[en] This chapter illustrates supervised classification of objects based on features measured on digital images, as it works in R using the zooimage and mlearning packages. It mainly focuses on plankton images, but it is also usable in different but similar contexts. Plankton is a diverse community of organisms that live in aquatic environments far away from hard substrate. Its diversity and the patchiness in its distribution, both in time and space, make it difficult to sample and to study. Automated classification of plankton digital images with machine learning algorithms in R is used since a few years to speed up the process of the large amounts of samples typically encountered in oceanographic campaign. Through the analysis of an example dataset of tropical zooplankton from Madagascar, we show how the zooimage R package, and the Zoo/PhytoImage software, contributed to the adoption of R for such a task. More particularly, we insist on the integration of data mining tools inside a larger workflow, from the processing of raw images to the calculation of derived statistics usable by the ecologist. Challenges and difficulties associated with this complex multi-class supervised classification application are also discussed.
