|
Abstract:
|
Multinomial Pattern Matching (MPM) is a statistical pattern recognition algorithm that has proved highly effective for open-set classification of sensor signal and image data given limited training exemplars. Originally developed by Sandia National Laboratories, MPM is widely used in national security applications involving automatic target recognition (ATR) for synthetic aperture radar (SAR), high-range resolution (HRR) radar, and other imaging modalities. MPM enables the type-level recognition of signature data using templates, or statistical characterizations learned from multinomial-domain representations of exemplar data. It can be trained from sparse sets of collected, surrogate, or simulated target-signature data. It is computationally efficient and has been implemented on a variety of computational platforms. It provides statistically meaningful match scores that lend themselves to human and machine interpretation, analysis, and fusion with other sources of information. We present details of the algorithm and its application, including its theoretical basis and mathematical formulation, its implementation, and appropriate training procedures.
|
