Radio-frequency (RF) sensing for deep awareness of human physical status

Abstract

With the advancement of technology, smart devices which can emit Radio Frequency (RF) signals are all around us. The purpose of the project is to improve upon RF sensing of human vital signs by using Deep Learning techniques. The dataset used for this project is collected from a XeThru X4 module connected to a Raspberry Pi while the ground truth is collected using a Neulog Respiration Monitor Belt logger sensor. The dataset is then trained on a proposed Convolutional Neural Network (CNN) model with the Swish activation function. Data augmentation is then performed on the dataset to further improve results. The best performing model achieves a validation loss of 0.38. Further efforts can be put into diversifying the dataset and combining other deep learning models such as Long Short Term Memory (LSTM) with CNN. Acknowledgement