Automatic image annotation of underwater video imagery with artificial intelligence techniques and machine learning algorithms
Abstract
The increasing need to gather large-scale data on the distribution and conservation status of deep-sea benthic species and habitats could resulted in the development of low-cost imaging tools to facilitate the access to the deep sea. In this context, the Azores Deep-sea research group developed the Azor drift-cam, a cost-effective video platform designed to conduct rapid appraisals of deep-sea benthic habitats. Built with off-the-shelf components, the Azor drift-cam is an effective, affordable, simple-to-assemble, easy-to-operate, resilient, operational and reliable tool to visually explore the deep sea to 1,000 m depth.
The Economic Exclusive Economic Zone of the Azores is vast, with about 1 million km2, and mostly deep, with the Mid-Atlantic Ridge and 100+ underwater features (seamounts/ridges) providing a complex geomorphological setting. In the past three years , the Azores Deep-sea research group have performed 400+ dives and video transects with the Azor drift-cam in 30+ seamount, ridges, and island slopes, covering a linear distance of about 200 km and summing over 300 hours of video footages. These images allowed us to better understand the spatial distribution of marine invertebrates (e.g. cold water corals and sponges) in the Azores, identify new areas that fit the definition of Vulnerable Marine Ecosystems, and compile valuable scientific information to inform the development of spatial management policies. In the coming years, the Azores Deep-sea research group aim to visit all geomorphological features inside the EEZ of the Azores shallower than 1,000m depth. These efforts will continue to increase the volume of deep-sea video imagery for the deep-sea of the Azores and are expected to produce video imagery from over 600 dives, summing over 500 hours of seafloor images.
The process of human assisted image annotation (i.e. adding vocabulary to an image) of deep-sea video imagery is very time consuming, hampering the timely analyses of the videos collected. The application of artificial intelligence techniques and machine learning algorithms have the potential to help with the automated detection of sediments types and megafauna. This PhD thesis aims to take advantage of the extensive database of deep-sea imagery and develop new tools and methodologies to advance automated detection systems using artificial intelligence techniques and machine learning algorithms that incorporates the expertise of marine biologists.
Candidate’s Profile
This PhD call is most suitable to “Computer scientist” with interest in natural sciences or a “natural scientist” with strong interests in advanced computing. The candidates should be able to demonstrate a strong background on computing sciences, have a certain degree of scientific independence, and an excellent level in oral and written communication in English.
Name(s) of supervisor(s)
Telmo Morato, Carlos Dominguez Carrió, Ocean Sciences Institute – Okeanos, University of the Azores
Josá Cascalho; full professor at the Group of Robotics and Artificial Intelligence, University of the Azores
Timm Schoening, DeepSea Monitoring, Information, Data and Computing Centre, and Research Data Management, GEOMAR, Germany
Identification of PhD Program
PhD Programme in Sciences of the Sea, University of the Azores, Portugal, Universidade dos Açores | (uac.pt)
Notice of the call (english version)
Notice of the call ( portuguese version)
We are no longer accepting applications for this scholarship. Thank you.