Missouri Botanical Garden Open Conference Systems, TDWG 2013 ANNUAL CONFERENCE

Font Size: 
Collaborative validation of visual data through the Pl@ntNet identification system
Samuel Dufour-Kowalski, Jeremy Salinier, Aurelien Peronnet, Jennifer Carré, Jean-Pascal Milcent, Hervé Goëau, Alexis Joly, Nozha Boujemaa, Pierre Bonnet, Julien Barbe, Jean-Francois Molino, Daniel Barthélémy

Building: Grand Hotel Mediterraneo
Room: America del Nord (Theatre I)
Date: 2013-10-28 03:10 PM – 03:30 PM
Last modified: 2013-10-05

Abstract


Plant identification plays a crucial role for a multitude of context (eg. Agronomy, Ecology, Forestry, Pharmacology, …). As part of the Pl@ntNet project, innovative approach based on picture indexation and comparison[1] has been developed. This identification system is available on the web and on mobile devices allowing direct identification on the field.

To improve the efficiency of such a system, the quality of the underlying database plays a crucial role. The more the database contains good pictures, for different species, at different stages of growing, the more the system is able to give accurate and relevant results.

Because it is not possible to create or to collect such a collection of images on our own, we organised a whole collaborative process, developing different applications to collect and validate together botanical observations. Each time a user submits a query to the system from its mobile, he can choose to contribute to the project. In this case, the images are stored and are reviewed through two collaborative validation applications. Once the data has been validated, they can be reused in the identification system. The more the people contribute, the more the user experience is enhanced.

The collaborative review of the data consists of two steps, one for the general image quality (PictoFlora) and one other for the accuracy of the taxonomic identification (IdentiPlante). PictoFlora offers an image gallery, where end-users can vote on the image quality and then give a score on different criteria. IdentiPlante is a botanical observation gallery dedicated to plant determination. Botanical records aggregated by the system can be with an incorrect or partial identification (i.e. at the family level), or even without any identification. IdentiPlante is then providing an easy way to suggest a new determination if necessary, to vote on propositions, or to discuss about identification of each botanical record. Once a record has collected enough votes and reached a minimal score, it can be reinjected in the identification system.

The Pl@ntNet-Mobile application is widely distributed today and reaches a very large public (65000 download in 6 months). Data quality becomes a major issue. Thanks to an active community, contributions can then be quickly reviewed. From its launch last spring 2013, PictoFlora has allowed to collect 55 000 pictures votes, and IdentiPlante 6200 identifications suggestions. This system creates a virtuous circle in data collection and validation, and provides an innovative approach to collect high quality biodiversity data.

 

[1] A. Joly, H. Goeau, P. Bonnet, V. Bakic, J. Barbe, S. Selmi, I. Yahiaoui, J. Carre, E. Mouysset, J.-F. Molino, N. Boujemaa, D. Barthelemy, 2013. "Interactive plant identification based on social image data". Ecological Informatics journal.