Machine learning applied to big data from marine cabled observatories: A case study of sablefish monitoring in the NE Pacific

Ocean observatories collect large volumes of video data, with some data archives now spanning well over a few decades, and bringing the challenges analytical capacity beyond conventional processing tools. The analysis such vast complex datasets can only be achieved appropriate machine learning Artificial Intelligence (AI) implementation AI monitoring programs for animal tracking classification becomes necessary in particular case deep-sea cabled observatories, as those operated by Networks Canada (ONC), where Petabytes are collected each every year since their installation. Here, we present machine-learning computer vision automated pipeline to detect count sablefish ( Anoplopoma fimbria ), key commercially exploited species N-NE Pacific. We used 651 hours footage obtained from three long-term sites NEPTUNE observatory, Barkley Canyon, on nearby slope, at depths ranging 420 985 m. Our proposed detection was tested validated an initial 4.5 month period (Sep 18 2019-Jan 2 2020), first step towards validation future decade-long Canyon. For period, trained YOLO neural network 2917 manually annotated frames containing images obtain automatic detector 92% Average Precision (AP) 730 test images, 5-fold cross-validation AP 93% (± 3.7%). then ran all material (i.e., period), automatically annotate sablefish. finally applied algorithm results, approximate counts individual fishes moving scene time series proxy abundance. Those abundance estimates among made using volume settings. discuss our results application program, particularly potential complementing fisheries management practices important species.