Rapid detection and classification of fungi contamination in rice grains based on electronic nose with chemometrics

摘要

Electronic nose (E-nose) coupled with chemometrics was used to detect the presence of the contamination by Aspergillus spp. in rice grains. Sterilized rice grains artificially inoculated with three Aspergillus strains (A. candidus, A. fumigatus, and A. clavatus) were stored for 6 days. The colony counts offungi were measured by traditional method as reference, gas chromatography-mass spectrometer (GC-MS) and E-nose response characteristics of inoculated rice grains were collected at different storagestages (day 0, 2, 4 and 6). GC-MS results explained the feasibility of E-nose detection. Linear discriminant analysis (LDA) was used to evaluate the classification performances of three feature extraction methods (the "area values ", the "80th s values", fusion of the "area values " and the "80th s values "), the result indicated that the fusion method of the "area values" and "80th s values" had the best classification performance with the first two scores explaining 87.8 % - 97.0 % of the variations. Extreme Learning Machine (ELM) was employed for further qualitative classification and the result showed that it was proposed to classify rice grains inoculated with individual fungal species. Partial least squares regression (PLSR) and ELM were used for quantitative monitoring of fungal colony counts in rice grains. The monitoring models based on ELM reached higher correlation coefficients (R2s) and lower root mean square errors (RMSEs) than the model based on PLSR. This work indicated that E-nose combined with ELM can be an alternative approach for fungal contamination monitoring.