The Inclusion of Correlation Effects in the Performance of Multiple Sensor and Classifier Systems

Classifiers from automatic target recognition systems to be fused are usually assumed to be independent. Some authors have investigated the level of covariance among the classifiers. However, much work still needs to be done to incorporate these correlation/covariance effects into the fused systems and thus its performance measures. In this paper we (1) examine and develop an expression for the covariance, and thus the correlation within the multiple sensor/classifier system, (2) to demonstrate that correlation affects the performance of the fused system. We observe that the best performance occurs when the correlation values vary