Operator Reliance on Automated Support for Target Recognition

Abstract

In machine-aided target recognition, human operators work with an automatic target recognition (ATR) system to locate targets in cluttered and degraded imagery. The operator must integrate his or her own visual judgment concerning whether a target is present in the image with the ATR's judgment, which is typically expressed numerically. We conducted a series of experiments in which subjects attempted to locate target shapes among non-targets based only on visual images and based on both visual images and supplementary numeric information such as an ATR might provide. Image quality was controlled as an independent variable through the use of distortion rates that randomly altered pixel values to degrade the image. We found that subjects maintained a constant false alarm rate as image distortion increased, at the expense of a lower hit rate. This result was found consistently in experiments where the subjects' task was to distinguish single targets from a blank background, to distinguish single targets from single non-targets, and to locate multiple targets in a multiple-object display. We also found a bias toward over reliance on image versus numeric information. As image distortion increased, subjects failed to make optimal use of supplementary numeric information and showed an unnecessary decrease in performance. The results suggest that operators may experience difficulty in working with an ATR that has a high false alarm rate, even if the ATR's hit rate is also high, and that numeric expressions of ATR judgment may be undervalued by operators in locating targets.