Electronic nose and data analysis for detection of maize oil adulteration in sesame oil

An “electronic nose” has been used for the detection of adulterations of sesame oil. The system, comprising 10 metal oxide semiconductor sensors, was used to generate a pattern of the volatile compounds present in the samples. Prior to different supervised pattern recognition treatments, feature extraction techniques were employed to choose a set of optimal discriminant variables. Principal component analysis (PCA), Fisher linear transformation (FLT), stepwise linear discriminant analysis (Step-LDA), selection by Fisher weights (SFW) were used, respectively. And then, linear discriminant analysis (LDA), probabilistic neural networks (PNN), back propagation neural networks (BPNN) and general regression neural network (GRNN) were applied as pattern recognition techniques for the electronic nose. As for LDA and PNN, FLT was the most effective feature extraction method, while Step-LDA was the most effective way for BPNN and FLT was more suitable for GRNN. With only one sample misclassified in our experiment, LDA is more powerful than PNN. Excellent results were obtained in the prediction of percentage of adulteration in sesame oil by BPNN and GRNN. After training for some time, BPNN could predict the adulteration quantitatively more precisely than GRNN, whereas with FLT as its feature extraction method and without iterative training, GRNN could also yield rather acceptable results.