Deep Learning-based Coral Reef Monitoring using UAV Imagery

Coral reefs support nearly 25% of marine biodiversity and are vital for coastal protection, fisheries, and tourism. However, they are under severe threat due to climate change, ocean acidification, and human activities, leading to widespread coral bleaching. Manual diver surveys and coarse satellite scenes slow coral-bleaching assessment. Authors answer this gap with a deep-learning pipeline that reads high-resolution images from low-cost Unmanned Aerial Vehicles (UAVs). A hybrid EfficientNetB3 + DenseNet121 model learns on 1,720 labelled reef images (865 healthy, 855 bleached) after resizing, normalizing, and augmenting the data. The network reaches 97.6% validation accuracy and an F1 score of 0.976. Each 4000 × 3000-pixel frame takes 45 milli seconds (ms) to classify, enabling near-real-time mapping of broad reef tracts. Field trials on the Great Barrier Reef and in the Maldives show that the workflow cuts monitoring costs by 60% versus a Landsat-based approach while keeping fine spatial detail. A Streamlit dashboard lets users upload flight batches, view geotagged health maps, and export results. The study delivers a quick, low-cost, non-invasive tool for early bleaching detection and long-term reef management.