Method for Optimizing Parameters for Detection Systems
There is tremendous interest in systems that can accurately detect radiological or nuclear threats. Many of these systems operate in extremely high throughput situations where delays caused by false alarms can have a significant negative impact. Todayís complex devices complicate the critical trade-off between detection and false alarm rates, not conforming to classical principles of statistical detection theory. This method provides optimization techniques that maximize the probability of detection for a selected number of false alarms.
Receiver operating characteristic (ROC) curves have long been used to depict the tradeoff between detection and false alarm rates. The methodology was first developed in the field of signal detection. In recent years it has been used increasingly in machine learning and data mining applications. Researchers at the Johns Hopkins Applied Physics Laboratory (JHU/APL) recognized that this methodology could be applied to threat detection systems. Given enough time and resources an estimation of all probability of detection, and false alarm (Pd, Pfa) pairs for an algorithmís parameter combinations at a reasonable granularity can be generated. These estimates can then be graphed on an ROC scatter-plot.
For algorithms with more than a handful of parameters performing this exhaustive search of the parameter space becomes virtually impossible. What is needed then is a method that overcomes these problems by empirically finding parameter values that maximize the chances of detection for a selected number of probabilities of false alarm. The JHU/APL method provides an apparatus and method to achieve this objective. This method allows the estimation of portions of an ROC curve for systems with many parameters that do not have tractable likelihood functions and do not have a one-to-one correspondence with each detection class. More particularly, this method can generate graphs which can be used to select parameters for threat detection systems by optimizing the probability of detection (Pd) with respect to several predetermined probabilities of false alarm (Pfa).
Patent Status: U.S. patent 8,494,808 issued.
Mr. E. Chalfin
Phone: (443) 778-7473