Vibration Signal Denoising Using Deep Learning

Structure vibration signals induced by footsteps are widely used for tasks like occupant identification, localization, human activity inference, structure health monitoring and so on. The vibration signals are collected as time series with amplitude values. However, the collected signals are always noisy in practice due to the influence of environmental noise, electromagnetic interference and other factors. The presence of noise affects the process of signal analysis, thus affecting the accuracy and error of the final tasks. In this paper, we mainly explore the denoising methods for footstep-induced vibration signals. We have considered different kinds of noise including stationary noises such as gaussian noises and non-stationary noises such as item-dropping vibration noise and music noises.