Abstract

Many economically important plants produce starch grains that, if distinctive in form, can be used as identifiers for particular taxa. The identification of starch to species or genera has become increasingly important in studies exploring plant use in ancient societies and also in the verification of plant origin for some plant-based medicines. However, identification of starch can be problematic, because of the considerable variability in the morphology of starch grains. As a result there has always been an element of subjective judgement when it comes to identifying a sample of grains. Here we present a novel system for identifying the plant species origin of unknown starch grains using image analysis of light micrographs. After manually obtaining a mask of the two-dimensional maximum-projection-area grain shape, features for each starch grain were determined automatically including the size metrics, circularity and Fourier transform signature. The starch grain features analysed were used to create classifiers for the grains. The relative performance of the different classifiers was evaluated, based on different combinations of the predictor variables (e.g. area, perimeter etc.), and the optimal classifier determined. The method was applied to a database of 1032 grains representing 8 geographically co-located known economic plant species. A classification tree using shape metrics and the Fourier signature produced the best separations. The morphological features were sufficient to obtain a high level of accuracy in attributing individual starch grains to plant species. The method enables the creation of effective classifiers to undertake a quantitative evaluation of starch grain morphologies, thereby reducing the need for subjective qualitative determinations. The system provides a robust framework in which plant microfossils of unknown species origin can be compared with reference grains for effecting identifications. The method is potentially useful not just for starch, but other microfossils of morphometric interest.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call