A vision-based LED defect auto-recognition system

Recognising different kinds of light-emitting diode (LED) defect is of great urgency for online manufacturers because such information not only assists in summarising the production variations but also in realising that the defects are caused either by fabricating process or by taping process. Although most of the studies nowadays engage in devising LED defect auto-inspection systems, they are rarely capable of auto-recognising the type of defects simultaneously. This study developed a vision-based defect auto-recognition system for three types of packaged surface-mounted device LEDs. A series of image processing are implemented to obtain the inherent feature of the LEDs. Then, a two-stage multiple discriminant analysis is used to recognise different kinds of defects. The recognition rates of up to 94.34% of LED1S, 95.17% of LED1B and 96.02% of LED2 are yielded, which demonstrate the efficiency and effectiveness of the proposed system.