Hyperspectral Image-Based Variety Discrimination of Maize Seeds by Using a Multi-Model Strategy Coupled with Unsupervised Joint Skewness-Based Wavelength Selection Algorithm

Discrimination of seed varieties is important in agricultural planting and plant breeding. This research aims to develop a rapid and highly accurate model for differentiating maize seeds through hyperspectral imaging. Hyperspectral reflectance images between 924 and 1657 nm were acquired for 1120 maize seeds from 14 varieties. The mean spectra were extracted from the region of interest on the hyperspectral reflectance image. Unsupervised joint skewness-based wavelength selection algorithm (JSWSA) was then used to select the optimal wavelengths. Finally, a multi-model strategy was developed based on least square support vector machine for variety discrimination. Experimental results showed that the multi-model for full-wavelength data achieved 98.18 % classification accuracy for the test set, which is higher than the 96.36 % classification accuracy of the single model. The multi-model also elicited 96.57 % classification accuracy, with an improvement of 4.96 %, relative to that of the single model when using 19 optimal wavelengths (only 8.68 % full wavelengths) selected by JSWSA. This study shows that the multi-model coupled with JSWSA exhibits high potential in rapid and highly accurate classification of seed varieties.