RAID: a neural network for rapid event identification in photon counting intensified CCDs

The design of a computational system based on a synchronous feedback neural network for the online event processing of a photon counting intensified CCD is presented. The detector head consists of a microchannel plate intensifier optically coupled to a fast read out CCD. Pixel data are sequentially fed to a layer of 3/spl times/3 neurons that is to analyse data from the CCD frame and identify photon events against spurious and/or noise events. Accepted events are then passed to a centroiding unit to determine the event centroids to sub-pixel accuracy, to be accumulated in an image memory. The performance of the system is evaluated in terms of result quality and algorithm robustness.