Abstract
Background
Over the past 5 or so years, many smartphone ‘apps’ for the plant identification (ID) have become available for use by amateur and professional botanists. There is, however, a need for good information on their reliability in practice.
Aims
To provide guidance on the criteria that can be used to evaluate the performance of plant ID apps. To assess the performance of the most popular free ID apps when used on a range of urban wild plants including both flowering and vegetative plants and also to compare the performance of current versions of the apps with those tested several years ago.
Methods
This study evaluated the performance of some of the more popular plant ID apps when tested on all the plants identifiable by a knowledgeable tester on a 2-day urban flora survey in East Scotland. In addition, the improvement in accuracy of the apps since a study in 2020 was assessed by a reanalysis of the original images.
Results
All apps tested have improved over the 3 years between 2020 and 2023 by around 20% points with regards to the percentage of samples identified correctly to species or to family. This improvement probably results from continued crowd-sourcing of images. The best apps tested, which include Flora Incognita and PlantNet, were found to identify correct to species 80%–90% of the plants tested here (which include a wide range of plants with or without flowers, and a range of monocots, herbaceous and woody plants).
Conclusion
We conclude that a specification for ideal apps is that they should provide an indication of confidence in the results and should minimise incorrect identifications, even if it is only the wrong species of two similar ones.
Over the past 5 or so years, many smartphone ‘apps’ for the plant identification (ID) have become available for use by amateur and professional botanists. There is, however, a need for good information on their reliability in practice.
Aims
To provide guidance on the criteria that can be used to evaluate the performance of plant ID apps. To assess the performance of the most popular free ID apps when used on a range of urban wild plants including both flowering and vegetative plants and also to compare the performance of current versions of the apps with those tested several years ago.
Methods
This study evaluated the performance of some of the more popular plant ID apps when tested on all the plants identifiable by a knowledgeable tester on a 2-day urban flora survey in East Scotland. In addition, the improvement in accuracy of the apps since a study in 2020 was assessed by a reanalysis of the original images.
Results
All apps tested have improved over the 3 years between 2020 and 2023 by around 20% points with regards to the percentage of samples identified correctly to species or to family. This improvement probably results from continued crowd-sourcing of images. The best apps tested, which include Flora Incognita and PlantNet, were found to identify correct to species 80%–90% of the plants tested here (which include a wide range of plants with or without flowers, and a range of monocots, herbaceous and woody plants).
Conclusion
We conclude that a specification for ideal apps is that they should provide an indication of confidence in the results and should minimise incorrect identifications, even if it is only the wrong species of two similar ones.
| Original language | English |
|---|---|
| Number of pages | 9 |
| Journal | Plant Ecology & Diversity |
| Early online date | 21 Mar 2025 |
| DOIs | |
| Publication status | E-pub ahead of print - 21 Mar 2025 |
Keywords
- Artificial intelligence (AI)
- automated apps
- citizen science
- ID apps
- identification keys
- image recognition
- plant identification