E-CONDOR: Efficient Contour-Based Detection Of Random Spatial Signals From UAV Observations Using Dual Stochastic Gradient

Abstract

This paper presents a novel efficient method for spatial monitoring of the distribution of correlated field signals, such as temperature, humidity, etc. using unmanned aerial vehicles (UAVs). The spatial signal is compressed to its iso-contour lines at a number of known levels that are introduced by data fusion center (DFC). The UAV traces a contour line of the field signal at a time, and reports the coordinates of its own traces to the DFC for spatial modeling. The DFC iteratively improves the spatial model of the field signal and assigns a new contour level to each UAV to trace and report its coordinates for spatial model improvement. The selected batch of levels and the start point of the search are introduced by the DFC. In order to reduce the required data for spatial modeling, and accordingly improve the algorithm data efficiency, dual stochastic gradient routines are used at the DFC to find a next proper number of contour levels in the batch, and to eliminate the redundant contour levels, in each iteration. The performance evaluation of the proposed algorithm based on computer simulations demonstrates significantly faster convergence, better signal estimation, and a higher data efficiency against when the stochastic gradient is not used.