ELSFYIER Nuclear Instruments and Methods in Physics Research A 404 (1998)445-454 NUCLEAR INSTRUMENTS 8 METHODS IN PHYSICS RESEARCH Section A Pattern recognition of particle tracks using principal component analysis and artificial neural network D. Duttaa, A.K. Mohanty”$*, R.K. Choudhury”, Phool Chandb a Nuclear Physics Division, Bhabha Atomic Research Centre, Mumbai 400085, India b Computer Division, Bhabha Atomic Research Cenhe, Mumbai 400085, India Received 29 May 1997; received in revised form 26 September 1997 Abstract A new method is suggested for pattern recognition of particle tracks based on a combined approach of both artificial neural network (ANN) and principal component analysis (PCA). It is seen that in high multiplicity environment, neither the PCA nor the ANN method is satisfactory when used separately as a track classifier. Best performance is achieved when the data are preprocessed using PCA technique, before it is fed to the backpropagated neural network. 0 1998 Elsevier Science B.V. All rights reserved. 1. Introduction Pattern recognition and track reconstruction is one of the important aspects in the detectors used in high energy heavy ion experiments. When the track multiplicities are high and the detectors are placed in the inhomogeneous magnetic fields, the track finding algorithm becomes very complex. In many situations, the tracks might also suffer mul- tiple scattering. It is, therefore, essential to develop some efficient algorithm taking into account the detector geometry, the inhomogeneity of the mag- netic field surrounding the detectors, multiple scat- tering and so on. In this work, we have tried to find out a track finding technique that will be well suited in tracking environments where one expects large track multiplicities, particularly in case of * Corresponding author. relativistic heavy ion collisions. We apply the methods to the muon tracking system of the PHENIX detector, which is being designed [1,2] for experiments with the Relativistic Heavy Ion Collider at BNL. The main purpose of the PHENIX muon arm detector is to detect muon pairs with high mass resolution originating from the vector meson decays or virtual photon produc- tion. Therefore, a highly efficient pattern recogni- tion technique is desirable as the momentum or the mass resolution of the muon pairs will depend on how well the tracks are identified from the set of measured co-ordinates. We generate simulated data for PHENIX muon arm using GEANT based simulation code PISA [3]. Conventionally, the Principal Component Analysis (PCA) method is used for pattern recognition and classification of true tracks [1,4]. It is found that the PCA method works well when the tracking distance is small, number of tracking stations are large and the track 016%9002/98/$19.000 1998 Elsevier Science B.V. All rights reserved PI1 SO168-9002(97)01139-X