Parameter Adaptation for Target Recognition in LADAR (PATROL)

Automatic Target Recognition (ATR) algorithms are extremely sensitive to differences between the operating conditions (OCs) under which they are trained and the OCs in which the fielded algorithms operate. For ATR algorithms to robustly recognize targets while retaining low false alarm rates, they must be able to identify the conditions under which they are operating and tune their parameters on the fly. In this proposal, we present a method for tuning the parameters of a model-based ATR algorithm using estimates of the current OC. The problem has two components: 1) identifying the current OCs and 2) using that information to tune parameters to improve performance. In Phase I, we demonstrated feasibility of using reinforcement learning techniques for parameter adaptation. We first define a set of valid states describing the world (the OCs of interest, such as the expected level of obscuration). Next, ATR parameter settings for that state are learned using an off-line reinforcement learning feedback loop. The result is a lookup table to select optimal parameter settings for each OC. In Phase II, we will also develop an OC detection method using Bayesian belief networks and self-organized clustering, and a continuous alternative to state look-up tables using function approximation.