FragSize: using computer vision models to monitor massive coral fragment growth in a Caribbean coral nursery in the Seaflower Biosphere Reserve
Gespeichert in:
| Autoren: | , , , |
|---|---|
| Format: | artículo original |
| Status: | Versión publicada |
| Publikationsdatum: | 2026 |
| Beschreibung: | Introduction: Monitoring the growth of coral fragments in restoration nurseries is essential for assessing their performance and the effectiveness of restoration interventions. Two-dimensional methods based on digital photography and ImageJ are commonly used, but their manual nature is time-consuming. Objectives: The aim of this study was to evaluate and test the performance of FragSizel, this open-source tool uses a computer-vision YOLOv8 segmentation model to automatically estimate the planar surface area of coral fragments from two-dimensional images. The evaluation of the performance of the model focused on assessing the degree of agreement between FragSize measurements and those obtained with traditional manual methods such as ImageJ. Methods: A YOLOv8 segmentation model was trained using photographs of fragments of D. labyrinthiformis, O. faveolata, and M. cavernosa that were kept an in-situ nursery at Nirvana Reef Garden, San Andrés Island (Colombian Caribbean). Agreement between FragSize and ImageJ measurements was assessed using Bland–Altman analyses. A total of 81 fragments of three species and 15 genotypes were measured across T0–T1 intervals ranging from 76 to 175 days. Results: Bland–Altman analyses showed strong agreement between FragSize and ImageJ, with low levels of bias and relative error in measurements performed at T₀ and T1 under both calibration methods evaluated. Conclusions: FragSize proved to be a reproducible tool for the automatic estimation of planar area in coral fragments from 2D images, showing agreement with ImageJ-derived measurements under both calibration strategies evaluated. Its open availability on GitHub facilitates its adoption in coral restoration programs requiring accessible and standardized methods for monitoring coral growth. |
| Land: | Portal de Revistas UCR |
| Institution: | Universidad de Costa Rica |
| Repositorio: | Portal de Revistas UCR |
| Sprache: | Inglés |
| OAI Identifier: | oai:portal.revistas.ucr.ac.cr:article/10958 |
| Online Zugang: | https://revistas.ucr.ac.cr/index.php/rrbt/article/view/10958 |
| Stichwort: | Caribbean; coral restoration; monitoring; computer vision Caribe; restauración coralina; monitoreo; visión artificial |