Processing CsI(Tl) 2D matrices by means of neural networks and Markov random fields

The present work is concerned with the automatic analysis of data coming from the multi-detector array Chimera, used in nuclear physics at intermediate energies. Each of Chimera's detection cells is a telescope made of a /spl Delta/E silicon detector and a CsI(Tl) crystal, thick enough to stop all the charged light particles. The signal produced in the CsI(Tl) scintillators can be subdivided into two components, fast and slow. These data are collected in form of bi-dimensional matrices (Fast-Slow matrices), particularly important for light particle identification. The proposed approach consists in applying image processing techniques, and in particular Grossberg's (1973) pre-attentive neural networks as first step, in order to isolate the regions of physical interest in the matrices and to roughly identify the directions depicted by the most intense lines, and then a successive step of filtering, based on Markov random fields.