William SalterKnowledge about biodiversity is critical for nature conservation. It is important to recognize which plant species surround us in order to protect them yet being able to accurately identify plants is an increasingly uncommon skill. One solution to improve knowledge about plants is using image-based identification applications powered by artificial intelligence. Several smartphone apps are now available, including Pl@ntNet, iNaturalist and Flora Incognita, that have been developed with a scientific context that allow users to quickly identify plants around them. There is however a need to test the possibilities of these apps and establish their limitations, especially before using them for research and nature management.
In their new Editor’s Choice study published in AoBP, Pärtel et al. investigated the accuracy of Flora Incognita, a research-based tool for automated plant image identification. In their work, the authors examined several thousand images of hundreds of plant species from Northern Europe. 1496 photos from the Estonian national curated biodiversity observations database were tested with the app. In addition, Flora Incognita was directly tested under field conditions in various habitats in Estonia, taking 1703 images of plant organs as guided by the application.
Flora Incognita is a free application that was developed by the Technical University Ilmenau and the Max Planck Institute of Biogeochemistry. Flora Incognita was originally developed for the German flora, however the application is now able to identify more than 4500 vascular plant species covering the Central European flora. Depending on the difficulty of identification, the application analyses one or several smartphone photos from predefined plant organs and perspectives. These images are gradually transferred to the Flora Incognita server until the plant can be identified to species level and the result is then transferred back to the user’s device. Sometimes several taxa are suggested; rarely there are no suggestions when similarity to all species in the application’s database is low.
The results of Pärtel et al. showed Flora Incognita to be highly accurate. For the database images 79% of species were correctly identified by the app; under field conditions species identification accuracy reached 85.3%. Overall, the correct genus was found for 89 % and the correct plant family for 95 % of the species. Accuracy varied among different plant families, life forms and growth forms. Images with reproductive organs or with only the target species in focus were identified with greater success. The authors concluded that plant identification apps, such as Flora Incognita, are already usable for several purposes, especially if their capabilities and limitations are known. They can improve the situation concerning plant blindness and also foster citizen science. As is common with most deep learning applications, even more accurate identification could be expected with additional high quality training data.
Nigel Chaffey
Editor Pat Heslop-Harrison
Unfortunately, if the rest of the 4851-species list is as poorly selected as it is for Abies (in the pic above), it is going to give very unreliable results. The Abies list shown above includes two ultra-rare exotics, Abies magnifica and Abies religiosa (neither found outside of major botanical collections), and three others that are distinctly rare (Aa. balsamea, fraseri, lasiocarpa), but does not include far more commonly cultivated (and often naturalised) species Abies procera, Abies cephalonica, Abies koreana, or even the native (in the far NE of Europe) Abies sibirica.