Determination of Moisture Content of Single Maize Seed by Using Long-Wave Near-Infrared Hyperspectral Imaging (LWNIR) Coupled With UVE-SPA Combination Variable Selection Method

Moisture content (MC) is one of the important factors to assess the quality of maize seeds. In this study, the feasibility of using long-wave near infrared (LWNIR) hyperspectral imaging (HSI) technique with the spectral range of 930-2548 nm for predicting MC of single maize seeds was observed. The averaged spectra extracted from whole and centroid regions in the embryo side of single maize seeds were pretreated by Savizky-Golay smoothing and first derivative (SG-D1). A combination of uninformative variable elimination (UVE) and successive projections algorithm (SPA) was proposed to select feature wavelengths (variables) from LWNIR hyperspectral data. The quantitative relationship between feature wavelengths and MC was established by partial least square (PLSR) and least square-support vector machine (LS-SVM), respectively. Results illustrated that the UVE-SPA-LS-SVM model established based on spectra of centroid region obtained the best performance for MC detection of the single maize seeds. The correlation coefficient of prediction (Rpre) and root mean square error of prediction (RMSEP) were 0.9325 and 1.2109, respectively. Finally, MC distribution of single maize seed was visualized by pseudo-color map. This study showed LWNIR HSI technique was feasible to measure MC of single maize seeds and a robust and accurate model could be established based on UVE-SPA-LS-SVM method with the spectra of centroid regions.