Recognition of events detected during nuclear research experiments

The principle of nuclear experiments is described. During an experiment, two kinds of events may arise-so called "electrons" and "jets". The target is to separate these two kinds of events as much as possible. For useful data (electrons), the "accept" signal is generated. Only these data are transferred for more detailed analysis. Each event has the form of a matrix of integers. These data are preprocessed and then classified by a neural classifier. There is need for rather high speed processing as these data arrive at a frequency of 100 kHz. A neural net with a layered architecture and step nonlinearity is described for such applications. Results showing the effectiveness of this neural net approach are presented and methods of hardware implementation are discussed.