The deep-water coral Lophelia pertusa is a common reef-building scleractinian coral, or stony coral, occurring in mid to deep waters around the world. The reefs they form are regarded as hot spots for biodiversity and carbon cycling, and play a key role in benthic ecosystems in Norwegian waters. The cold-water reefs are however under increasing anthropogenic pressure due to human activities and a changing environment. Development of methods that enable time-and cost-effective monitoring of these reefs is therefore important. We propose using synthetic aperture sonar (SAS) on-board autonomous underwater vehicles (AUVs) as a means to detect the presence of assemblages of corals. Automated segmentation of areas with coral is presented using a convolutional neural network (CNN).