科研之友
微信
新浪微博
Facebook
分享链接
摘要
Fungal contamination is one of the main sanitary hazards for the cereal supply chain industry as pathogenic fungi can produce toxic secondary metabolites such as mycotoxins. The objective of the study is to apply near infrared spectroscopy to identify and discriminate Fusarium isolates, grown on solid culture medium, without preparation of the sample. This approach should allow discrimination of Fusarium species most abundant in the corn (maize): Fusarium graminearum, Fusarium proliferatum, Fusurium subglutinans, Fusarium verticillioides. The infrared spectra of 58 strains belonging to these four species Fusurium, cultured on solid media, were collected on a FOSS NIRSystem 6500 spectrometer, in reflectance, with replicates. The dimensionality of the infrared spectra was reduced by applying a principal component analysis (PCA) and the twenty first principal components were used in a discriminant factorial analysis (DFA) and as input nodes in artificial neural networks. The DFA model gave good results. The best neural networks architecture was [20-35-4] for the species discrimination. With this model, the correct classification rate on the external validation set was very good (98.8%). Near infrared spectroscopy can be a potential tool to be used in combination with the morphological diagnostic, in order to faster and validate it.
- 出版日期2010-10
