Robust mine detection and classification with target physics-derived features and classifiers

One of the most difficult challenges in shallow-water mine countermeasure (MCM) is active classification for (1) robust mine detection with an acceptable P_FA at long ranges and (2) differentiation of various mine types to facilitate an effective mine neutralization for rapid breaching capability. In this paper, we deal with the first issue of mine-versus-clutter separation over the entire 360 degree aspect coverage. Naturally, selecting key parameters or features constitutes a critical element in designing a robust active classification algorithm. To achieve good mine detection and classification (MDC) performance: (1) features must be derived from real-world events, rather than from peculiarities specific to the training data and (2) they need to be explainable in physical terms. Therefore, it is crucial that we understand and exploit acoustic wave scattering characteristics closely associated with the physical properties of mines in the feature-extraction process. We investigate how unique echo structures can be projected onto various transformation spaces so that we can capture important physical phenomena as compactly as possible. This compact representation is crucial for robust classification performance at low signal-to-noise ratio (SNR). By judiciously combining good features with an appropriate decision logic, we achieve significant improvement in both clutter reduction and mine classification performances. To address the relationship among target physics, features, and the MDC performance, we explore mid- and high-frequency echo formation processes and their role in active classification using the scaled model data