Rapid and nondestructive method for identification of molds growth time in wheat grains based on hyperspectral imaging technology and chemometrics

Mildew on wheat is a undesired quality decrease process, producing a lot of toxic metabolites. In order to reduce economic loss and ensure the safety during wheat storage and transportation, the monitoring of mold growth in early stage is particularly important. This research was to study the different growth stages of five molds in wheat grains based on the hyperspectral imaging system. The visible-near-infrared hyperspectral information of wheat grains, with different molds growed from 1 day to 9 day, were collected to establish the model for the mold growth time discrimination. The spectroscopy in 50 wheat grains were collected for 9 days. Then, the SVM model was built with orthogonal signal correction (OSC) treated full-spectra for identification, with the coefficient of determination (R2 CV) ranged from 0.916 to 0.962, and from 0.507 to 0.748 for the root mean square error of cross-validation (RMSECV), from 0.940 to 0.964 for the determination coefficient of prediction (R2 P) and from 0.492 to 0.638 for the root mean error of prediction (RMSEP). In order to simplify the prediction models, suc-cessive projection algorithm (SPA) was employed to obtain the characteristic wavelengths. SVM models, which were better than PLS modes, were built by orthogonal signal correction (OSC)-treated characteristic wave-lengths, R2 CV and RMSECV were 0.890-0.923 and 0.716-0.862 respectively, and R2 P and RMSEP of the models were 0.882-0.932 and 0.682-0.895 respectively. It can be concluded that hyperspectral imaging tech-nology could be used to distinguish the growth time of different molds on wheat grains with high accuracy.