Deep learning in deep nets: Helping fish farmers feed the world
summary
Aquabyte builds underwater camera rigs and image pipelines that run CNNs for body‑part detection, depth mapping and 3D weight estimation, deployed on working farms to replace crane weighing and weekly netting. The system processes 10 million images/day (about 1.5 petabytes/month) from 10 customers, detects sea lice that drive ~25% of costs and earned an approved sea lice counter; pellet detection reaches ~99% to cut feed waste and A/B feeding experiments show ~2–3% efficiency gains.
AI-written summary · about
chapters (28)
- 0:00Introduction
- 0:47Fish farms
- 1:18Overfishing
- 1:49Fish farming
- 2:56Health benefits of fish
- 3:30Impact of fish farming
- 4:28Goal of the company
- 5:48About the company
- 7:20Challenges of fish farming
- 8:18Fish farming is global
- 9:46First fully working product
- 10:31From one farm to many farms
- 11:38What we are working on
- 12:22A single fish
- 12:52The technology
- 14:04Sea lice
- 15:34Fish infestation patterns
- 16:12Fish weighing
- 16:47Simpler approach
- 17:10Underwater environment
- 18:05Depth mapping algorithm
- 18:40Weight distribution algorithm
- 20:11Pellet detection
- 20:26Fish identification
- 21:24Fully autonomous farm
- 22:17Indoor fish farms
- 23:05Image processing
- 24:21Goal
related talks
24mAI in maritime - Autonomous ships
58mDeep Learning in Real Life
46mCombatting illegal fishing with Machine Learning and Azure
42mH2O Deep Water with Python early sneek
40mMachines are Learning
41mPractical Aspects of Applying Modern Computer Vision Technologies
29mDeep Learning, Neuroscience and the future of AI
24mMachine Intelligence gets real