Improved SPRT detection using localization with application to radiation sources

We consider the problem of detecting a source with a scalar intensity inside a two-dimensional monitoring area using intensity sensor measurements in presence of a background process. The sensor measurements may be random due to the underlying nature of the source and background as well as due to sensor errors. The Sequential Probability Ratio Test (SPRT) can be used to infer detections from measurements at the individual sensors. When a network of sensors is available, these detection results may be combined using a fusion rule such as majority rule. We propose a detection method that first utilizes a robust localization method to estimate the source parameters and then employs an adaptive SPRT based on estimates to infer detection. Under Lipschitz conditions on the source and background parameters and minimum size of the packing number of state-space, we show that this method provides better performance compared to: (a) any SPRT-based single sensor detection with fixed threshold, and (b) majority and certain general fusers of SPRT-based single sensor detectors. We analyze the performance of this method for the case of detecting point radiation sources, and present simulation and testbed results.