Classification using Efficient LU Decomposition in Sensornets

Z.H. Kamal, A. Gupta, L. Lilien, and A. Khokhar (USA)


Classification, LU decomposition, sensor networks, collaborative processing


We consider the popular application of detection, classification and tracking, and their feasibility in resource-constrained sensornets. We concentrate on classification by Maximum A Posteriori (MAP) classifier. The MAP classifier requires computation of inverse of matrices. We show how to use LU decomposition to overcome the computationally intensive and sometimes unstable inverse computation. We use a clustering and folding approach that allows LU decomposition in load balanced procedures that are computationally simpler. We present algebraic and power consumption analysis for this LU decomposition and show its feasibility in sensornets. Our analyses indicate that the investigated LU decomposition algorithms may not yet be practical for sensornets, at least the ones consisting of lightweight Mica type motes.

Important Links:

Go Back