CNN Multi-Position Wearable Sensor Human Activity Recognition Used in Basketball Training

Area of Science:

• Artificial Intelligence
• Sensor Technology
• Sports Science

Background:

• Human activity recognition using wearable sensors is crucial for AI applications.
• Optimizing convolutional neural network (CNN) structures is key to improving efficiency.
• Basketball training benefits from accurate activity recognition.

Purpose of the Study:

• To propose a novel convolutional network entity model with shared parameters for human activity recognition.
• To optimize network structure and reduce training parameters in CNNs.
• To validate the model's effectiveness in multi-position wearable sensor-based activity recognition for basketball.

Main Methods:

• Developed a convolutional network entity model utilizing shared main parameters.
• Analyzed the model's performance in multi-position wearable sensor human activity recognition for basketball training.
• Verified effectiveness using metrics like total sensor count and single-class recognition accuracy.
• Employed Support Vector Machine (SVM) algorithm and motion simulation for validation.

Main Results:

• The proposed model effectively reduces the total number of main training parameters.
• Maintained high recognition accuracy, even with fewer sensors.
• Simulation results confirmed the model's effectiveness and the SVM algorithm's performance.
• Demonstrated the model's practical application in optimizing basketball training.

Conclusions:

• The shared parameter convolutional network entity model offers an efficient approach to human activity recognition.
• The model successfully reduces computational load while preserving recognition accuracy.
• This technology can enhance the quality and effectiveness of sports training through scientific exercise methods.