Integration of multiple adaptive algorithms for parallel decision fusion

The Chair-Varshney rule for parallel binary decision fusion requires knowledge of the a priori probabilities of the hypotheses and the performance of the sensors (probabilities of false alarm and missed detection). In most applications, this information is not available. Five methods were developed so far for estimating the unknown probabilities. However, none of them is the best under all circumstances. We present an algorithm that selects the best of these five methods. The algorithm estimates roughly the value of the a priori probabilities and the sensor performance from input data, and seeks support from a data base that provides archival data from the five methods at this operating point. In simulation, the algorithm performed on average better than each one of the five existing methods operating alone.