When: Thursday 19th of Oct, 1pm AEDT
Where: This seminar will be partially presented at the Rose Street Seminar area (J04) and partially online via Zoom. RSVP
Speaker: Heather Doig
Title: Improving Classification and Detection of Benthic Morphospecies Using Unsupervised Domain Adaptation and Other Method
Regular visual surveys of the ocean environment by robotic vehicles provide scientists with a valuable resource to characterise and monitor the composition and diversity of the benthos or seafloor. The use of machine learning models like classifiers and object detectors for this task is currently limited. Three key issues reduce their effectiveness. Marine species are hard to identify due to wide variations within a species and similarities between species. There are very low numbers of annotations to help train a robust model. Thirdly, using a machine learning model on images from different conditions compared to those used for training can experience reduced performance due to domain shift. This talk will discuss two applications to address these issues. The first uses unsupervised domain adaptation to address domain shift for the classification of marine species. The second uses synthetic data and domain adaptation to detect sea urchins. The application of methods like these could provide a valuable tool for scientists to leverage more information from the thousands of images being captured of the seafloor.
Heather Doig is a third year PhD student with the Marine Lab at the Australian Centre for Robotics. She is working on improving automated observations of the underwater environment by applying current machine learning techniques to underwater images. Heather has a Bachelor of Engineering – Mechanical and Manufacturing (Honours) from the University of Melbourne and more recently a Masters of Data Science and Executive Masters in Arts and Social Sciences from the University of Sydney.