A recursive correction FDA method based on ICA combined with STAW of vinegar E-nose data

In order to eliminate the drift signal caused by the sensor aging and environmental change, and improve the robust detection capability of the electronic nose, six kinds of vinegar were used as the research object, a recursive correction Fisher discrimination analysis method based on Fast independent component analysis combined with the sample test amount window was proposed. Firstly, the electronic nose signal was decomposed by Fast independent component analysis to obtain mutually independent signal components; Secondly, according to the wavelet energy value of each independent component, the drift signal component was eliminated; Finally, after removing the drift information, the E-nose signal was reconstructed and without or with little drift information was obtained. Based on the reconstructed signal, the feature vectors were extracted and optimized, then the recursive correction Fisher discrimination analysis model was established based on the different size of sample test amount window. By comparing the correct rate of the recursive correction Fisher discrimination analysis model under different sample test amount window, the results show that when the sample test amount window was set to 100 samples, the discriminant correct rate of recursive correction Fisher discrimination analysis model was in the range of 90.67-95.83%, and the robust identification of 6 kinds of vinegar samples was realized. The results provided a new idea for the construction of long-term robust model of food using E-nose. (C) 2020 Elsevier Ltd. All rights reserved.