Abstract:
Accurate and scalable mapping of coral reefs is important for effective conservation and ecological monitoring. This study investigates the application of deep learning-based instance segmentation for predicting individual Scleractinian coral colonies on the Pelorus reef in the Great Barrier Reef, leveraging high-resolution orthomosaic maps. Traditional methods of coral colony detection involve manual annotation and visual inspection, which are labor-intensive and time-consuming, particularly for large-scale reefs. In this study, supervised instance segmentation offers a promising alternative by automating the detection process and scaling up the predictions of coral colonies. A state-of-the-art convolutional neural network (CNN) method is utilized to develop an instance segmentation model to train on a dataset of annotated orthomosaic tiles, each labeled with bounding boxes and segmentation masks at different coral morphology level. The objective is to scale the prediction process from a manually annotated subset to the entire reef site, thereby enhancing mapping efficiency and coverage. Experiments such as data augmentation are performed to boost the model performance to improve prediction accuracy. The ultimate goal is to establish a scalable and efficient workflow that can be applied to new reef sites, thus providing a standardized and consistent approach to coral community mapping.
You may join online at this link: https://uni-bremen.zoom.us/j/93546917122?pwd=K2k3Zm4rZFd1Wit3SFFlV3pieHprdz09