Identification of Myosotis seeds by means of digital image analysis

Abstract

Despite their name, the identification of seeds of Myosotis species (forget-me-not) has hitherto received little attention from archaeobotanists. In an attempt to assemble a collection of reliable identification criteria, digital image analysis was applied to photographs of Myosotis seeds by means of Fovea Pro 4.0. This program computes 23 features that describe the size and shape of the seeds shown in scale-normalized photographs. We computed the features for 1,453 individual seeds, and performed statistical analyses of the resulting data set with Discriminant Analysis, Correspondence Analysis, and t-Distributed Stochastic Neighbour Embedding (t-SNE). The combination of analyses provides clues as to how most of the seven western European species of Myosotis can successfully be distinguished. Using these clues, an identification key was developed for the identification of waterlogged Myosotis seeds.