1. More than rapid identification—Free plant identification apps can also be highly accurate
- Author
-
Michael Rzanny, Anke Bebber, Hans Christian Wittich, Alice Fritz, David Boho, Patrick Mäder, and Jana Wäldchen
- Subjects
automated identification ,botany ,Flora Incognita ,identification accuracy ,plant ID ,plant identification apps ,Human ecology. Anthropogeography ,GF1-900 ,Ecology ,QH540-549.5 - Abstract
Abstract Hart et al. (2023) conducted a study to evaluate the accuracy of five plant identification apps based on snapshot images as used in practice by field ecologists. Their results revealed varying accuracies per app, ranging from 86.9% to 46.4%. We explore the reasons why apps failed to deliver the expected result. We re‐evaluated the image dataset using another plant identification app (Flora Incognita) in order to understand the discrepancies between ground truth and app predictions. We found that mismatches between the given and returned labels can arise due to incorrect app prediction, incorrect ground truth, multiple species per image or taxonomical inconsistencies. For some images depicting early developmental plant stages, the ground truth could not be verified, resulting in some cases where both the ground truth and the app predictions could neither be confirmed nor refuted. After accounting for these aspects, Flora Incognita reached an accuracy of 98.8% on the same image dataset. Our results highlight the untapped potential of plant ID apps, as they can be highly accurate. As shown here, one area of application could be spotting misidentifications in scientific image collections, especially if multiple apps disagree with the given label. Read the free Plain Language Summary for this article on the Journal blog. more...
- Published
- 2024
- Full Text
- View/download PDF