⌘K
ScopePeriod

Computer Vision Center

FFB quality inspection at the loading ramp · simulated vision pipelineAll operations
Trucks graded (7 d)
1,284
Mill A ramp pilot · 3 bays
Mill A overripe share
10.6%
from 6.1% seven days ago
Grader agreement
91%
vs manual 10% sample
Avg inference time
180ms
Edge GPU · per frame

FFB inspection · Ramp bay 2 · Truck TR-011

Estate B · Div 2 → Mill A · 28 Sep 07:18 · net 8.6 t (weighbridge) · Morning load, mixed rounds

Ripe 97%Ripe 97%Overripe 80%Ripe 87%Overripe 75%Ripe 96%Ripe 90%Ripe 83%Ripe 90%Damaged 78%Ripe 91%Ripe 90%Damaged 89%Ripe 88%Ripe 83%Ripe 96%Ripe 83%Ripe 88%Overripe 72%Ripe 85%Unripe 88%Ripe 95%Ripe 95%Ripe 95%Ripe 82%Ripe 94%Ripe 83%Ripe 84%Unripe 80%Ripe 89%Ripe 84%Ripe 88%Unripe 84%Overripe 69%Ripe 86%Ripe 86%Ripe 86%Ripe 88%Overripe 91%Ripe 94%Unripe 90%Ripe 95%Ripe 93%Ripe 94%Ripe 92%Damaged 74%Ripe 96%Ripe 88%Ripe 97%Ripe 92%
SYNTHETIC SCENE50 detections · conf ≥ 0.60
Sample loads
Synthetic images · simulated detections

Inspection result

Ripeness · grade · weight · decision

FFB Grading v1.3
Ripeness score
86/100
Quality grade
B
Estimated weight
8.6 t
Rejected bunches
0.0%
ACCEPT WITH WARNINGConfidence88%
  • • Unripe proportion exceeds preferred operational threshold.
Class distribution · 50 bunches detectedclick to highlight

Grading thresholds

Operational acceptance policy · applied live

PolicyThresholds are illustrative; the mill grading SOP remains the reference. Unripe > 2.5× threshold triggers reject.

Recent truck inspections — Mill A

Automated ramp grading · last 7 days

OER root cause
Overripe share, daily %
Link to Mill A OER decline
Confidence87%
Overripe share at Mill A rose from 6.1% to 10.6% in 7 days (ramp-camera grading) — the largest contributor (32%) to the OER drop from 22.3% to 21.6%. Harvest-to-mill time also increased (17 h → 26 h). Recommend tighter ramp thresholds for Estates B & E and same-day evacuation from CPs.
InspectionDate · timeTruckOriginOverripeUnripeGradeDecision
INS-918728 Sep · 06:00TR-011Estate B · Div 29.4%3.6%BACCEPT
INS-916427 Sep · 11:17TR-018Estate A · Div 18.0%3.7%AACCEPT
INS-914126 Sep · 16:34TR-025Estate E · Div 19.7%4.1%BACCEPT
INS-911825 Sep · 10:51TR-012Estate B · Div 18.2%4.6%AACCEPT
INS-909524 Sep · 15:08TR-019Estate A · Div 28.4%3.3%AACCEPT
INS-907223 Sep · 09:25TR-026Estate E · Div 27.2%2.6%AACCEPT
INS-904922 Sep · 14:42TR-013Third-party supplier6.0%3.4%AACCEPT
INS-902621 Sep · 08:59TR-020Estate B · Div 28.4%3.0%AACCEPT

Model card

FFB Quality Grading (Vision) · v1.3.2

Pilot
mAP@0.5
0.87
F1 (macro)
0.84
Per-class F1
Ripe0.91
Unripe0.86
Overripe0.82
Empty bunch0.88
Rotten0.76
Damaged0.71
Algorithm
YOLO-style detector + ripeness classifier
Training data
48k annotated bunch images (loading ramp cameras)
Last validated
05 Sep 2026 · drift Stable
Approval
Pilot — Mill A ramp
Prototype model metrics — demonstration only. Known limitations: occlusion in dense loads, low light after 18:00, rotten vs overripe confusion. Grader confirmation required for rejects.

Vision use cases across the enterprise

Stages illustrative
FFB quality grading
Loading-ramp cameras · Mill A pilot
You are here · Pilot
PPE & safety detection
48 CCTV feeds · edge inference
Pilot
Loose fruit detection
Harvester app photos at CP
Validation
Drone canopy counting
Palm census & vacancy mapping
Development
Truck plate recognition
Weighbridge ANPR · ticket matching
Pilot
V-TEKI — IT & Business ConsultingDeveloped by V-TEKI · IT & Business ConsultingPrime Agri AI Intelligence Platform — prototype. All operational records are synthetic; model outputs are simulated; financial impacts are scenario estimates; recommendations require human validation. No actual company confidential data is used.