The app helps identify plants, but only up to a point
Marco Gaber/Digital Vision/Getty Images
Smartphone apps that identify plants from photos are only 4% accurate, can endanger people in search of food, and can even eradicate endangered plants by misclassifying them as weeds. I have.
Julie Peacock of the University of Leeds (UK) and her colleagues evaluated the six most popular apps: Google Lens, iNaturalist, Leaf Snap, Pl@ntNet, Plant Snap and Seek. They sought to identify 38 species of plants in their natural habitat at her four locations in Ireland, per app. The team found that some apps scored very low, and even the best apps did not reach 90% accuracy.
“There are many reasons why it is important that the apps are accurate or that people are aware that even though these apps are guides, they are not complete,” says Peacock. For example, people may misidentify important native species as invasive and remove them from their gardens, or may consume potentially dangerous wild plants as harmless varieties.
But Peacock doesn’t think people shouldn’t use these apps as long as they understand the limitations. say.
These apps use artificial intelligence algorithms trained on a huge number of captioned plant photos. During training, the AI learns not only to recognize training photos, but also to find similarities between them and new photos, enabling it to identify plants.
In general, all apps were better at identifying flowers than leaves. The researchers say this is due to the diversity of flower shapes and colors that provide more cues to the AI. However, this was not always the case. The iNaturalist app was able to correctly identify 3.6% of flowers and 6.8% of leaves. Plant Snap correctly identified 35.7% of flowers and 17.1% of leaves. The best accuracy was achieved by Pl@ntNet with 88.2%.
Alexis Joly of Inria, Montpellier, France, is one of the researchers behind the non-profit project Pl@ntNet, whose success is contributed by botanists, scientists and informed amateurs. , states that it rests on a classified data set. Along with an algorithm that tries to balance the bias towards common species and instead rank several possible candidates per search.
“People like to see one result with 100% confidence, even if it’s not the correct one, so instead of showing three species with 33% chances, each with 33%. It’s a thankless job at times because it represents a reality about photography, without it being taken,” he says. “But our strategy seems to be working.”
Stephen Harris of the University of Oxford says Peacock’s concerns are legitimate, saying he too has experienced problems with such apps and instead relies on good reference books. It’s because uploaded images are often mislabeled, he says.
“People tend to image similar things. If you do get a plant, but it happens to be a silly little thing that doesn’t have very attractive flowers or anything like that, Harris said, “People scramble in the pond, or raking weeds in ponds or taking pictures.”
Google declined an interview request, but the other app creator didn’t respond.
topic:
