THENLINE us
LINE usภาษาไทย
Back to all work
Vision AIClient work, poultry plant

ChickenQC Conveyor AI

A camera watches chicken on the conveyor. Foreign objects and bad pieces get blown off by air, all on one Raspberry Pi 5.

Year
2026
Role
Data, training and hardware wiring
Status
Deployed on site
GitHub
ChickenQC Conveyor AI

The problem

The line needed people staring at the belt all day. Paper, plastic or glass could slip through with the chicken.

What we built

  1. 01

    Collected tens of thousands of real belt images and trained YOLO11 for drumstick, breast and foreign objects

  2. 02

    Added HSV color checks for pieces with abnormal color

  3. 03

    Timed an air valve to kick rejects off the belt as they pass

  4. 04

    Everything runs on one Raspberry Pi 5, no PC or GPU box

Numbers

99%Drumstick accuracy on test set
97%Breast accuracy on test set

Watch it work

0 pass / 0 reject

Simulated with real dataset images. Good pieces pass, foreign objects get blown off.

Screenshots

Model output, drumstick vs breast with confidence
Model output, drumstick vs breast with confidence
Confusion matrix on the test set
Confusion matrix on the test set
Training curves
Training curves
Flow from camera frame to air reject
Flow from camera frame to air reject

Stack

YOLO11OpenCVRaspberry Pi 5ESP8266