ORCID
- William Jay: 0009-0007-6191-5801
Abstract
The recent expansion of wild Pacific oysters already had negative repercussions on sites in Europe and has raised further concerns over their potential harmful impact on the balance of biomes within protected areas. Monitoring their colonisation, especially at early stages, has become an urgent ecological issue. Current efforts to monitor wild Pacific oysters rely on “walk-over” surveys that are highly laborious and often limited to specific areas of easy access. Remotely Piloted Aircraft Systems (RPAS), commonly known as drones, can provide an effective tool for surveying complex terrains and detect Pacific oysters. This study provides a novel workflow for automated detection, counting and mapping of individual Pacific oysters to estimate their density per square meter by using Convolutional Neural Networks (CNNs) applied to drone imagery. Drone photos were collected at low tides and altitudes of approximately 10 m across a variety of cases of rocky shore and mudflats scenarios. Using object detection, we co pared how different Convolutional Neural Networks (CNNs) architectures including YOLOv5s, YOLOv5m, TPH-YOLOv5 and FR-CNN performed in the detection of Pacific oysters over the surveyed areas. We report the precision of our model at 88% with a difference in performance of 1% across the two sites. The workflow presented in this work proposes the use of grid maps to visualize the density of Pacific oysters per square meter towards ecological management and the creation of time series to identify trends.
Publication Date
2024-01-01
Publication Title
Ecological Informatics
ISSN
1574-9541
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Mata, A., Moffat, D., Almeida, S., Radeta, M., Jay, W., Mortimer, N., Awty-Carroll, K., Thomas, O., Brotas, V., & Groom, S. (2024) 'Drone imagery and deep learning for mapping the density of wild Pacific oysters to manage their expansion into protected areas', Ecological Informatics, . Available at: 10.1016/j.ecoinf.2024.102708