ASABE Technical Library - Abstract

Abstract. Moisture content is an important quality parameters of wheat seed kernel, the traditional physical and chemical methods of measuring this parameter are time-consuming with complicated operations. As a rapid and nondestructive method, Visible near infrared (VIS/NIR) spectroscopy is widely used in the quality detection and classification of wheat in recent years. But it can reflect only a small part of the sample and cannot display the detail conditions, which lead to some limitations. In this study, the moisture content of wheat kernel samples was predicted by using hyperspectral technology and the visual display of moisture content of the whole sample could be realized. A linear pushbroom hyperspectral imaging system in reflectance mode was developed for prediction. And the hyperspectral images in the range of 400nm to 1100nm were collected from 45 units of wheat kernel samples. After the collection of hyperspectral data, the moisture content of the samples was determined by national standard methods. And then the Partial least squares regression (PLSR) models of moisture was developed based on the average reflective spectra data of region of interest (ROI). The models lead to good coefficients of determination (R_p=0.9762) and low standard errors estimated by validation set. The final PLSR model executed via full-wavelength was then transferred to each pixel of the image to predict moisture content in all spots of the sample. The results obtained in the research demonstrate that the hyperspectral technology has the potential to predict moisture content non-destructively in a reasonable accuracy, and the visualized results of moisture content was convenient for identification and classification of wheat.