Advanced hyperspectral detection based on elliptically contoured distribution models and operator feedback

In autonomous hyperspectral remote sensing systems, the physical causes of false alarms are not all understood. Some arise from vagaries in sensor performance, especially in non-visible wavelengths. Consequently, many false target declarations are characterized simply as outliers, anomalies conforming to no physical or statistical models. Other false alarms arise from clutter spectra too similar to target spectra. To eliminate the recurrence of such difficult errors, deployed systems should allow operator feedback to their signal processing systems. Here we describe how a hyperspectral system using even advanced detection algorithms, based on a elliptically contoured distribution models, can be enhanced by allowing it to learn from its mistakes.