Classification of animals and people based on radio-sensor network

Abstract

Personnel detection embedded in foliage is extremely important to border patrol, perimeter protection and search-and-rescue operations. In this paper, we explore the utility of radio-sensor network (RSN) to distinguish between humans and animals. We explore the phenomenon that signals are always affected by the presence of obstacles and identify human based on the received signals by transceivers, which leads to a potential low-cost way for personnel detection without specific sensors. In our study, the impulse radio ultra-wideband (IR-UWB) technology is selected for the RF transceiver due to the fact that it is not only energy efficient, but also robust against interferences. The principle component analysis (PCA) is applied to extract the feature vector, and a support vector machine is used as the target classifier. Experiment result with an average accuracy of 97.5% based on actual data collected in a cornfield indicates that this approach has a good capability to distinguish between human and animals in a foliage environment.