Nondestructive prediction of pepper seed viability using single and fusion information of hyperspectral and X-ray images

Pepper is one of the most important vegetable crops grown worldwide, and pepper fruits are consumed as vegetables and spices. Seed quality is an important factor in the production of crops, including peppers. Among the quality indicators of seeds, viability indicates whether seeds can grow into normal seedlings. In this study, hyperspectral and X-ray imaging techniques were applied to nondestructively predict the viability of pepper seeds. Various machine learning methods, including CNN-based deep learning, have been applied to evaluate single-information and information-fusion models. In addition, by modeling each criterion according to seedling conditions, it was evaluated which classification criterion can be successfully predicted. Both the X-ray and hyperspectral models showed the highest performance in classifying seed germination. The hyperspectral model showed an accuracy of 88.99% for germination prediction, which was higher than that of the X-ray model (75.33%). The ensemble-based information fusion model showed an accuracy of 92.51%, which was higher than those of single information models. Furthermore, by comparing evaluation metrics by cultivar, it was confirmed that each piece of information showed different trends depending on the condition of the seed lot.