모바일 환경에 적합한 수면 신호 디노이징 네트워크

With low birthrates and extended life expectancy worldwide, the healthcare industry is speeding up the development of biometric devices that monitor various biological signals using wearable devices or sensors. However, the biological signals are likely to be perturbed by noise due to the physical limitations of the device. This can result in severe degradation on the devices’ performance and even become threat to the user’s safety. Therefore, pre-processing steps of denoising on the acquired signals from biometric devices are essential. In this paper, we assume a situation where noise is added to the input signals for the sleep staging network, which is used to develop mobile devices for monitoring and staging sleep states of the user. We propose a deep learning based denoising network with noisy signals as input and use it as pre-processing stage before the sleep staging network. Experiment shows that the proposed denoising network improves decreased staging accuracy when noisy signals are used to original sleep staging network.