A neural net implementation of SPCA pre-processor for gas/odor classification using the responses of thick film gas sensor array

Abstract

In this paper, an artificial neural net (ANN) implementation of SPCA pre-processing is presented for its use with a neural classifier trained with SPCA transformed data. Here, a SPCA transforming neural stage (Net I SPCA) is placed before a SPCA trained neural classifier stage (Net IISPCA). Accordingly, newer sensor array response of respective gas/odor can now be classified, more precisely, using Net IISPCA fed through (Net ISPCA). This way, newer sensor response gets transformed to corresponding SPCA transformation, with conformity, for its classification using Net IISPCA. Efficacy of this scheme is demonstrated by considering thick film tin oxide sensor array response data for four gases/odors (viz. acetone, carbon tetra-chloride, ethyl methyl ketone and xylene). Experimentally, respective nets, Net ISPCA and Net IISPCA comprising of 11 and 04 neurons were, respectively, trained in just 78 and 12 epochs of 42 × 4 training vectors for the aforesaid gases/odors. The SPCA transformation derived mathematically and obtained through Net ISPCA, carried a mean squared error (MSE) of 1.81 × 10-9 for 18 test sensor responses not used for training Net ISPCA. As well as, 100% correct classification was achieved for the aforesaid 18 independent samples using Net IISPCA with a MSE of 1.22 × 10-9. © 2010 Elsevier B.V.