Binary Classification of Coral Reefs Using Deep Learning for Enhanced Monitoring

Coral reefs are among the planet’s most crucial biodiverse and economically valuable ecosystems. Yet, they face unprecedented threats from climate change, which drives ocean warming, acidification, and increased frequency of extreme weather events, leading to widespread coral bleaching and ecosystem degradation. Monitoring coral reef health is critical for conservation efforts, but traditional methods are laborintensive and time-consuming, requiring the expertise of divers, thus making large-scale assessment challenging. This study uses two distinct datasets composed of high-resolution underwater images capturing diverse coral species, bleaching states, and environmental conditions to investigate the application of deep learning, specifically transfer learning models, for binary classification of (healthy and bleached coral images) coral reef health. We evaluated the performance of three pre-trained models-ResNet-152, EfficientNet-B0, and DenseNet-121-on the Bleached Corals and Healthy Corals Classification dataset (BC-HC dataset) collected from Kaggle and Coral Reef Health Identifier dataset (CRHI dataset) collected from Roboflow. Accuracy is used as the primary metric to compare model performance. After fine-tuning, our results demonstrate that transfer learning is highly effective for coral reef health classification, with EfficientNet-B0 achieving the highest accuracy of 99% on BC-HC dataset and 96.9% on CRHI dataset. This research highlights the potential of deep learning as a scalable and efficient tool for coral reef monitoring, thus enabling timely interventions to mitigate climate change impacts while at the same time, offering significant benefits to marine biologists, policymakers, and coastal communities.