Drift Elimination Method Of Electronic Nose And Its Application In Food Detection

Electronic nose(in short E-nose) is a very potential identification tool which has been widely applied in the field of food detection. But so far, it has been still in the laboratory stage, and has a gap in practicability. The main reason is that the drift phenomenon exists in E-nose signal. In order to reduce the influence of the E-nose drift, the drift was respectively studied by two methods in the thesis. On the one hand, according to Fourier Transform, the E-nose signal is transformed into the frequency domain signal, and the drift of E-nose was eliminated or reduced by constructing a filter function. On the other hand, with help of multivariate statistical analysis method, the drift signal was separated and removed from E-nose data. At the same time, for testing the availability of the proposed methods, six kinds of spirit and six kinds of vinegar were respectively selected as identification objects.Firstly, a kind of elimination method based on Fast Fourier transform(FFT) was putting forward. The method is that a threshold filter function is given, and the FFT coefficients of E-nose signals are updated by the filter function; then the updated coefficients are reconstructed by inverse FFT, so the E-nose drift will be reduced or eliminated. When the integral value of E-nose signal was extracted, the correct identification rates of Fisher discriminant analysis(FDA) to spirit and vinegar samples are respectively from 38.6%, 76.8%(non-reconstruction signals) up to 99.9%, 100%(reconstruction signals), and the cross-validation rates were also from 35.0% and 74.3% up to 99.9% and 100%. The results show clearly that the elimination method is very effective.Secondly, a drift elimination/compensation method based on independent component analysis(ICA) coupled to wavelet energy threshold was also puts forward. The first is that the E-nose signals were decomposed to individual components using ICA. The second is that some independent components were selected by their wavelet energy with the help of wavelet analysis so as to eliminate those independent components correspond to drift signals. The third is that the selected independent components were reconstructed to E-nose signals again. When the integral values were extracted from before and after drift elimination for E-nose signal to two kinds of samples, the FDA correct identification rates of white spirit and vinegar samples were respectively from 38.6% and 75.7% up to 100% and 99.7%, and the cross-validation rates were also from 35.0% and 74.3% up to 100% and 98.6%, respectively. The FDA results clearly show the proposed method is very effective. Compared with other ICA algorithm, the proposed method does not require prior information, so it is very simple method and more suitable for practical application. The study result of this thesis is not only suitable in the identification of white wine and vinegar samples, but also provides a reference for other types of sample identification.