Use of partial least squares discriminant analysis on visible‐near infrared multispectral image data to examine germination ability and germ length in spinach seeds

Abstract

Because of the difficulties in obtaining homogenous germination of spinach seeds for baby leaf production, the possibility of using partial least squares discriminant analysis (PLS‐DA) on features extracted from multispectral images of spinach seeds was investigated. The objective has been to discriminate between different seed sizes, as well as to predict germination ability and germ length. Images of 300 seeds including small, medium, and large seeds were taken, and the seeds were examined for germination ability and germ length. PLS‐DA loadings plots were used to reduce the multidimensional image features to a few important features. The PLS‐DA prediction resulted in an independent test set not only providing discrimination of seed size but also demonstrating how germination ability and germ length vary according to seed size. The result indicated that larger seeds had both a significantly higher germination potential and germ length compared with smaller seeds. The variable importance for projection method showed that the near infrared (NIR) wavelength region is important for germination predictability. However, the PLS‐DA model did not improve when only the NIR region was used. Copyright © 2012 John Wiley & Sons, Ltd.