Coral Health Identification using Image Classification and Convolutional Neural Networks

The health of the coral reefs is at risk due to the increase in water pollution and climate change. The destruction of these coral reefs becomes relevant as the Philippines has a huge coral reef presence making it the second largest in Southeast Asia. The early detection of coral bleaching and deterioration may be able to decrease the death rate of corals. This study focuses on the use of Convolutional Neural Networks (CNN) to identify a coral’s state: (1) Healthy, (2) Dead and (3) Bleached. The model developed in this study will then classify these corals to determine which corals need harvesting or regeneration. There are two datasets used in this study for comparison purposes. To build the dataset for this study, the researchers compiled images from known datasets used in other studies. The images for alive corals were taken from the RSMAS, dead corals from EILAT while bleached corals were taken from ReefBase. The results of the second dataset have an accuracy of 84.93% which is better than the first dataset that have an accuracy of 68.75%. The results showed that datasets which have larger sample size perform better than smaller datasets. It also showed that for machine learning models, the quantity outperforms the quality of image data.