| Summary: | 碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 106 === The PCB(Printed Circuit Board) has become a key component of electronic products nowadays. For PCB makers, how to eliminate defective products before packaging process is a very important issue. This dissertation presents the design of an automatic detection system with a suitable light source and CCD, and a set of image inspection processing to detect surface defects. The major manufacturing defects addressed in this dissertation on conductive areas of PCB include: oxidation, contamination, etc. These defects will affect the functions on finishing goods.
A conventional method for detecting PCB surface defects is manual visual inspection. This method may have inconsistent criteria for different people, and the human eye is prone to fatigue. In this dissertation, an automatic detection system, called HybridDetect, will be designed to detect the PCB conductive area to solve the insufficiency problem of detection accuracy with human-eye detection, and to improve the detection rate and reduce the chance of misjudgment.
In this dissertation, ImageRange of the image processing library OpenEVision and several image processing techniques such as simple subtraction, moving subtraction, average statistics, and closed average statistics as well as the AlexNet convolutional neural network are evaluated to detect the defects in the conductive area of the PCB. The experimental results show that HybridDetect with the integrated moving subtraction and closed average statistical detection can have better detection rate and test accuracy than the others.
|
|---|
