AI-Driven Personalized Physical Education and Improvement of Key Physical Fitness Characteristics

This study proposes an AI-driven adaptive framework for personalized physical training based on multimodal sensing, real-time signal fusion, and hierarchical decision optimization. A heterogeneous sensing architecture integrating smartphone vision sensors and wearable physiological devices is developed to acquire synchronized kinematic and physiological information. A multimodal state perception layer combines motion-derived features, physiological measurements, and temporal synchronization mechanisms to generate unified state representations for continuous individual monitoring. To characterize long-term behavioral evolution, an incremental clustering strategy is introduced to dynamically update user profiles, while a weighted knowledge graph establishes quantitative relationships between physical fitness attributes and training actions. Furthermore, a hierarchical reinforcement learning framework is constructed, where a deep Q-network regulates training-load allocation and a proximal policy optimization network generates adaptive action sequences. An edge-deployable feedback module incorporating real-time motion analysis and augmented-reality-assisted correction is integrated to form a closed-loop sensing and control architecture. Experimental evaluation involving 1,286 students over a 12-week intervention period demonstrates an average improvement of 12.3 points in key physical fitness indicators, while the individual adaptability index increases from 0.732 to 0.901. The proposed framework establishes a unified methodology for multimodal signal acquisition, adaptive information processing, and closed-loop feedback optimization, providing an engineering-oriented solution for large-scale intelligent sensing and human-centered monitoring systems.