EMSI proposes to demonstrate the feasibility of extending our machine-learning-based SAR classifiers to provide real-time high confidence maritime target classification from ISAR data collected and processed onboard airborne ISR platforms. To this end, we will modify our classifier architectures to accommodate target motion and ISR platform constraints, simulate realistic ISAR data sets, and determine: which ISAR features are salient to ship classification and to which physical features they correspond; and what classification performance is achievable from an airborne ISR platform, as functions of ship type, environmental and radar parameters, collection geometry, and amount and fidelity of training data.
