Application of random-point processes to the detection of radiation sources

In this report the mathematical theory of random-point processes is reviewed and it is shown how use of the theory can obtain optimal solutions to the problem of detecting radiation sources. As noted, the theory also applies to image processing in low-light-level or low-count-rate situations. Paralleling Snyder's work, the theory is extended to the multichannel case of a continuous, two-dimensional (2-D), energy-time space. This extension essentially involves showing that the data are doubly stochastic Poisson (DSP) point processes in energy as well as time. Further, a new 2-D recursive formulation is presented for the radiation-detection problem with large computational savings over nonrecursive techniques when the number of channels is large (greater than or equal to 30). Finally, some adaptive strategies for on-line ''learning'' of unknown, time-varying signal and background-intensity parameters and statistics are present and discussed. These adaptive procedures apply when a complete statistical description is not available a priori