Human Activity Classification With Transmission and Reflection Coefficients of On-Body Antennas Through Deep Convolutional Neural Networks

Abstract

We propose to classify human activities based on transmission coefficient (S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</sub> ) and reflection coefficient (S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11</sub> ) of on-body antennas with deep convolutional neural networks (DCNNs). It is shown that spectrograms of S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</sub> and S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11</sub> exhibit unique time-varying signatures for different body motion activities that can be used for classification purposes. DCNN, a deep learning approach, is applied to spectrograms to learn the necessary features and classification boundaries. It is found that DCNN can achieve classification accuracies of 98.8% using S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</sub> and 97.1% using S <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11</sub> . The effects of operating frequency and antenna location on the accuracy have been investigated.