Neural network based recognition of flow injection patterns

Abstract

Response patterns produced from a flow injection (FI) system employing sodium, potassium and calcium ion-selective electrodes as detectors were used for training and testing back-propagation neural networks. A variety of different network parameters were investigated including a study of the effects of variation of the learning rate coefficient and the momentum on the rate of training. The networks studied demonstrated the ability to recognize the identity of sample components even in FI patterns severely distorted by noise addition, peak height variation and baseline shift.