SUMATRA: Supervised Modeling of ATR Algorithms

Automated and assisted target recognition (ATR) is an important aspect of an Intelligence Surveillance and Reconnaissance (ISR) systems. The role of ATR systems is to detect, classify, recognize and identify targets, based on input sensor data from a variety of sources. ATR system performance varies with the quality of the sensor data, the difficulty of the environment being assessed, and the separability of the target signatures in some feature space. In addition, the potency of the ATR algorithm in the extraction of feature information from the data is of importance. In order to maintain a high vehicle to operator ratio in ISR systems a key criterion is the false alarm rate. Receiver operating characteristic (ROC) curves relate the probability of detection of a target to the probability of a false alarm. The position on the ROC curve is a function of response bias which is directly related to the threshold set in the ATR detectors. The goal of this proposal is to develop a methodology for performance prediction and modeling of operator in the loop ATR systems. The proposed SUMATRA system (Supervised Modeling of ATR Algorithms) will relate physics and geometry-based vectors to false alarm rates. BENEFIT: In order to maintain a high vehicle to operator ratio in ISR systems a key criterion is the false alarm rate. Receiver operating characteristic (ROC) curves relate the probability of detection of a target to the probability of a false alarm. The position on the ROC curve is a function of response bias which is directly related to the threshold set in the ATR detectors. One way of achieving a lower false alarm rate is to lower the threshold. This shifts the operating point and lowers the false alarm rate, but it comes at the expense of a concomitant lower probability of detection. Of course this is undesirable. A better option would be to move to another curve instead which can reduce the false alarm rate without the associated drop in probability of detection. In order to achieve this, performance modeling of ATR algorithms, which can include the operator in the loop is critical. The proposed SUMATRA system will model the performance of ATR algorithms and therefore be a vary valuable tool for the Air Force. I addition, civil use in security systems, such as airports, public transportation and even closed video surveillance is anticipated.