CORAL–Classification of Reefs and Analysis using Learning algorithms of image processing

Abstract

Coral bleaching, driven primarily by rising sea temperatures, poses a severe threat to coral reefs and the millions who depend on these resources. This study investigates the potential of deep learning for automated bleaching detection, a crucial step towards effective monitoring and conservation. Study evaluated five classification algorithms, each paired with three feature extractors, using a publicly available dataset of 1150 coral images. The findings demonstrate the superior performance of the DenseNet-Logistic Regression model, achieving the highest average accuracy (83 %), F1-score (0.84), and precision (0.85), highlighting its effectiveness in capturing subtle bleaching indicators. While this research underscores the promising outcomes of deep learning for this critical task, further investigation with larger, more diverse datasets is warranted to develop highly accurate and generalisable models for safeguarding these vital ecosystems.