Robust signal detection by using the EEF

In detection theory, the optimal Neyman-Pearson rule applies when the characteristics of the signal and the noise are completely known. However, in many practical scenarios such as multipath or moving targets, only partial knowledge of the signal can be obtained. In this paper, we examine the case when the alternative hypothesis has multiple candidate models, and apply the multimodal sensor integration technique based on the exponentially embedded family to detection. It is shown that our method is asymptotically optimal as it converges to the true underlying model. Furthermore, this method is computationally efficient. We also compare the proposed method with existing classifier combining rules by simulations.