Using Machine Learning to Automate Benthic Transect Image Analysis
The Living Oceans Foundation is working with the Pacific Blue Foundation and the University of California San Diego’s CoralNet cloud-based tools to automate the analysis of images from coral reef photo transects. These benthic transects are a critical tool scientists use to assess the health of a reef. They can provide scientists with important metrics such as coral cover, algal cover, and benthic diversity. Analyzing images from benthic photo transects typically is a labor-intensive process that involves highly trained individuals. We should know, we’ve analyzed tens of thousands of them from the Global Reef Expedition (GRE).
This project uses KSLOF’s large repository of annotated and unannotated images from Lau Province, Fiji, and our expertise in benthic photo transects, to improve the CoralNet machine learning platform. CoralNet is an open source resource for benthic image analysis. It uses machine learning and deep neural networks to automate the annotation of benthic transect images. It also serves as a data repository and collaboration platform to connect coral reef researchers around the world with each other (and with their data).
By using expertly-analyzed images from our research on the GRE to train the CoralNet algorithm, we will be able to improve the automation of image analysis from benthic transects and evaluate if CoralNet returns high accuracy of benthic point identification. Early tests indicate that the revised algorithm can return automated results with over 90% accuracy.
By improving the robustness of the CoralNet machine learning platform, we can enable low-skilled indigenous communities to collect digital images at low cost, link to the cloud, and have the images processed and data delivered back that is relevant for indigenous community reef management as well as to provide data for scientific studies and government management.
This technology could be globally transformative, allowing for very large-scale collection of images at low cost with strong objective analysis using the CoralNet machine annotator platform. Once the CoralNet algorithm is properly trained, the machine can annotate an image more than 1,000 times faster than a human.
Once the algorithm has been shown to work effectively on the reefs of Lau Province, we intend to turn our efforts toward analyzing coral reef transect images from Beqa Lagoon to support Pacific Blue’s Beqa Lagoon Initiative. This community-based project employs a social-ecological systems approach to conserve the biological and cultural diversity of the lagoon while sustainably managing the natural resources of the marine, coastal and terrestrial environments.
The Pacific Blue Foundation (PBF) is a nonprofit organization based in Fiji that works to promote the biological and cultural diversity of the region. They provide research, education, encouragement, and dissemination of sustainable practices in coastal areas.
CoralNet is a resource for benthic images analysis run by scientists at the University of California San Diego (UCSD). The site deploys deep neural networks which allow fully and semi automated annotation of images. It also serves as a data repository and collaboration platform.