Abstract
Coral reef monitoring is a highly laborious process which involves taking images of corals repeatedly over a period of time for data collection, followed by data processing by manually colour correcting, measuring the coral size, and identifying the coral species. The coral reef monitoring problem is further challenging in Singapore's non-ideal water conditions which provides murky visual data. To address this issue, we propose an automated three-stage parallel deep learning data processing pipeline consisting of (a) colour correction, (b) coral segmentation and (c) coral species classification. This pipeline is supported by a graphical user interface (GUI) with a human-computer interaction element allowing modification and validation of results. Experimental results demonstrated excellent performance for colour correction, coral segmentation and coral species classification on Singapore's underwater data. In addition, the entire post processing is sped up to a few seconds per image which can potentially take up to three weeks for manual processing of the data from a typical dive session.