, including three convolutional layers Conv2D for image feature extraction. Each convolution layer is followed by a MaxPooling2D layer to reduce the dimension of the feature map. The convolutional layers are configured with 32, 64 and 128 filters respectively, and the filter size is set to 3x3. The ReLU activation function is used to introduce nonlinearity to capture the key features of the abnormal action images and generate a preliminary feature data set. Based on the preliminary feature data set, the network structure is iteratively optimized, two batch normalization layers are added, and two fully connected layers Dense are introduced to learn and classify deep features. The first fully connected layer is configured with 256 nodes and uses the ReLU activation function. The second fully connected layer is used as the output layer. The number of nodes corresponds to the number of categories of the juggling skill score. The softmax activation function is used for classification decision making. The evaluation factors of the juggling skill are learned and identified from the image features to generate an optimized feature learning model. Based on the optimized feature learning model, the model is trained and parameter optimized. During the training process, the batch size is set to 32, the number of iterations is set to 100, the Adam optimizer is used to adjust the learning rate to 0.001, and the loss function uses categorical_crossentropy to perform multi-classification loss calculation. By training and optimizing the model parameters, the model can extract key features from abnormal action images, reflect the scoring indicators of juggling skills, and generate a key evaluation factor model.

7. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the physical response data of the ball and the entropy value information of the juggling action, a time series analysis method is used to analyze the time interval and duration of each ball contact during the juggling process, a statistical analysis technique is used to extract characteristic parameters of the juggling rhythm, and a Fourier transform method is applied to identify and analyze the juggling rhythm pattern. The specific steps of generating a juggling rhythm characteristic parameter set are as follows: Based on the physical response data of the ball and the entropy value information of the juggling action, the pandas library in the Python environment is used to perform processing operations on the time point data, and the to_datetime function and the diff method are used to perform format conversion and time interval calculation, so as to provide a timestamp for each contact in the juggling process and generate processed time series data; Based on the processed time series data, the numpy library is used to perform basic statistical analysis of the juggling rhythm, and the mean, standard deviation, minimum and maximum values ​​of the time interval are calculated through the mean, std, min and max functions to provide a quantitative feature description for the juggling rhythm and generate a basic statistical parameter set of the juggling rhythm; Based on the basic statistical parameter set of juggling rhythm, Fourier transform is applied to analyze the frequency characteristics of juggling rhythm, and the fft function in the numpy library is used to convert time series data from time domain to frequency domain to identify the key rhythm patterns of juggling actions. By acquiring and analyzing the rhythm change rules during the juggling process, a juggling rhythm characteristic parameter set is generated.

8. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average and standardization processing methods are applied to evaluate the juggling skills, and dynamic skill tracking and feedback optimization are performed to generate the juggling skills comprehensive scoring results as follows: Based on the abnormal action identification and analysis results, the key evaluation factor model and the juggling rhythm characteristic parameter set, the weighted average of multiple indicators is calculated using the numpy and pandas libraries in the Python environment, including assigning a weight to each indicator based on its impact on the juggling skill. The abnormal action identification is weighted according to the impact on the skill performance. The numpy.average method is used to input each indicator data and its corresponding weight, calculate the weighted average, and generate a weighted average performance score. Based on the weighted average performance score, use the StandardScaler function of the sklearn library to convert it into a distribution with zero mean and unit variance, pass the data as input values, calculate and apply the mean and standard deviation, standardize the scoring data, and generate a standardized scoring result; Based on the standardized scoring results, the juggling skill scores are analyzed regularly, and by comparing the current scores with historical scoring data, the continuous optimization of the training plan is guided to generate a comprehensive juggling skill scoring result.

9. The method for generating juggling scores based on a neural network according to claim 1, characterized in that: Based on the comprehensive scoring results of the juggling skills, the decision tree and rule engine technology are used to analyze the player's skills and formulate a training plan, and the results are tracked through the training plan. The specific steps for generating juggling scores and technical improvement feedback results are as follows: Based on the comprehensive scoring result of the dribbling skill, a decision tree algorithm is adopted, the input features include ball possession time, number of dribbling times and error rate, the splitting standard is set to information gain, the data is trained by the fit method, model training and analysis are performed to identify skill weaknesses, and the player skill weakness analysis results are generated; Based on the analysis results of the players' skill weaknesses, a training plan is formulated using rule engine technology. The Drools rule engine is used to define training rules based on skill weaknesses. The rule engine automatically matches the training plan according to the analysis results to generate a dedicated training plan. Based on the exclusive training program, the results are tracked through the training plan, and decision tree analysis is used to compare the differences in skill scores before and after training. Through effect tracking and formulation of technical improvement plans, juggling results and technical improvement feedback results are generated.

10. A juggling score generation system based on a neural network, characterized in that: According to the method for generating juggling scores based on a neural network according to any one of claims 1 to 9, the system comprises a contact mechanics analysis module, an entropy value analysis module, an abnormal action recognition module, a feature extraction module, a rhythm analysis module, a skill scoring module, a training program formulation module, and a feedback optimization module; The contact mechanics analysis module uses molecular dynamic analysis technology based on the motion tracking data captured by the camera device, defines a force field model and uses the molecular dynamics simulation software LAMMPS to perform numerical simulation and adjust the friction coefficient parameters and elastic modulus parameters to generate physical response data of the sphere; The entropy value analysis module uses the Shannon entropy calculation method based on the physical response data of the ball, performs time series processing on the data through the pandas library in the Python environment, and uses the entropy function in the SciPy library to calculate the information entropy to generate the entropy value information of the juggling action; The abnormal action recognition module is based on the entropy information of the juggling action, adopts an abnormal detection model based on deep learning, uses the long short-term memory network and isolation forest algorithm built by TensorFlow and Keras libraries, processes time series data, identifies abnormal actions, and generates abnormal action identification and analysis results; The feature extraction module uses reverse engineering technology and convolutional neural network models based on abnormal action identification and analysis results, uses TensorFlow and Keras libraries to build models, including multi-layer convolution and pooling layers to extract image features, iteratively optimizes the network structure, and generates a key evaluation factor model; The rhythm analysis module uses a time series analysis method based on the physical response data of the ball and the entropy value information of the juggling action, performs basic statistical analysis and Fourier transform through the numpy library in the Python environment, identifies the characteristics of the juggling rhythm, and generates a juggling rhythm characteristic parameter set; The skill scoring module applies weighted average and standardization processing methods based on abnormal action identification and analysis results, key evaluation factor model and juggling rhythm characteristic parameter set, calculates the weighted average of multiple indicators through numpy and pandas libraries and performs standardization processing to generate a comprehensive juggling skill scoring result; The training program formulation module analyzes the player's skills based on the comprehensive scoring results of the juggling skills, adopts a decision tree algorithm, uses the DecisionTreeClassifier class in the Scikit-learn library, performs model training and analysis to identify skill weaknesses, and adopts rule engine technology to formulate a training program to generate a dedicated training program; The feedback optimization module is based on a dedicated training program, tracks the results through the training plan, adopts decision tree analysis, uses the DecisionTreeClassifier class in the Scikit-learn library, compares the skill score differences before and after training, and generates juggling results and technical improvement feedback results.

Images