Fusing heterogeneous data for detection under non-stationary dependence

In this paper, we consider the problem of detection for dependent, non-stationary signals where the non-stationarity is encoded in the dependence structure. We employ copula theory, which allows for a general parametric characterization of the joint distribution of sensor observations and, hence, allows for a more general description of inter-sensor dependence. We design a copula-based detector using the Neyman-Pearson framework. Our approach involves a sample-wise copula selection scheme, which for a simple hypothesis test, is proved to perform better than previously used single copula selection schemes. We demonstrate the utility of our copula-based approach on simulated data, and also for outdoor sensor data collected by the Army Research Laboratory at the US southwest border.