Non-Contact Human Activity Classification using DCNN based on UWB Radar

Deep convolutional neural networks (DCNN) is applied in non-contact human activity classification based on ultra-wideband (UWB) radar system. A weighted time-range-frequency transform (WRTFT) method was used to get the spectrograms combining time, range and frequency information from human activity signals. Then DCNN is utilized to extract features and classification boundaries from spectrograms. Extensive experiments were conducted to compare the classification performance between the physical empirical feature method and DCNN method. DCNN method can achieve a 92.8% classification accuracy for classifying six typical human activities and show a good robustness facing individual diversity.