Oyster Spat Counter for a Small Aquaculture Farm

Results at a glance
Automatic counting worked in the pilot
Counted accurately even with the seed clumped and the water murky
Showed a clear path from accepted ±20% variance to a camera that just watches
Challenge
A camera counts 10mm oyster seed in murky, moving water — seed a crew was counting by hand, full-time, against a ±20% variance the trade had accepted as normal. The only alternative anyone was selling the farm was a $150,000 imported grader.
Constraints
- check_circleReal water — murky, uneven light, never holding still
- check_circleCounts had to happen on the spot, out on the farm
- check_circleHad to be accurate enough for harvest planning and the compliance paperwork
- check_circleA $150,000 imported grader was the visible alternative, far beyond what a small farm should need for counting
Approach
A camera rig that watches the seed and counts it automatically — no tweezers, no hand-tallying, no imported grader as the only path forward.
Architecture
- 01The system spots each tiny oyster, traces its outline, and tells it apart from everything else in frame
- 02Counts the seed as the camera passes over it
- 03Handles the seed even when it's clumped together or half-buried
Implementation
- 01Built it to work in real water — murky, moving, badly lit
- 02Tied the camera to simple water-condition sensors
- 03Built around the camera-counter idea the operator had wanted for 8 years
- 04Proved it on a working farm, in a pilot
Results
- check_circleAutomatic counting worked in the pilot
- check_circleCounted accurately even with the seed clumped and the water murky
- check_circleShowed a clear path from accepted ±20% variance to a camera that just watches
Business impact
- check_circleCould free a full-time worker from counting by hand all day
- check_circleLets the farm watch beds too big to ever count by hand
- check_circleAvoids treating a $150,000 imported grader as the only way to modernize counting
- check_circleLive counts help time the harvest and meet the paperwork rules
Technologies
The research is public.
Benchmarks run on machines I own, measured against frontier systems — with the failed approaches reported next to the results. Read it before you take my word for anything.