Intrusion detection and classification in forest area using inter-sensor communication signals and SVM

Using wireless sensor networks (WSNs) for intrusion detection in critical area and border surveillance has been an active research topic in the last decade. In this paper, we present a novel approach for intrusion identification in the forest region. This approach is based on the ultra wideband (UWB) WSNs model. It shows the potential for exploiting off-the-shelf UWB transceivers or existing WSNs to detect the intrusion, since the proposed method only used the inter-sensor communication signals to identify the targets. For increasing the identification accuracy, we introduce a new combination of Principal Component Analysis (PCA) coefficients and channel characteristic parameters derived from received signal waveform, with support vector machine (SVM) based classifier, to classify the types of intrusion. By this method, barehanded intruder and armed intruder can be effectively distinguished. In addition, both static and moving targets are tested in the field experiment.