Project overview
An advanced computer vision system that automatically identifies and classifies defects in manufactured products, focusing on cups and television sets on a production line.
The project uses the YOLOv5 object detection architecture, which offers the real-time processing capability that production line integration actually requires. What makes it distinctive is the custom dataset: hundreds of images of cups and TVs carrying various types of manufacturing defect, collected and annotated by hand. That manual data acquisition is what let the model train on examples which genuinely represent real-world quality control.
The system detects subtle defects such as scratches, dents, colour inconsistencies and structural abnormalities, classifying each item as good or defective with clear visual indicators. Training and optimisation ran on local hardware, showing that an effective solution can be built inside a tight timeframe without relying on expensive cloud resources.
Project gallery
Click any image to enlarge.