Abstract
Pl@ntNet is a free web and mobile platform dedicated to automated, image-based plant identification and to collaborative gathering of plant observations (http://identify.plantnet-project.org/). It relies on crowdsourcing approaches and machine learning techniques for data production, validation and enrichment. The initial version of the application, launched in 2013, covered 800 French native species. It now covers a large part of European flora (6,200 species) and has been extended to other floristic regions, such as the Mascarene Islands, the Guiana Shield and Maghreb. Through its iPhone and Android apps (> 3 million downloads and 10,000-50,000 daily users), Pl@ntNet gathers increasingly large amounts of botanical observations voluntarily contributed by an array of people who are often novice in plant identification. These observations are continually checked and amended (for identifications and image quality) by hundreds of amateur botanists, through Pl@ntNet’s collaborative web tools. We recently worked on several different datasets shared by national and international institutions (such as a visual dataset from Encyclopedia of Life), in the aim to improve efficiency and taxonomic coverage of Pl@ntNet application. This has allowed the adaptation of Pl@ntNet to several new floras, such as the North American flora, a part of the Caribbean and Hawaiian flora, and the Tropical Andes flora. We propose to present (i) recent developments dedicated to data aggregation and enrichment (notably the semi-automated plant image annotation), (ii) the limits of this approach, (iii) as well as the perspectives of improvements, based on both the users feedback, and on analyses of the data already collected. This emphasize on one side, the potential of new technologies for botanical and ecological activities, and on the other side, the capacity of multi-disciplinary projects to address societal needs at large scale