Comparison of Pennatula aculeata sea pen abundance at a fixed site over an extended period using machine learning, image annotation, and image cataloguing / Claude Nozères, Laurence H. De Clippele, Jinshan... : Fs97-4/3300E-PDF

"A benthic platform documented retraction behavior of Pennatula aculeata colonies in The Gully, Nova Scotia, Canada. The fixed site had 12,054 images taken every 30 min between October 2022 and July 2023. Three methods were used to quantify visible sea pens. A machine learning model using RootPainter was trained on cropped images of several colonies, with the areas classified as sea pens serving as a proxy for presences. The BIIGLE online platform was used to annotate one image per week, with colonies ranging from 10 to 52 per image. Finally, a catalogue with Adobe Lightroom was used to label 22 P. aculeata colonies individually, with 2 to 21 visible per cropped image. Similar values indicating presences over the months were produced across the three methods. Cataloguing required the most effort, though it was estimated to be faster than BIIGLE annotation and easier to produce counts than with RootPainter. Overall, RootPainter was preferred for its efficiency and Lightroom for its utility, while BIIGLE was not recommended because of the higher manual workload"--Abstract, page iv.

Lien permanent pour cette publication :
publications.gc.ca/pub?id=9.947230&sl=1

• Format MARC XML
• Format MARC HTML

Demander des formats alternatifs