Using an information-theoretic sensor placement algorithm to assess classifier robustness

Abstract

In this paper, we use an information theoretical sensor placement algorithm to assess the impact of target camouflage, concealment, and deception (CCD) effects on classifier performance. Physics-based target models are constructed to exhibit varying CCD effects of a single target class. An information theoretical sensor placement algorithm is used to identify potential sensor locations yielding highly probable discrimination of test targets representing non-CCD and CCD targets. Platforms are positioned according to the identified sensor locations for classification processing. Classification performance results are presented and discussed in the context of the modeled CCD effect. Results demonstrate the effectiveness of the placement algorithm to identify sensor locations void of the intended CCD effects.