Modeling Parameters for Target Identification: A Critical Features Analysis

摘要

For decades, the modeling and prediction of human target identification hasrelied on physical parameters generalized over the whole target, such as physical size, range to the sensor, and apparent thermal contrast, defined as six cycles on the target in the ACQUIRE model. Identification performance for targets that meet this criteria are on average accurate. However, variation in the identifiability of objects meeting the ACQUIRE criterion is so wide that it suggests that some other factor, something bound up in the way people perceive and identify objects, is also influencing identification performance. The evidence for this is that some targets are easier to identify than the model would predict, while others are much more difficult; some aspect angles are more difficult, while others are more readily identified. Many perception experiments which had been performed for sensor design yielded general results of value to system design parameters but resistant variance remained in the results to baffle those demanding definitive results for specific target configurations. For the last few years, NVESD has embarked upon a strategy of understanding the human perception of thermal imagery from the standpoint of neuroscience theory, the most prominent of which are Recognition-by-Components (Biederman, 1987) and computational vision (e.g., Wilson, 1995; Fiser, et al., 1995; Lades et al., 1993).