Target classification based on sensor fusion in multi-channel seismic network

Abstract

Target classification plays a vital role for outdoor security applications. The main focus of this paper is to describe a strategy to classify a target in a multi-channel seismic network. A technique of sensor level fusion is applied in a seismic network. This technique is based on correlation method. The method determines the weights of each seismic sensor present in the network. These weights are then adjusted adaptively as the change of correlation is observed among the sensors for real-time data. The self-clustering of the sensors is then evaluated based on the Euclidean distance measure of these weighted values in a network. This technique is not only helpful to reduce the computational cost of the network since the features of a target is extracted only from a fused signal but also to identify the failure state of the sensor. The shape statistics and peak values in a frequency domain are extracted as the features of the target. Principal component analysis is used to optimize the feature vectors. Then, the AdaBoost classifier is applied on these feature vectors for target classification.