Machine learning applied to big data from marine cabled observatories: A case study of sablefish monitoring in the NE Pacific
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Ocean observatories collect large volumes of video data, with some data archives now spanning well over a few decades, and bringing the challenges of analytical capacity beyond conventional processing tools. The analysis of such vast and complex datasets can only be achieved with appropriate machine learning and Artificial Intelligence (AI) tools. The implementation of AI monitoring programs for animal tracking and classification becomes necessary in the particular case of deep-sea cabled observatories, as those operated by Ocean Networks Canada (ONC), where Petabytes of data are now collected each and every year since their installation. Here, we present a machine-learning and computer vision automated pipeline to detect and count sablefish ( Anoplopoma fimbria ), a key commercially exploited species in the N-NE Pacific. We used 651 hours of video footage obtained from three long-term monitoring sites in the NEPTUNE cabled observatory, in Barkley Canyon, on the nearby slope, and at depths ranging from 420 to 985 m. Our proposed AI sablefish detection and classification pipeline was tested and validated for an initial 4.5 month period (Sep 18 2019-Jan 2 2020), and was a first step towards validation for future processing of the now decade-long video archives from Barkley Canyon. For the validation period, we trained a YOLO neural network on 2917 manually annotated frames containing sablefish images to obtain an automatic detector with a 92% Average Precision (AP) on 730 test images, and a 5-fold cross-validation AP of 93% (± 3.7%). We then ran the detector on all video material (i.e., 651 hours from a 4.5 month period), to automatically detect and annotate sablefish. We finally applied a tracking algorithm on detection results, to approximate counts of individual fishes moving on scene and obtain a time series of proxy sablefish abundance. Those proxy abundance estimates are among the first to be made using such a large volume of video data from deep-sea settings. We discuss our AI results for application on a decade-long video monitoring program, and particularly with potential for complementing fisheries management practices of a commercially important species.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.004 | 0.016 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it