Neural network based electronic nose for the classification of aromatic species

In this work, an aroma identification system has been developed. Based on an array of semiconductor tin dioxide gas sensors and neural network processing algorithms, the system has proven a 100% success rate in the discrimination of five different aromatic species. An initial nine sensor array was simplified to seven after a PCA analysis detected redundancy between three of the sensors. Data processing and classification performed by a feedforward artificial neural network with a hidden layer and trained with a backpropagation algorithm showed no significant performance differences between the complete and reduced sensor array which confirms the redundancy detected by the PCA analysis. Our results show that a reliable Electronic Nose system can be designed using inexpensive and poorly selective chemical semiconductor gas sensors.