Wavelet-based neural network and statistical approaches applied to automated visual inspection of LED chips

Abstract

This research explores automated visual inspection of surface defects in a light-emitting diode (LED) chip. One-level Haar wavelet transform is first used to decompose a chip image and extract four wavelet characteristics. Then, wavelet-based back-propagation network (WBPN) and wavelet-based Hotelling statistic (WHS) approaches are respectively applied to integrate multiple wavelet characteristics. Finally, back-propagation algorithm of WBPN or Hotelling test of WHS judges existence of defects. Two proposed methods achieve detection rates of above 90.8% and 92.4%, and false alarm rates below 4.4% and 6.1%, respectively. A valid computer-aided visual defect inspection system is contributed to help meet quality control needs of LED chip manufacturers.