Monitoring changing intertidal fucoid communities using unmanned aerial vehicle imaging and deep learning algorithms Annulé - Canceled" class="responsive-img img-fluid">
In Nova Scotia, the invasive brown seaweed Fucus serratus is becoming more widely distributed both geographically and ecologically. At several sites on the Atlantic coast this species can occupy up to 40% of the area of the intertidal zone and it can be a dominant species in shallow subtidal water. To best quantify its spatial distribution and abundance, we use unmanned aerial vehicle (UAV) surveys and a U-Net deep learning algorithm to efficiently and automatically quantify cover of Fucus serratus in these new locations. The U-Net is a deep learning algorithm developed for applications in medical imaging that can return robust results with small amounts of training data, which makes it uniquely suitable for our study. Here we summarize the methodology and give preliminary results from selected field sites. We flew a Mavic 2 Pro recreational UAV ten to twenty meters above the shore at Chebogue Point in southwestern Nova Scotia to collect high quality images at low tide. Images were used to train a U-Net algorithm to differentiate between Fucus serratus and co-occurring seaweeds, in particular Fucus vesiculosus and the economically important Ascophyllum nodosum. The final trained U-Net is expected to have applications in long-term monitoring of Fucus serratus encroachment in Nova Scotia and ongoing assessment of economically important seaweeds in Nova Scotia.
