Modeling ATR processes to predict their performance by using invariance, robustness and self-refusal approach

A tremendous variety of types of targets, sensors, and environments is a great challenge for modern ATR systems, applications, and technologies. This makes extensive large-scale experimentation to evaluate performance of Automated Target Recognition (ATR) systems nearly impossible and motivates the proposed approach and algorithms to overcome this problem by using predictive modeling. This paper presents a methodology of ATR performance prediction called Predictive Modeling of Invariance and Robustness (PMIR). The key idea is to use parameters of ATR algorithms for predicting ATR performance. The levels of robustness and invariance of parameters are used here as predictive indicators of ATR performance along with self-refusal capabilities of the algorithms.