This project investigates various machine learning methods to monitor seagrass meadows, aiming to determine seagrass and turf density and to capture data on seasonal changes. Utilizing USV and UAV technology enables fast and efficient monitoring.

The datasets used in the study include UAV maps from different months, underwater images from a USV, and various ground truth data collected manually or created using GIS tools. The project covers methods for training and preprocessing, including augmentation techniques and programming packages.

The results show that a CNN model can effectively annotate images from a USV, demonstrating the ability to differentiate between different seabed habitats. For UAV images, both SVM and FCN models were tested. While the SVM model had lower accuracy and produced speckled predictions, the FCN model achieved higher accuracy. Some postprocessing of the FCN model was conducted, including area calculation and coverage percentages maps, which biologists can use for analysis.