Effectiveness of plant identification applications at identifying common plant species in forage systems
Applied Research
Amanda Grev
Forage Extension Specialist
University of Maryland Extension
Keedysville
Abstract
Accurate plant species identification is essential in making management decisions for hay and pasture systems. A wide variety of mobile phone applications offer users the potential to quickly and easily identify plant species; however, the accuracy of mobile phone applications at correctly identifying plant species under field conditions is unclear. The objective of this project was to test the accuracy of nine popular mobile phone identification applications at identifying common plant species found in forage systems. A total of 30 different plant species (27 broadleaf, 3 grass) were tested in 2024. Target plant species included common broadleaves and grasses located in pastures and hayfields. All plants were photographed on farms throughout the growing season under normal field conditions. For each plant species, three unique images were selected, with priority given to images depicting whole plants in a vegetative state. Each image (n=90) was then individually run through nine different automated identification applications, including: PictureThis, iNaturalist, Seek by iNaturalist, PlantSnap, LeafSnap, PlantNet, Plantum, Google Lens, and Apple Visual Look Up. For each individual image, identification application performance was scored as follows: 4=top suggestion correct; 3=second suggestion correct; 2=third suggestion correct; 1=genus correctly identified but not species; and 0=no correct identification. Data was analyzed using one-way ANOVA, with significance at p ≤ 0.05. Across all applications, 61% of images were identified correctly on the first suggestion and 74% were identified correctly within the first three suggestions. PictureThis was the most accurate application (average ± SD; 3.83 ± 0.7; p < 0.05), identifying 94% of tested images correctly on the first suggestion. This was similar to Plantum (3.77 ± 0.7) and iNaturalist (3.58 ± 1.0), which identified 89% and 79%, respectively, of images correctly on the first suggestion. PlantSnap was the least accurate application (1.55 ± 1.7; p < 0.05), identifying only 26% of images correctly on the first suggestion. Results from this assessment indicate that plant identification applications can be a useful tool for those wanting rapid identification of plant species in forage systems. Future testing will include additional plant species and additional photos representing a broader range of plant growth stages.
Poster has NOT been presented at any previous NACAA AM/PIC
This poster is being submitted for judging. It will be displayed at the AM/PIC if not selected as a State winner. The abstract will be published in the proceedings.
Click to view Poster
Authors: Amanda Grev, Raven Herron
-
Grev, A. Forage Extension Specialist, University of Maryland Extension, Maryland, 21756
-
Herron, R. Student/Intern, University of Maryland, Maryland, 21756