Tidal energy is one potential source of green energy but the impact of placing large turbines on the seabed is not fully understood. In particular, how are sea mammals coping with such changes in their environment? Do we need to put automatic mitigations in place – is it even possible? For over a year, we have been annotating more than 60 terabytes of data, finding these rare events of interest. We’ve looked at whether Artificial Intelligence, combined with sonar, can automatically spot and classify interesting underwater objects. We’ve tried several approaches, some more successful than others, but there are promising results with movement detection and segmentation.
This talk covers our initial findings - where we went wrong and one promising approach that we are still actively researching. I'll be talking about U-Nets, convolutional neural networks, sonars and the unknown entities that may lurk in the depths!
Licensed to the public under https://creativecommons.org/licenses/by-sa/4.0/