An Efficient Machine Learning Algorithm For Spatial Tracking Of Correlated Signals In Wireless Sensor Field

Abstract

An efficient machine learning algorithm based on stochastic gradient is proposed and discussed for spatial tracking of correlated spatial signals from the sensor observations in wireless sensor field. The proposed algorithm can be used for environmental monitoring applications such as efficient temporal monitoring of temperature in hot island, or efficient monitoring of the distribution of pollutant gasses in wide areas for example terrain of large cities, etc. The proposed algorithm is computationally efficient and is low cost. In this paper the number of reporting sensors in tracking of signal is defined as cost. The spatial signal is compressed into a number of its isocontours at specific levels and the sensors whose sensor readings are in given margin of these contour levels, report their sensor readings to the fusion center (FC). The algorithm is done in two phases of spatial modeling and spatial tracking, where it uses the correlation between the spatial signal before and after variation of the spatial signal and updates the new set of contour levels for spatial tracking. The proposed machine learning algorithm finds the modeling parameters during the spatial modeling phase, and then updates them in spatial tracking phase. The performance analysis of the proposed algorithm shows that it successfully tracks the spatial variations of the signal at low cost with similar modeling performance to the spatial modeling.