Substance classification and measure for low-cost electronic noses

In this paper, a new approach to classify and quantify substances is presented to be suitable for low-cost electronic noses. An appropriate architecture based on multi-layer perceptron neural networks is proposed to shorten training set and improve accuracy if a substance is clearly detected. Elaboration is suitable to be implemented in an eight-bit microcontroller due to its simplicity. Experimental results, reported in presence of mixture of ethanol and methanol, shows a classification error within 10% and a quantification error in the order of 10% of full scale