Using Features to Reduce LFA and CFLA Clutter - MP 14-13

Metron has developed a detector-tracker for Mid Frequency Active (MFA). This detector-tracker computes likelihood functions and likelihood ratio surfaces from the un-normalized matched filter output of the MFA system. While doing this we have discovered a number of features that significantly reduce false alarms. The process involves identifying a feature, characterizing its statistical behavior, and developing a likelihood ratio function based on the probability distribution of the feature"s response when a target is present to the distribution when no target is present. The power and virtue of working with likelihood ratios is that there is a principled and optimal way to combine this feature information with the likelihood ratio surface produced from the matched filter output, namely multiply the likelihood ratios together to form a cumulative likelihood ratio surface. Peaks in this new surface become candidates for detections. When combined in this fashion, the likelihood ratios from a well-constructed feature will reinforce the peaks due to targets and reduce those due to clutter. This will reduce the false alarm rate without lowering detection probability. We plan to adapt and apply this process to the LFA/CLFA tracker Metron is developing under ONR funding.