Zoo/PhytoImage version 1

Planktonic organisms constitute the base of many aquatic food webs. Indeed zooplankton is the key mediator between energy synthesized by phytoplankton and higher trophic levels. Because plankton can vary quickly in term of abundance and biomass according to variations of environmental conditions, it constitutes a significant bio-indicator of global changes like increasing of atmospheric CO 2 , global warming or anthropogenic eutrophication. The taxonomic composition, the distribution and the abundance of planktonic groups are thus fundamental parameters of ecosystem structures and mechanisms and can be used to understand and to quantify the contribution of these planktonic organisms to these processes. To understand these phenomena and because of the heterogeneity of plankton distribution it is necessary to increase spatial and temporal resolution of sampling method. In most of the case, the amount of samples can be quickly important. Unfortunately, the analysis of plankton samples is traditionally associated with long sessions of counting planktonic organisms under the binocular with formaldehyde vapours floating around. This limitation increases the time of samples treatment and constitutes the bottleneck of planktonology. Moreover, taxonomists are increasingly rare. Although this picture of a planktonologist will probably remain for a while, it seems to be another way to gather data about plankton: computer-assisted analysis of plankton digital images. This alternative and complementary method is considered since the 80's but was limited by the quality and the resolution of existing devices. Now, the combination of current powerful computers with the quality of digitalized devices and the efficiency of machine learning algorithms provide a potent tool to biologists and planktonologists. The computer-assisted analysis of plankton can provide rapid enumeration and identification of plankton samples. In order to help planktonologists in their works, a whole suite of hardwares to take pictures of plankton, both in situ and/or from fixed samples, the use of a digital camera on top of a binocular or with a macro lens. Digital images of plankton are barely usable as such; they must be analyzed in a way that biologically and ecologically meaningful features are extracted from the pixels. But, all images digitalized by these hardwares need the intervention of taxonomists for plankton identification and classification. The machine learning method uses extracted features of some representative particles to automatically identify and classify all particles images in different taxonomical / ecological groups. A software doing such an analysis is thus an essential phase in digital image processing. From this …