Optimization of automatic target recognition with a reject option using fusion and correlated sensor data

• Authors: Kenneth W. Bauer, Jr.;Trevor I. Laine
• Venue: Optimization of automatic target recognition with a reject option using fusion and correlated sensor data
• Year: 2005

In many pattern recognition applications, significant costs can be associated with various decision options. Often, a minimum acceptable level of confidence is required prior to making an actionable decision. Combat target identification (CID) is one example where the incorrect labeling of Targets and Non-targets has substantial costs; yet, these costs may be difficult to quantify. One way to increase decision confidence is through fusion of data from multiple sources or from multiple looks through time. Numerous methods have been published to determine optimal rules for the fusion of decision labels or to determine the Bayes' optimal decision if prior probabilities along with decision costs can be accurately estimated. This research introduces a mathematical framework to optimize multiple decision thresholds subject to a decision maker's preferences. The decision variables may include rejection thresholds to specify Non-declaration regions and ROC thresholds to explore viable true positive and false positive Target classification rates. This methodology yields an optimal class declaration rule subject to decision maker preferences without using explicit costs associated with each type of decision. This optimization framework is demonstrated using various generated and collected sensor data. The experiments using generated data were performed to gain insight of the potential effects of fusing data with various degrees of correlation. The optimization framework is then applied to assess two competing fusion systems across four test sets of radar data. The fusion methods include Boolean logic and probabilistic neural networks for the fusion of collected 2-D SAR data processed via 1-D HRR moving target algorithms. Excursions are performed by varying the prior probabilities of Targets and Non-targets and varying the correlation between multiple sensor looks. In addition to optimizing thresholds according to decision maker preferences, an objective function is presented to facilitate comparison between CID systems, where the time associated with each look is incorporated.