Hand Gesture Recognition Using Input Impedance Variation of Two Antennas with Transfer Learning

This paper investigates the possibility of classifying hand gestures using the impedance variation of two antennas through transfer learning. When a hand gesture is made near an antenna, the impedance of the antenna varies over time due to the near-field perturbation. By recognizing the pattern of impedance variation, the gesture can be identified. We proposed to use two monopole antennas to measure signatures at different locations. A network analyzer measured the input impedance of the two antennas for ten hand gestures. The impedances were collected at 2.4 GHz and then were transformed to spectrograms to capture the time-varying signatures as imagery. The spectrogram images were classified using deep convolutional neural networks. In particular, we employed transfer learning to overcome the issue of the small data set available. Pre-trained networks, such as AlexNet and VGG-16 that were trained for general optical images were able to maximize the classification accuracy of our hand gesture recognition problem.