Evaluation and classification of five cereal fungi on culture medium using Visible/Near-Infrared (Vis/NIR) hyperspectral imaging

In order to detect and identify fungal infection in cereals timely even at its early stage of spore germination and development, a visible/near-infrared hyperspectral imaging (HSI) system with a wavelength range between 400 and 1000 nm was utilized to determine fungal growth. Five common cereal fungi, Aspergillus parasiticus, Aspergillus flavus, Aspergillus glaucus, Aspergillus niger and Penicillium sp., were selected and cultivated on Maize Agar medium individually for 6 d, HSI images were captured every 24 h for each fungus. Firstly, to classify the growth days of the five fungi, spectral characteristics were analyzed and principal component analysis (PCA) was performed, from which the growth of each fungus can be roughly divided into four growth stages, i.e., the control group-D1, D2, D3, D4-D6. Then support vector machine (SVM) model of each fungus for inoculation days were established with the first four PCs as inputs. Optimal wavelengths were then selected by successive projection algorithm (SPA) to create corresponding multispectral classification models. Overall results were satisfactory, in which accuracies of A. niger and A. glaucus were both higher than 95.87%. To further differentiate fungal species as early, the HSI images of five fungi for only one day growth were analyzed, and all five species can be distinguished well with an average accuracy of 98.89% and 0.97 for Kappa coefficient using SPA-SVM method. The results proved that VNIR hyperspectral imaging could be used to evaluate growth characteristic of cereal fungi.