Incremental PNN classifier for a versatile electronic nose

Abstract

Due to robustness of the probabilistic neural network (PNN) architecture, it has been widely used for pattern classification tasks. Commonly used PNN algorithms are not capable of incremental learning. The classifiers having the incremental learning ability can be of great benefit by automatically including the newly presented patterns in the training dataset without affecting class integrity of the previously trained classifier. This signifies that, the incremental classifiers have the ability to accommodate new classes and new knowledge within an already trained model. Under the present study, an electronic nose anchored aroma characterization model based on PNN classification strategy has been developed whereby the sensor array outputs of the electronic nose can be co-related to the sensory panel (tea tasters) quality scores for black tea. The whole study has been done in few tea gardens in north-east India. In pursuit of development of optimal strategy for data collection from dispersed locations followed by dynamically augmenting the training data corpus of the already trained PNN model, the incremental leaning mechanism has bee suitably grafted to the PNN model to have efficient co-relation of electronic nose signature with tea tasterspsila scores. The incremental PNN classifier promises to be a versatile pattern classification algorithm for black tea grade discrimination using electronic nose system.