High-resolution underwater imagery can provide a detailed view of coral reefs and thus facilitate insight into important ecological metrics of the health and well-being of the reefs. In recent years, anthropogenic stressors, including those related to climate change have altered the community composition of coral reef habitats around the world. Currently the most common method of quantifying the composition of these communities is through benthic quadrat surveys, which requires manual annotation of numerous randomly projected points superimposed on each image. This is a time-consuming task that does not scale well for large studies. “Deep Learning” continues to show its usefulness in a variety of fields, and our research investigates the potential for its application to automatically characterize the contents of coral habitat imagery data. By repurposing existing point-based annotation data, trained deep learning models can be used to classify class categories at the pixel-level and are capable of generalizing to similar habitats, making it trivial to compute the change in community composition across space and time. We also showcase how a trained deep learning model can be used in conjunction with structure-from-motion photogrammetry to easily quantify the community composition of a reef in 3D.
