The work aims to propose a sleep staging method based on a radial basis function (RBF) neural network and use it to design an intelligent wake-up system that can adjust the wake-up time according to the user's recovery state, in order to optimize the user's sleep duration and reduce their discomfort after waking up. Based on theoretical knowledge of heart rate variability and sleep staging, the electrocardiogram (ECG) signal was collected from the human body through a low-power heart rate band, and the optimal wavelet transform was selected to precisely denoise the collected ECG signal. The radial basis function (RBF) neural network was trained repeatedly to filter out 10 key feature vectors, so as to build a model of sleep staging. The sleep staging information was transmitted to the mobile phone client via a STM32 processor, and the system woke up the user according to designed optimized wake-up mechanism when the user's body and mind recovered to the optimal state. The results showed that the algorithm based on the sleep staging model had an average accuracy of 88.9% with a Kappa coefficient of 0.839, which was higher than that of other algorithms. The intelligent wake-up system has a lower collection cost, a simpler and more efficient algorithm, and a scientific and reasonable wake-up mechanism, which enables the user to wake up comfortably and is of great significance in improving the user's state after awakening.