Real-time monitoring system of college students’ physical exercise physiological indexes based on deep learning and sensor data

The decline of college students’ physical health is a big problem. The development of wearable devices and deep learning technology provides a new path for sports physiological monitoring. In this study, a real-time monitoring system was built. Smart watches collected multimodal data such as heart rate, acceleration and skin conductance of 120 college students during exercise, and CNN-LSTM mixed model was used to classify fatigue and predict heart rate. The average absolute error of heart rate prediction on the test set is 3.2 times/min, the accuracy of fatigue classification is 88.3%, and the response time is 187 ms. There is a linear relationship between exercise intensity and heart rate (8.2 beats/min km/h), and the heart rate of overweight students is 5–8 beats/min higher. The system has been verified in physical education class, independent exercise and other scenes, and the early warning accuracy rate is 94.1%. Deep learning combined with sensor data can realize high-precision real-time physiological monitoring and provide quantitative tools for scientific exercise.