Basketball shooting result prediction method and system based on machine learning

Abstract

The invention relates to the technical field of intelligent sports, in particular to a basketball shooting result prediction method and system based on machine learning, and the system comprises a data collection module, a data preprocessing module, a feature extraction module, a shooting result prediction model module, a real-time prediction module and a result application module. Multi-modal data in the shooting process are collected through a data collection module, the collected multi-modal data are preprocessed through a data preprocessing module, feature extraction is conducted on the multi-modal data through a feature extraction module, a shooting result prediction model module predicts a shooting result based on extracted feature vectors, and a real-time prediction module predicts the shooting result according to the prediction result. According to the intelligent basketball shooting system and method, the training effect and competition performance of basketball players can be improved, and development of the intelligent sports field is promoted.

Description

Technical Field

[0001] The present invention relates to the field of intelligent sports technology, and in particular to a method and system for predicting basketball shooting results based on machine learning. Background Art

[0002] Basketball is a widely played competitive sport around the world, and shooting accuracy directly affects game results and athletes' performance. With the development of science and technology, especially the continuous advancement of sensor technology, computer vision and deep learning algorithms, data-driven basketball training and analysis methods are gradually being applied.

[0003] Although the prediction methods of basketball shooting results based on machine learning and deep learning technologies have made progress to a certain extent, they still face some technical challenges. First, the existing multimodal data collection methods rely on sensors and cameras, but due to problems such as environmental noise, equipment accuracy and sensor errors, the collected data may have certain inaccuracies and omissions, which affects the accuracy of subsequent analysis. Secondly, in terms of feature extraction, the existing technologies mostly use traditional feature engineering methods. Although they can extract some basic information such as joint positions and shooting angles, it is often difficult to fully and accurately capture the characteristics of athletes' shooting movements when dealing with complex spatiotemporal dynamic changes. Especially when combining spatial features with temporal features, the effect of existing methods is relatively limited, resulting in unsatisfactory model training results. Finally, although deep learning algorithms have strong prediction capabilities in theory, the generalization ability and real-time response capabilities of existing models are still limited due to insufficient training data or inaccurate annotations. In a rapidly changing game environment, the prediction effect of the model may be unstable and cannot provide accurate tactical decision support. Summary of the invention

[0004] The present invention provides a basketball shooting result prediction method and system based on machine learning.

[0005] The method for predicting basketball shooting results based on machine learning includes the following steps:

[0006] S1, data acquisition: through the high-precision camera equipment, inertial measurement unit and sensor deployed in the basketball court, multi-modal data in the shooting process is collected in real time, and the multi-modal data includes the athlete's joint position information, shooting parameters and initial motion parameters of the basketball, among which;

[0007] The joint position information includes the spatial position and movement trajectory of the shoulder, elbow and wrist;

[0008] The shooting parameters include shooting speed, shooting angle and shooting point position;

[0009] The initial motion parameters include initial velocity, rotation rate and launch angle;

[0010] S2, data preprocessing: preprocess the collected multimodal data, including noise removal, missing value filling and standardization;

[0011] S3, feature extraction: Convolutional neural network is used to extract spatial features of multimodal data, including joint angles, limb position changes and spatiotemporal patterns of shooting actions. Long short-term memory network is used to process the temporal features in multimodal data, capture the temporal dynamic changes during the action process, and combine spatial features with temporal features to generate feature vectors.

[0012] S4, model construction: based on the deep learning algorithm, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data. The trained shooting result prediction model predicts the shooting result based on the input feature vector;

[0013] S5, real-time prediction: during the game or training, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability, and a shooting hit rate report is generated based on the prediction result;

[0014] S6, Application of results: In personalized training, provide athletes with shooting action adjustment suggestions based on real-time prediction results to improve their shooting accuracy. In games, provide coaches with tactical decision-making support and provide tactical guidance based on shooting accuracy probability.

[0015] Optionally, the S1 includes:

[0016] S11, collecting shooting parameters: using a dedicated sensor to collect shooting parameters in real time during the shooting process, including shooting speed, shooting angle and shooting point position;

[0017] S12, collecting joint position information: installing an inertial measurement unit and a motion capture system on the athlete's shoulder, elbow, and wrist to measure the athlete's joint position information and motion trajectory in real time during the shooting process;

[0018] S13, collecting initial motion parameters: installing a dedicated sensor on the basketball to collect the initial motion parameters of the basketball in real time, including initial velocity, rotation rate and launch angle;

[0019] S14, data synchronization and transmission: Synchronize the data from various sensors (including joint position information, shooting parameters and initial motion parameters) through wireless communication technology.

[0020] Optionally, S2 includes:

[0021] S21, noise removal: During the data collection process, due to sensor errors, environmental factors or other external interference, noise may be generated, affecting the accuracy of the data. Therefore, noise is removed from the collected multimodal data;

[0022] S22, missing value filling: During the data collection process, some data may be missing due to sensor failure or other reasons. Therefore, linear interpolation is used to fill the missing data in the collected multimodal data;

[0023] S23, standardization processing: In order to eliminate the dimensional differences between different features and improve the efficiency of model training, the Z-score standardization method is used to standardize the multimodal data;

[0024] S24, data verification: after completing the data preprocessing, verify the processing results.

[0025] Optionally, the S3 includes:

[0026] S31, spatial feature extraction: In the feature extraction process, convolutional neural network is used to extract spatial features from multimodal data;

[0027] S32, spatial feature optimization: The extracted features are integrated through the fully connected layer of the convolutional neural network to obtain the spatial feature vector.

[0028] Optionally, the S3 further comprises:

[0029] S33, time series feature extraction: In the feature extraction process, the time series features in the multimodal data are processed using a long short-term memory network to extract the time series features;

[0030] S34, temporal feature integration: Integrate the temporal features extracted by the long short-term memory network with the spatial feature vector extracted by the convolutional neural network to generate the final feature vector.

[0031] Optionally, the S4 includes:

[0032] S41, model construction: construct a shooting result prediction model based on a deep neural network, and the shooting result prediction model structure includes an input layer, a hidden layer and an output layer;

[0033] S42, Loss function: Select cross entropy as the loss function of the shooting result prediction model.

[0034] Optionally, S4 further includes:

[0035] S43, model training: After the shooting result prediction model is constructed, the shooting result prediction model is trained using historical multimodal data.

[0036] S44, model verification: after the training is completed, the shooting result prediction model is verified;

[0037] S45, shooting result prediction: after the verification is completed, the shooting result prediction is performed on the input real-time multimodal data using the trained shooting result prediction model.

[0038] Optionally, the S5 includes:

[0039] S51, Real-time data acquisition: During the game or training, the multi-modal data of athletes are collected in real time through high-precision cameras, inertial measurement units and sensors deployed in the basketball court.

[0040] S52, feature vector generation: feature extraction is performed on the multimodal data collected in real time through the feature extraction module, spatial features are extracted using a convolutional neural network, and temporal features are processed using a long short-term memory network, and the two are combined to generate a feature vector.

[0041] S53, prediction of shot hit probability: inputting the generated feature vector into the trained shot result prediction model, and the model outputs the shot hit probability value;

[0042] S54, hit rate report generation: based on the shooting hit probability value, automatically generate a shooting hit rate report, the report includes the athlete's shooting performance in the current game or training, such as shooting hit rate, hit probability distribution, etc.

[0043] Optionally, the S6 includes:

[0044] S91, Application in Personalized Training: In the process of personalized training, based on the shooting hit probability in the real-time prediction results, provide athletes with targeted shooting action adjustment suggestions;

[0045] S92, Application in the game: During the game, the real-time prediction results of the shooting probability provide the coach with accurate tactical decision support.

[0046] The basketball shooting result prediction system based on machine learning is used to implement the above-mentioned basketball shooting result prediction method based on machine learning, and includes the following modules:

[0047] Data acquisition module: Through high-precision cameras, inertial measurement units and sensors deployed in the basketball court, multi-modal data during the shooting process is collected in real time. The multi-modal data includes the athlete's joint position information, shooting parameters and the initial motion parameters of the basketball;

[0048] Data preprocessing module: preprocess the collected multimodal data, including noise removal, missing value filling and standardization;

[0049] Feature extraction module: Convolutional neural network is used to extract spatial features from multimodal data, long short-term memory network is used to process temporal features in multimodal data, spatial features are combined with temporal features to generate feature vectors;

[0050] Shooting result prediction model module: Based on the deep neural network, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data;

[0051] Real-time prediction module: During a game or training session, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability. Based on the prediction results, a shooting hit rate report is generated.

[0052] Result application module: In personalized training, it provides athletes with shooting action adjustment suggestions based on real-time prediction results. In the game, it provides tactical decision-making support for coaches and provides tactical guidance based on the probability of shooting success.

[0053] Beneficial effects of the present invention:

[0054] The present invention uses high-precision multimodal data collection, including athlete's joint position information, shooting parameters and initial motion parameters of basketball, combined with deep learning algorithms to accurately extract and analyze these data. By extracting spatial features through convolutional neural networks and capturing temporal features using long short-term memory networks, it is possible to effectively capture the dynamic changes of athletes during the shooting process, form accurate feature vectors, and train the shooting result prediction model. The training of the prediction model is based on historical data, so that the prediction results have high accuracy, and can accurately predict the probability of shooting hits during actual training or competitions, and provide athletes with real-time action adjustment suggestions. Therefore, the present invention has significant advantages in real-time prediction and shooting hit rate improvement.

[0055] The present invention can provide athletes with targeted shooting action adjustment suggestions according to the real-time shooting hit probability of the athletes in personalized training. By continuously optimizing the shooting angle, shooting speed, shooting point position and other action elements during the training process, athletes are helped to gradually improve their shooting hit rate. Provide accurate tactical decision support for coaches. According to the shooting hit probability of each athlete, the coach can dynamically adjust the offensive strategy, select the most advantageous shooting plan or adjust the offensive layout, thereby improving the team's overall scoring ability. This combination of personalized training and accurate tactical support helps athletes and coaches make more efficient decisions in actual games. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0058] Figure 2 Schematic diagram of a system flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0060] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0061] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0062] like Figure 1 As shown, the basketball shooting result prediction method based on machine learning includes the following steps:

[0063] S1, data collection: through the high-precision camera equipment, inertial measurement unit and sensor deployed in the basketball court, multi-modal data of the shooting process is collected in real time. The multi-modal data includes the athlete's joint position information, shooting parameters and the initial motion parameters of the basketball, among which;

[0064] Joint position information includes the spatial position and motion trajectory of the shoulder, elbow, and wrist;

[0065] Shooting parameters include shooting speed, shooting angle and shooting point location;

[0066] Initial motion parameters include initial velocity, rotation rate, and launch angle;

[0067] S2, data preprocessing: preprocess the collected multimodal data, including noise removal, missing value filling and standardization;

[0068] S3, feature extraction: Convolutional neural network is used to extract spatial features of multimodal data, including joint angles, limb position changes and spatiotemporal patterns of shooting actions. Long short-term memory network is used to process the temporal features in multimodal data, capture the temporal dynamic changes during the action process, and combine spatial features with temporal features to generate feature vectors.

[0069] S4, model construction: based on the deep learning algorithm, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data. The trained shooting result prediction model predicts the shooting result (hit or miss) based on the input feature vector.

[0070] S5, real-time prediction: during the game or training, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability, and a shooting hit rate report is generated based on the prediction result;

[0071] S6, Application of results: In personalized training, provide athletes with shooting action adjustment suggestions based on real-time prediction results to improve their shooting accuracy. In games, provide coaches with tactical decision-making support and provide tactical guidance based on shooting accuracy probability.

[0072] S1 includes:

[0073] S11, collecting shooting parameters: using a dedicated sensor to collect shooting parameters in real time during the shooting process, specifically including shooting speed, shooting angle and shooting point position, the steps specifically include:

[0074] Shot speed: Use high-speed camera equipment or radar speed meter (such as Doppler radar) to measure the speed of the basketball when it leaves the player's hand in real time. Radar speed meter can measure the moving speed of an object (such as a basketball) through reflected signals, which is particularly suitable for accurately capturing the initial velocity of the basketball;

[0075] Release angle: The inertial measurement unit or angle sensor measures the launch angle of the basketball when it leaves the player's hand, that is, the angle between the basketball and the ground;

[0076] Release point location: Use a laser rangefinder or positioning system (such as GPS or indoor positioning system) to record the spatial coordinates of the athlete's hand when shooting to determine the specific location of the release point;

[0077] S12, collecting joint position information: installing an inertial measurement unit and a motion capture system on the athlete's shoulder, elbow, and wrist to measure the athlete's joint position information and motion trajectory in real time during the shooting process, including:

[0078] Joint spatial position: The inertial measurement unit combines accelerometers, gyroscopes and magnetometers to measure the spatial position of joints such as shoulders, elbows and wrists in real time, and obtain the displacement and posture changes of each joint during the shooting action;

[0079] Joint motion trajectory: Through the motion capture system (such as VICON system or camera equipment), the motion trajectory of the shoulder, elbow and wrist is tracked in real time to generate joint displacement curves in three-dimensional space in order to accurately analyze the athlete's shooting action;

[0080] S13, collecting initial motion parameters: by installing a special sensor on the basketball, collecting the initial motion parameters of the basketball in real time, including initial velocity, rotation rate and launch angle, the steps include:

[0081] Initial velocity: The speed of the basketball when it leaves the player's hand is measured by Doppler radar speedometer or high-speed camera equipment. The speedometer can capture the initial velocity of the basketball, especially at the moment of release;

[0082] Spin rate: By installing a rotation sensor (such as a gyroscope) on the basketball, the rotation rate of the basketball is measured while it is flying in the air, especially at the moment it leaves the player's palm;

[0083] Launch angle: The launch angle of the basketball when it leaves the player’s hand is measured through an inertial measurement unit (IMU) or angle sensor, that is, the launch angle of the basketball relative to the ground;

[0084] S14, data synchronization and transmission: synchronizing the data from each sensor (including joint position information, shooting parameters, and initial motion parameters) through wireless communication technology, and ensuring the consistency of the data in time;

[0085] By deploying high-precision cameras, inertial measurement units, sensors and other equipment, we collect real-time information on athletes' joint position (spatial position and motion trajectory of the shoulder, elbow and wrist), shooting parameters (shooting speed, shooting angle and shooting point position) and the initial motion parameters of the basketball (initial velocity, rotation rate and launch angle), forming a multimodal data set, which provides comprehensive and accurate basic data support for subsequent data processing, feature extraction and shooting result prediction.

[0086] S2 includes:

[0087] S21, noise removal: During the data acquisition process, due to sensor errors, environmental factors or other external interference, noise may be generated, affecting the accuracy of the data. Therefore, the collected multimodal data is subjected to noise removal. The specific method is to use a low-pass filter to smooth the data to remove high-frequency noise and retain the main trend of the signal. The frequency response function of the filter is:

[0088]

[0089] Where f is the frequency of the signal, f c is the filter cutoff frequency, n is the filter order, and H(f) is the filter frequency response. This method removes high-frequency noise from the data and retains important low-frequency signals;

[0090] S22, missing value filling: During the data collection process, some data may be missing due to sensor failure or other reasons. Therefore, the linear interpolation method is used to fill the missing data in the collected multimodal data. The linear interpolation method interpolates between adjacent known data points to infer the value of the missing data. The specific calculation formula is:

[0091]

[0092] Among them, X1 and X2 are known adjacent data points, t1 and t2 are the timestamps of adjacent data points, and t missing is the timestamp of missing data, X missing is the estimated value of the missing data. This method can effectively fill in the missing values ​​and ensure the continuity and integrity of the data;

[0093] S23, standardization processing; in order to eliminate the dimensional differences between different features and improve the efficiency of model training, the Z-score standardization method is used to standardize the multimodal data. The Z-score standardization converts the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The specific calculation formula is:

[0094]

[0095] Among them, X is the original data, μ is the mean of the feature, and σ is the standard deviation of the feature. Through this method, all data features have the same scale, which facilitates the convergence and stability of subsequent model training;

[0096] S24, data verification: After completing data preprocessing, verify the processing results by calculating the statistical characteristics of the data (such as mean, standard deviation, skewness, kurtosis, etc.) to check whether the preprocessed data conforms to the expected distribution and whether there are any anomalies. If it is found that the data does not meet expectations, the data processing process is corrected using a backtracking method to ensure the quality of the final processed data.

[0097] The quality of multimodal data is effectively improved through noise removal, missing value filling and standardization. Noise removal uses a low-pass filter to remove high-frequency noise, missing value filling uses linear interpolation to restore data gaps, and standardization uses Z-score standardization to unify the data to the same scale. These steps ensure the integrity, consistency and availability of the data, providing high-quality input data for subsequent feature extraction and model training.

[0098] S3 includes:

[0099] S31, spatial feature extraction: In the feature extraction process, convolutional neural network is used to extract spatial features from multimodal data. The specific steps are as follows:

[0100] The athlete joint position information, shooting parameters, and initial basketball motion parameters obtained from the data collection step are input into the convolutional neural network. The convolutional neural network extracts local spatial features from the data through a series of convolutional and pooling layers, mainly including joint angles, limb position changes, and spatiotemporal patterns of shooting movements.

[0101] The convolutional layer is used to extract spatial features such as joint angles and limb position changes, such as changes in the position of key parts such as shoulders, elbows, and wrists, as well as changes in the relative position of various parts of the body.

[0102] In the output layer of the convolutional neural network, the features are nonlinearly transformed through activation functions (such as ReLU) to further enhance the expressiveness of the features. The structure of the convolutional neural network is defined as:

[0103] F out =σ(W2(σ(W1F in +b1))+b2);

[0104] Among them, F in is the input feature, W1 and b1 are the weight and bias of the first convolution layer, σ is the activation function, W2 and b2 are the weight and bias of the second convolution layer, and F out is the extracted feature;

[0105] S32, spatial feature optimization: The extracted features are integrated through the fully connected layer of the convolutional neural network to obtain a spatial feature vector, which will be input into the subsequent model building stage to predict the shooting results;

[0106] By using convolutional neural networks to extract spatial features from multimodal data, we can effectively capture spatial information such as the athlete's joint angles and limb position changes. This step provides key spatial features for subsequent prediction of shooting results.

[0107] The S3 in question also includes:

[0108] S33, time series feature extraction: In the feature extraction process, the time series features in the multimodal data are processed using a long short-term memory network to extract the time series features. The specific steps are as follows:

[0109] The athlete action data obtained from the data collection step (such as the trajectory of limb movement when shooting, the speed of the shot, etc.) is input into the long short-term memory network. The long short-term memory network has the ability to capture long-term dependencies in the time series and handle the dynamic changes of the shooting action.

[0110] The LSTM network gradually extracts the timing features of the shooting action from the historical time series data through a mechanism of input gate, forget gate and output gate, and updates the data at each time step. In this way, the LSTM network can capture the temporal dynamic changes of the athlete's action during the shooting process.

[0111] The calculation formula of long short-term memory network is expressed as:

[0112] f t =σ(W f x t +U f h t-1 +b f );

[0113] i t =σ(W i x t +U i h t-1 +b i );

[0114] c t =f t ⊙C t--1 +i t ⊙tanh(W c x t +U c h t-1 +b c );

[0115] o t =σ(W o x t +U o ht-1 +b o );

[0116] h t =o t ⊙tanh(c t );

[0117] Among them, f t ,i t , c t and t They are forget gate, input gate, cell state and output gate respectively, x t is the current input data, h t-1 is the hidden state of the previous moment, W, U, b are the corresponding weight matrices and bias terms; S34, time series feature integration: the time series features extracted by the long short-term memory network are integrated with the spatial feature vector extracted by the convolutional neural network to generate the final feature vector as the input of the shooting result prediction model. The feature vector contains the time series and spatial information of the shooting process, which helps to more accurately predict whether the shot is successful or not;

[0118] By extracting temporal features from multimodal data using a long short-term memory network, we can effectively capture the temporal dynamic changes in shooting movements, especially the subtle temporal features of the athlete's movements. Combined with spatial features, the extraction of temporal features further improves the expressiveness of features and provides richer input data for predicting shooting results.

[0119] S4 includes:

[0120] Select a deep learning algorithm suitable for the task of predicting shooting results. In order to accurately predict the shooting results, a deep neural network is selected as the shooting result prediction model. The deep neural network consists of multiple layers of fully connected neurons and can effectively process multi-dimensional input features, especially suitable for extracting complex nonlinear relationships from high-dimensional features;

[0121] Specifically, the deep neural network contains multiple hidden layers. The input layer receives the feature vector generated from the feature extraction step. The hidden layer extracts higher-order feature representations layer by layer through activation functions (such as ReLU or Sigmoid). The output layer predicts the probability of a shot through the Softmax or Sigmoid function.

[0122] S41, model construction: construct a shooting result prediction model based on a deep neural network, the shooting result prediction model structure includes an input layer, a hidden layer and an output layer, wherein;

[0123] The input layer receives the generated feature vector, which includes spatial and temporal features extracted from the player's joint position information, shooting parameters, and initial motion parameters.

[0124] The hidden layer consists of several neurons and uses the ReLU activation function to learn complex nonlinear relationships in the data. The Dropout layer can be used to prevent overfitting and improve the generalization ability of the model.

[0125] The output layer uses a Sigmoid activation function to output the probability of a shot being successful, expressed as a hit (value close to 1) or a miss (value close to 0).

[0126] S42, loss function: Select cross entropy as the loss function of the shooting result prediction model. The cross entropy loss function is suitable for binary classification tasks and can measure the difference between the probability distribution of the model output and the true label (hit or miss):

[0127]

[0128] Among them, y i is the true label, is the predicted probability of shooting success, N is the number of samples, and θ is the model parameter. The optimization algorithm uses the Adam optimizer, which performs well in large-scale data and high-dimensional feature space, and can adaptively adjust the learning rate to accelerate the convergence process;

[0129] Building a shooting result prediction model through a deep neural network can effectively extract useful information from the high-dimensional features of the input, and learn complex nonlinear relationships through multi-layer neural networks, providing an accurate model foundation for predicting shooting results.

[0130] S4 also includes:

[0131] S43, model training: After the shooting result prediction model is constructed, the shooting result prediction model is trained using historical multimodal data. The training process includes the following steps:

[0132] The multimodal features in the historical multimodal data (including the athlete's joint position information, shooting parameters, initial motion parameters, etc.) and the corresponding shooting results (hit or miss) are used as training data. The training data is divided into a training set and a validation set, and the model is trained using the training set data.

[0133] The model is trained using batch gradient descent or mini-batch gradient descent optimization algorithms. Each training sample will go through forward propagation (calculating the predicted value) and backpropagation (updating the network weights).

[0134] During the training process, the model parameters are optimized through multiple rounds of iterations so that the model can minimize the loss function and gradually improve the prediction accuracy.

[0135] S44, model verification: After the training is completed, the shooting result prediction model is verified through the verification set, and the performance of the model is further optimized by adjusting hyperparameters such as learning rate and batch size.

[0136] The validation set is used to evaluate the accuracy of the shooting result prediction model, avoid overfitting, and ensure that the model can also show good prediction ability on new data;

[0137] Cross-validate the shooting result prediction model and select the best model parameter combination to achieve the best prediction effect;

[0138] S45, shooting result prediction: After the verification is completed, the trained shooting result prediction model is used to predict the shooting result of the input real-time multimodal data, including:

[0139] The multimodal data collected in real time during the shooting process is converted into a feature vector and input into a trained shooting result prediction model for prediction;

[0140] The output layer of the shooting result prediction model will give the shooting hit probability, and judge whether the shot is hit or not based on this probability. Specifically, when the hit probability is greater than the set threshold, it is judged as a hit, otherwise it is judged as a miss;

[0141] By training and validating the deep neural network, we can ensure that the model can fully learn the complex relationships in multimodal data to accurately predict the results of shots. This training process can optimize the model parameters, improve the model's predictive ability, and apply it to real-time shot prediction.

[0142] S5 includes:

[0143] S51, real-time data acquisition: during the game or training, the multi-modal data of the athletes are collected in real time through high-precision camera equipment, inertial measurement units and sensors deployed in the basketball court. These data include the athlete's joint position information, shooting parameters and initial motion parameters of the basketball. The multi-modal data is transmitted to the data processing unit in real time;

[0144] Specifically include:

[0145] Joint position information, such as the position and movement trajectory of the shoulder, elbow, and wrist.

[0146] Shooting parameters, such as shooting speed, shooting angle, shooting point location, etc.

[0147] Initial motion parameters, such as the initial velocity and rotation rate of the basketball.

[0148] S52, feature vector generation: extract features from multimodal data collected in real time through a feature extraction module, extract spatial features using a convolutional neural network, process temporal features using a long short-term memory network, and combine the two to generate a feature vector;

[0149] This feature vector includes the spatiotemporal pattern of the player's current shooting action and historical action information, which serves as the input of the shooting result prediction model.

[0150] S53, prediction of shot hit probability: input the generated feature vector into the trained shot result prediction model, and the model outputs a shot hit probability value, which indicates the possibility of the player's shot hit. The prediction process is completed through forward propagation, in which the shot hit probability value is calculated based on the input features;

[0151] S54, generating a shooting percentage report: based on the shooting hit probability value, automatically generating a shooting percentage report, the report including the athlete's shooting performance in the current game or training, such as shooting hit percentage, hit probability distribution, etc., and providing the athlete with real-time feedback on the shooting hit trend;

[0152] The report can also be analyzed by time, position or movement to help athletes understand their performance in each link and identify potential technical problems.

[0153] S6 includes:

[0154] S91, Application in Personalized Training: In the process of personalized training, based on the shooting hit probability in the real-time prediction results, provide athletes with targeted shooting action adjustment suggestions. The specific steps are as follows:

[0155] Evaluate the athlete's current shooting performance and identify potential deficiencies or issues through shooting percentage reports, generating feedback for each shot taken by the athlete;

[0156] Provide optimization solutions based on real-time prediction results, such as shooting angle, shooting speed, shooting point location, etc. For example, if the hit probability is low, it is recommended to adjust the shooting angle, increase the shooting speed, or optimize the shooting point selection;

[0157] Based on multiple rounds of feedback and historical data, we adjust the athletes' technical movements and optimize their shooting accuracy, thereby improving their overall hit rate. During the training process, we help athletes quickly improve their shooting skills through repeated training by constantly adjusting and optimizing their shooting movements.

[0158] S92, Application in the game: During the game, based on the shooting probability in the real-time prediction results, provide the coach with accurate tactical decision support. The specific steps are as follows:

[0159] Analyze each player's shooting performance in the game based on their real-time shooting success rate, especially the success rate at critical moments;

[0160] Based on the probability of shooting success, tactical guidance is provided to coaches to help them decide when and where to arrange shots or passes. For example, when the probability of shooting is high, the coach can decide to let the athlete take more shots; when the probability of shooting is low, it can be recommended to adjust tactics and choose a more efficient offensive method or a more advantageous position.

[0161] like Figure 2 As shown, the basketball shooting result prediction system based on machine learning is used to implement the above-mentioned basketball shooting result prediction method based on machine learning, and includes the following modules:

[0162] Data acquisition module: Through high-precision cameras, inertial measurement units and sensors deployed in the basketball court, multi-modal data during the shooting process is collected in real time. The multi-modal data includes the athlete's joint position information, shooting parameters and the initial motion parameters of the basketball;

[0163] Data preprocessing module: preprocess the collected multimodal data, including noise removal, missing value filling and standardization;

[0164] Feature extraction module: Convolutional neural network is used to extract spatial features from multimodal data, long short-term memory network is used to process temporal features in multimodal data, spatial features are combined with temporal features to generate feature vectors;

[0165] Shooting result prediction model module: Based on the deep neural network, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data;

[0166] Real-time prediction module: During a game or training session, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability. Based on the predicted results, a shooting hit rate report is generated.

[0167] Result application module: In personalized training, it provides athletes with shooting action adjustment suggestions based on real-time prediction results. In the game, it provides tactical decision-making support for coaches and provides tactical guidance based on the probability of shooting success.

[0168] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0169] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A basketball shooting result prediction method based on machine learning, characterized in that: The following steps are involved: S1, data collection: through the camera equipment, inertial measurement unit and sensor deployed in the basketball court, multimodal data in the shooting process is collected in real time, and the multimodal data includes the athlete's joint position information, shooting parameters and initial motion parameters of the basketball, wherein; The joint position information includes the spatial position and movement trajectory of the shoulder, elbow and wrist; The shooting parameters include shooting speed, shooting angle and shooting point position; The initial motion parameters include initial velocity, rotation rate and launch angle; S2, data preprocessing: preprocess the collected multimodal data, including noise removal, missing value filling and standardization; S3, feature extraction: Convolutional neural network is used to extract spatial features of multimodal data, including joint angles, limb position changes and spatiotemporal patterns of shooting actions. Long short-term memory network is used to process the temporal features in multimodal data, capture the temporal dynamic changes during the action process, and combine spatial features with temporal features to generate feature vectors. S4, model construction: based on the deep learning algorithm, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data. The trained shooting result prediction model predicts the shooting result based on the input feature vector; S5, real-time prediction: during the game or training, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability, and a shooting hit rate report is generated based on the prediction result; S6, Application of results: In personalized training, provide athletes with shooting action adjustment suggestions based on real-time prediction results. In competitions, provide coaches with tactical decision-making support and provide tactical guidance based on the probability of shooting success.

2. The method for predicting basketball shooting results based on machine learning according to claim 1, characterized in that: The S1 includes: S11, collecting shooting parameters: using a dedicated sensor to collect shooting parameters in real time during the shooting process, including shooting speed, shooting angle and shooting point position; S12, collecting joint position information: installing an inertial measurement unit and a motion capture system on the athlete's shoulder, elbow, and wrist to measure the athlete's joint position information and motion trajectory in real time during the shooting process; S13, collecting initial motion parameters: installing a dedicated sensor on the basketball to collect the initial motion parameters of the basketball in real time, including initial velocity, rotation rate and launch angle; S14, data synchronization and transmission: synchronize the data from various sensors through wireless communication technology.

3. The method for predicting basketball shooting results based on machine learning according to claim 2, characterized in that: The S2 includes: S21, noise removal: performing noise removal on the collected multimodal data; S22, missing value filling: linear interpolation is used to fill in the missing data in the collected multimodal data; S23, standardization: Z-score standardization method is used to standardize multimodal data; S24, data verification: after completing the data preprocessing, verify the processing results.

4. The method for predicting basketball shooting results based on machine learning according to claim 3, characterized in that: The S3 includes: S31, spatial feature extraction: In the feature extraction process, convolutional neural network is used to extract spatial features from multimodal data; S32, spatial feature optimization: The extracted features are integrated through the fully connected layer of the convolutional neural network to obtain the spatial feature vector.

5. The method for predicting basketball shooting results based on machine learning according to claim 4, characterized in that: The S3 in question also includes: S33, time series feature extraction: In the feature extraction process, the time series features in the multimodal data are processed using a long short-term memory network to extract the time series features; S34, temporal feature integration: Integrate the temporal features extracted by the long short-term memory network with the spatial feature vector extracted by the convolutional neural network to generate the final feature vector.

6. The method for predicting basketball shooting results based on machine learning according to claim 5, characterized in that: The S4 includes: S41, model construction: construct a shooting result prediction model based on a deep neural network, and the shooting result prediction model structure includes an input layer, a hidden layer and an output layer; S42, Loss function: Select cross entropy as the loss function of the shooting result prediction model.

7. The method for predicting basketball shooting results based on machine learning according to claim 6, characterized in that: The S4 further comprises: S43, model training: after the shooting result prediction model is constructed, the shooting result prediction model is trained using historical multimodal data; S44, model verification: after the training is completed, the shooting result prediction model is verified; S45, shooting result prediction: after the verification is completed, the shooting result prediction is performed on the input real-time multimodal data using the trained shooting result prediction model.

8. The method for predicting basketball shooting results based on machine learning according to claim 7, characterized in that: The S5 includes: S51, real-time data collection: during the game or training, the multi-modal data of athletes are collected in real time through the camera equipment, inertial measurement unit and sensors deployed in the basketball court; S52, feature vector generation: extracting features from multimodal data collected in real time through a feature extraction module to generate a feature vector; S53, prediction of shot hit probability: inputting the generated feature vector into the trained shot result prediction model, and the model outputs the shot hit probability value; S54, generating a shooting percentage report: automatically generating a shooting percentage report based on the shooting probability value.

9. The method for predicting basketball shooting results based on machine learning according to claim 8, characterized in that: The S6 includes: S91, Application in Personalized Training: In the process of personalized training, based on the shooting hit probability in the real-time prediction results, provide athletes with targeted shooting action adjustment suggestions; S92, Application in the game: During the game, the real-time prediction results of the shooting probability provide the coach with accurate tactical decision support.

10. A basketball shot result prediction system based on machine learning, used to implement a basketball shot result prediction method based on machine learning as described in any one of claims 1 to 9, characterized in that: Includes the following modules: Data acquisition module: Through the camera equipment, inertial measurement unit and sensors deployed in the basketball court, multi-modal data of the shooting process is collected in real time. The multi-modal data includes the athlete's joint position information, shooting parameters and the initial motion parameters of the basketball; Data preprocessing module: preprocess the collected multimodal data, including noise removal, missing value filling and standardization; Feature extraction module: Convolutional neural network is used to extract spatial features from multimodal data, long short-term memory network is used to process temporal features in multimodal data, spatial features are combined with temporal features to generate feature vectors; Shooting result prediction model module: Based on the deep neural network, a shooting result prediction model is constructed. The shooting result prediction model uses the generated feature vector and is trained in combination with historical multimodal data; Real-time prediction module: During a game or training session, the multimodal data collected in real time is input into the shooting result prediction model to generate a predicted value of the shooting hit probability. Based on the prediction results, a shooting hit rate report is generated. Result application module: In personalized training, it provides athletes with shooting action adjustment suggestions based on real-time prediction results. In the game, it provides tactical decision-making support for coaches and provides tactical guidance based on the probability of shooting success.

Images