Machine-Learning Approach for Automatic Detection of Wild Beluga Whales from Hand-Held Camera Pictures
Bibliographic record
Abstract
A key aspect of ocean protection consists in estimating the abundance of marine mammal population density within their habitat, which is usually accomplished using visual inspection and cameras from line-transect ships, small boats, and aircraft. However, marine mammal observation through vessel surveys requires significant workforce resources, including for the post-processing of pictures, and is further challenged due to animal bodies being partially hidden underwater, small-scale object size, occlusion among objects, and distracter objects (e.g., waves, sun glare, etc.). To relieve the human expert’s workload while improving the observation accuracy, we propose a novel system for automating the detection of beluga whales (Delphinapterus leucas) in the wild from pictures. Our system relies on a dataset named Beluga-5k, containing more than 5.5 thousand pictures of belugas. First, to improve the dataset’s annotation, we have designed a semi-manual strategy for annotating candidates in images with single (i.e., one beluga) and multiple (i.e., two or more belugas) candidate subjects efficiently. Second, we have studied the performance of three off-the-shelf object-detection algorithms, namely, Mask-RCNN, SSD, and YOLO v3-Tiny, on the Beluga-5k dataset. Afterward, we have set YOLO v3-Tiny as the detector, integrating single- and multiple-individual images into the model training. Our fine-tuned CNN-backbone detector trained with semi-manual annotations is able to detect belugas despite the presence of distracter objects with high accuracy (i.e., 97.05 mAP@0.5). Finally, our proposed method is able to detect overlapped/occluded multiple individuals in images (beluga whales that swim in groups). For instance, it is able to detect 688 out of 706 belugas encountered in 200 multiple images, achieving 98.29% precision and 99.14% recall.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".