The purpose of this paper is to propose a defect detection method of bare printed circuit boards (PCB) with high accuracy.
First, bilateral filtering of the PCB image was performed in the uniform color space, and the copper-clad areas were segmented according to the color difference among different areas. Then, according to the chaotic characteristics of the spatial distribution and the gradient direction of the edge pixels on the boundary of the defective areas, the feature vector, which evaluates quantitatively the significant degree of the defect characteristics by using the gradient direction information entropy and the uniform local binary patterns, was constructed. Finally, support vector machine classifier was used for the identification and localization of the PCB defects.
Experimental results show that the proposed algorithm can accurately detect typical defects of the bare PCB, such as short circuit, open circuit, scratches and voids.
Considering the limitations of describing all kinds of defects on bare PCB by using single kind of feature, the gradient direction information entropy and the local binary patterns were fused to build a feature vector, which evaluates quantitatively the significant degree of the defect features.
