TÜBİTAK 2209-B: Visual Defect Detection in Packaged Foods
A deep learning quality control system for packaged food production, detecting packaging tears, label printing errors, liquid leakage and deformation at pixel level.
- Type
- Research project
- Period
- April 2025 - June 2026
- Role
- Principal investigator
- Stack
- Python, PyTorch, YOLOv11, FastAPI, Next.js, Docker
A quality control system for packaged food production, built to catch packaging tears, label printing errors, liquid leakage and deformation. I led it as principal investigator from TÜBİTAK approval in April 2025 to completion in June 2026.
A YOLOv11 segmentation model marks defects at pixel level, and a ratio-based scoring algorithm turns each result into a numerical quality score. After 100 epochs the model reached 94.78% classification accuracy.
The system runs on a FastAPI backend with a Next.js dashboard and Docker Compose. Results were published as a full-text paper at an international congress.
Highlights
- Led the project as principal investigator from TÜBİTAK approval to completion.
- Built a deep learning quality control system detecting defects at pixel level with a YOLOv11 segmentation model.
- Assigned numerical quality scores with a ratio-based scoring algorithm; the model reached 94.78% classification accuracy after 100 epochs.
- Built a scalable architecture with a FastAPI backend, a Next.js dashboard and Docker Compose.
- Published the results as a full-text paper at an international congress.