Global Oceans
2024 - Ongoing
Computer Vision AI, Deep Learning
Marine AI Institute
Phase 1 - Database Development
Oceanic whitetip sharks (Carcharhinus longimanus) have declined by over 95% in many ocean basins due to overfishing and bycatch. Understanding their population dynamics, migration patterns, and habitat use is critical for effective conservation, but traditional tagging methods are expensive, invasive, and limited in scale.
The Longimanus AI Photo-ID program leverages cutting-edge computer vision and deep learning to identify individual sharks from photographs of their unique fin patterns, markings, and scars. Like human fingerprints, each oceanic whitetip has distinct natural markings that remain stable over time, enabling non-invasive photo-identification.
Our AI system processes thousands of images submitted by researchers, divers, and citizen scientists worldwide, automatically matching individuals across sightings. This creates a global database revealing migration routes, residency patterns, population estimates, and critical habitats—insights essential for protecting this magnificent apex predator.
Train deep learning algorithms to accurately identify individual oceanic whitetip sharks from photographs with >95% precision, matching human expert performance.
Build a comprehensive photo-identification catalog of 1,000+ individual sharks across major ocean basins, enabling population assessment and connectivity analysis.
Document long-distance movements and seasonal migration patterns through multi-year, multi-location resightings of identified individuals.
Apply mark-recapture statistical models to photo-ID data to generate scientifically robust population estimates for regional subpopulations.
Develop user-friendly platforms for divers and ocean enthusiasts to submit sightings, democratizing shark conservation and expanding data collection capacity.
Translate photo-ID insights into actionable conservation recommendations including marine protected areas, fisheries management, and international policy.
Underwater photographs of oceanic whitetips are collected from research expeditions, dive tourism operations, and citizen scientists globally. Images undergo quality screening for clarity, lighting, angle, and coverage of identifying features (dorsal fin, body markings, scars).
Marine biologists manually annotate key anatomical features and unique markings on high-quality images. This curated dataset trains the neural network to recognize biologically relevant identification patterns. Initial training set includes 5,000+ annotated images of 300+ individuals.
Convolutional neural networks (CNNs) learn to extract distinctive features from shark images and compute similarity scores between individuals. Transfer learning from pre-trained computer vision models accelerates development. Models are iteratively refined through validation against expert identifications.
New images are automatically processed through the AI pipeline: shark detection, fin segmentation, feature extraction, and database matching. Top candidates are returned with confidence scores. Human experts verify high-confidence matches and adjudicate uncertain cases.
Confirmed identifications populate a centralized database with sighting metadata: date, location, depth, behavior, associated species, and observer information. Spatial analysis reveals movement patterns, site fidelity, and habitat preferences.
Photo-ID capture histories feed into mark-recapture statistical models (e.g., Jolly-Seber, spatially-explicit capture-recapture) estimating abundance, survival rates, and recruitment. Annual reports disseminate findings to scientists, managers, and policymakers.
Open-source computer vision toolkit achieving 96.3% identification accuracy on test datasets. Software processes 1,000 images per hour with cloud-based scalability.
Catalog of 847 individually identified oceanic whitetip sharks with 12,400+ sighting records spanning 28 countries and 4 ocean basins. Publicly searchable interface available to researchers.
Web and mobile applications enabling divers to upload shark photos, receive instant ID results, and contribute to conservation science. Platform has 3,200 registered users contributing 40% of data.
Interactive web maps visualizing long-distance movements of 127 sharks tracked across multiple years. Longest recorded movement: 4,800 km over 18 months between Red Sea and Indian Ocean.
Regional abundance estimates for Red Sea (120 individuals, 95% CI: 95-158), Hawaiian waters (78 individuals), and Caribbean (43 individuals) with evidence of population connectivity.
6 peer-reviewed papers on AI methodology, population genetics linkage, behavioral ecology, and conservation recommendations. Results inform IUCN Red List assessments.
Evidence-based proposals submitted to regional fisheries management organizations for seasonal closures in critical habitats and bycatch reduction strategies.
Individual animals catalogued
Across 28 countries
Matching human expert performance
Contributing observations
Multi-year movement data
Red Sea to Indian Ocean
Published in peer-reviewed journals
Informed by program data
Lead Scientific Partner
Provides field expertise, diving support, and long-term ecological data integration. Co-leads site selection and survey planning.
Field Research Partner
Coordinates global network of dive operators and researchers submitting photo-ID data. Facilitates field expeditions to key shark aggregation sites.
Academic Partner
Multi-university collaboration (Stanford, Miami, James Cook) providing graduate researchers, genetic sampling, and statistical modeling expertise.
