With the deep integration of Internet of Things (IoT) technology and wearable devices, the field of sports training is undergoing a gradual transformation toward intelligent and data-driven paradigms. This study aims to enhance core football skills. It addresses key limitations of traditional training methods, namely insufficient action quantification, delayed feedback, and lack of personalized plans. This study presents a smart training framework based on multi-modal IoT sensing and adaptive algorithms. A network of wearable devices integrates high-precision Inertial Measurement Units (IMU), flexible pressure sensors, and bioelectric signal collection modules. These devices capture real-time data on athletes’ limb movements, ball contact mechanics, and physiological states. An improved Spatio-Temporal Graph Convolutional Network++ (ST-GCN++) and Long Short-Term Memory (LSTM) network are integrated to construct a dynamic action recognition model. This model enables multi-dimensional quantitative analysis of core skills like passing accuracy, shooting power, and agility in direction changes. An innovative Reinforcement Learning-Adaptive Training Strategy (RL-ATS) algorithm is developed and integrated into the framework. This algorithm dynamically optimizes training intensity and motion combinations, thus accommodating individual differences and facilitating adaptive training load regulation. A low-latency edge computing architecture ensures real-time data processing and feedback. To validate the model’s effectiveness, 48 amateur football players were recruited for a 12-week controlled experiment. The experimental group uses the proposed system, while the control group follows traditional methods. Results show that the experimental group’s passing accuracy improved by 19.3% (from 72.1% to 91.4%), the standard deviation of shooting speed decreased by 28.5% (indicating enhanced action stability), 30-meter sprint times improved by an average of 0.32 s, and muscle fatigue recovery cycles shortened by 15%. Comparing ST-GCN + + with classic Convolutional Neural Network (CNN) models, the system achieves 94.7% accuracy in complex action classification tasks (higher than CNN’s 86.2%), with a false detection rate reduced to 3.1%. Additionally, the RL-ATS algorithm reduces training plan adjustment response times from hours to under 5 min, with an 89% adoption rate for personalized plans. This study realizes the systematic integration of IoT wearable devices and adaptive algorithms, thereby bridging the “data-decision” gap in traditional training. This study further establishes a quantifiable, reusable technical framework for enhancing football skill development.
Ma et al. (Fri,) studied this question.