Digit Recognition Using FMCW and UWB Radar Sensors: A Transfer Learning Approach
Jaehoon Jung, Sohee Lim, Jihye Kim, Seong-Cheol Kim
Abstract
In this article, we propose a digit recognition method for an air-writing system based on transfer learning of different types of radars and data representations. Two types of radars, the frequency-modulated continuous-wave (FMCW) and ultra-wideband (UWB) radars, are used to collect air-written digit data. The received radar signals are further processed to remove undesired clutter signals and to transform the data into a suitable format. Two data representations, the range-time map and the Doppler-time map, which represent the variation in the range and Doppler over time, are considered. These data representations are used as inputs to a convolutional neural network (CNN) to learn the nonlinear characteristics of each digit data and classify the radar signal images. The classification results demonstrated that the proposed digit recognition method can identify air-written digits from 0 to 9 with an accuracy greater than 98%. Compared to the standard CNN method, the proposed transfer learning method showed superior performance in distinguishing each digit data, even for training datasets with limited size.