The difficulty of physical exercise promotion in community public health is not to give general recommendations, but to continuously identify residents’ exercise status and improve long-term compliance. In this paper, an intelligent monitoring and personalized intervention framework for community scenes is constructed, which integrates wearable sensor signals, mobile logs and follow-up information, and conducts multi-source modeling of residents ‘exercise behavior. After data preprocessing and feature representation, the residents’ motion portraits are further generated, and the activity types, load changes and execution trends are distinguished by combining the time series recognition and state evaluation model. The motion types, intensity, duration and reminder timing are adjusted through the dynamic feedback mechanism. The experimental results show that the proposed method achieves 93.8% Accuracy and 92.4% Macro-F1 on the behavior monitoring task, and the plan completion rate is improved to 84.9%. The duration of moderate to high intensity physical activity per week of residents is significantly increased. The results show that integrating behavior monitoring, state judgment and compliance optimization into the unified computing closed loop can more effectively support daily exercise management in community public health.
