Why it matters
Without AI, grading relies on manual sampling that can miss subtle deformities, colour faults, or rising cull rates, leading to inconsistent fry quality and increased losses.
From design to safety
Design
AI-guided layout for grading workflow
Define data flows, from imaging sensors to the cull decision engine, and determine required record-keeping for fry lots, size bins, and defect categories.
- fry grading station
- size sorter
- deformity detector camera
- color fault analyzer
Build
Install grading and sorting infrastructure
Integrate imaging mounts, conveyors, and calibration rigs with the RAS control panel, setting thresholds for size, color, and deformity classifications.
- fry grading station
- size sorter
- deformity detector camera
- color fault analyzer
Operation
Real-time grading and culling in live rearing
Capture fry images, extract features, apply AI scores for size, deformities, and color, and execute cull or retention actions with a traceable log.
- fry grading station
- size sorter
- deformity detector camera
- color fault analyzer
Maintenance
Continuous calibration and data health checks
Schedule sensor cleaning, recalibrate imaging focus, verify threshold drift, and review AI model drift against control fry batches.
- fry grading station
- size sorter
- deformity detector camera
- color fault analyzer
Safety
Traceability and health compliance
Ensure notifiable disease and welfare records tie to fry lots, with sign-off by Fish Health Officer and compliance checks.
- fry grading station
- cull-rate dashboard
- color fault analyzer
- hatchery water-quality probe
Part by part
- 1
fry grading station
- Watch for
- drift in size-score distributions, missed small fry, inconsistent defect classifications, or new deformity patterns across batches
- AI action
- adjust size thresholds and defect models in real time; trigger human review for unexpected changes
- 2
size sorter
- Watch for
- bin misalignment, jammed conveyor, inaccurate bin counts, or misrouted fry streams
- AI action
- recalibrate sorting gates and update bin routing rules; log bin-level accuracy metrics
- 3
deformity detector camera
- Watch for
- false positives/negatives in deformity scoring, blurred images, or occluded fry
- AI action
- trigger focus and exposure recalibration; raise flag for human review on ambiguous cases
- 4
color fault analyzer
- Watch for
- color balance drift, misclassified colour faults, or lighting inconsistency
- AI action
- update color tolerance windows and flag high-risk fry batches for secondary inspection
- 5
cull-rate dashboard
- Watch for
- rising cumulative cull rate, uneven cull by batch, or seasonal shifts in defect profile
- AI action
- alert supervisor and suggest targeted culling by defect profile and batch history
- 6
hatchery water-quality probe
- Watch for
- ammonia/nitrite spikes, DO drop, temperature deviation, or pH drift
- AI action
- adjust feeding schedule or send maintenance ticket; correlate with fry health drift
Worked example
Culling cue from color faults in pink-gold koi fry
Illustrative scenario with sample data.
A mid-cycle batch shows rising cull rate due to subtle colour faults; AI flags potential batch risk before grading, enabling pre-emptive sorting and record-keeping actions.
Signal in
batchId: FK-412, date: 2026-11-01, fryCount: 3200, sizeMean: 2.1mm, colorScore: 0.72, deformityScore: 0.08, ammonia: 0.03, nitrite: 0.01, temperature: 26.5C, batchHistory: BK-407/6
System out
alert: colorFaultRisk; batch FK-412; action: isolate 400 fry; notify hatchery manager; generate QA report; update cull-rate forecast; attach image set and sensor logs.
Human sign-off: Hatchery manager reviews the alert, confirms the cull count, approves the proposed selective culling plan, and signs the QA report for regulatory traceability.
How to roll it out
Phase 1
Weeks 1 to 4: Baseline
Install imaging, sorting, and logging to establish baseline grading performance and data capture across all fry lots.
Phase 2
Weeks 5 to 12: Pilot
Run pilot batches with AI thresholds; adjust defect categories and cull rules based on end-to-end feedback and traceability outcomes.
Phase 3
Weeks 13 onward: Full deployment
Scale AI-driven grading across all fry batches; implement continuous calibration and QA reporting with notifiable-condition linkage.
Where people stay in charge
- Decisions about culling must remain with the farm manager or fish health officer, because welfare and regulatory compliance depend on expert judgment.
- Color fault and deformity classifications cannot be finalized without human review if anomaly rates exceed defined thresholds.
- All changes to AI thresholds require an audit trail and sign-off by the Fish Health Inspectorate when notifiable conditions are suspected.
Questions owners and crews ask
How does AI determine which fry to cull based on size and deformities?
AI analyzes images to extract size metrics, detect deformities, and score color faults; the system then applies predefined culling rules and generates a batch disposition; staff verify before action.
What records are created for fry grading decisions?
Dispositions by batch, grade counts, defect categories, thresholds, timestamped sensor readings, imaging metadata, and QA notes are stored in the hatchery information system with full audit logs.
Who approves culling recommendations from AI?
The farm manager, with input from the fish health officer, approves or overrides AI-generated culling decisions before action and signs the disposition record.
How is regulatory compliance tracked in AI fry grading?
Notifiable condition flags, health certifications, quarantine records tie to fry lots; inspectors review logs, QA reports, and traceability summaries during audits.