Research On Defect Detection Method Of Printed Circuit Board Based On Machine Vision

The development trend of electronic product miniaturization makes the manufacturing process of printed circuit board(PCB)become more and more complicated,and the higher requirement of product quality detection is also put forward.Machine vision inspection technique provides possibility for the efficient detection in volume production of PCB due to advantages of fast,non-contact,etc.But the existing related image processing algorithms still have several faultiness in specific application just like poor reliability,vulnerable to environmental interference and low detection accuracy.In addition,most algorithms can only process gray level images,which limits the advantage release of PCB visual inspection technique.Considering to improve the detection accuracy,generality and stability of algorithm,a defect detection method for PCB color images is proposed in this paper to achieve the purpose of efficient detection in PCB manufacturing.In aspect of image preprocessing,the limitations of applying RGB color model to image processing is analyzed with the characteristics of PCB color image,and algorithm chooses to preprocess images in CIE L*a*b* color space;the bilateral filtering algorithm is used to protect the edge details efficiently in removing the image noise;by comparing results of three segmentation algorithms,the K-means clustering method is chosen to segment the wiring layer and pads area accurately in order to reduce the amount of calculation in subsequent operations.In extraction of defect features,through analyzing the gradient direction distributions and local binary patterns(LBP)of edge pixels in image specially,typical image features in defect regions are discovered,in which the gradient directions of edge pixels are chaotic within certain range,and the corresponding LBPs are non-uniform patterns mostly.On this basis,the conception of neighbor gradient direction information entropy and improved LBP algorithm are defined to quantitatively describe the defect information,then both of them constitute the feature vector of pixels in region of interest.In defect recognization,algorithm uses support vector machine(SVM)classifier for the binary classification of feature patterns,which means the classifier obtained by training can extract defect area pixels as detection results from target area,and defect region is also located for the intuitive display of defect information.Algorithm is realized by software programming to verify the true performance oftwo feature extraction operators.Experimental results of PCB images show that the comprehensive detection accuracy of visual detection algorithm proposed in this paper is above 98% for several PCB defects such as short circuit,open circuit,scratch,hole,residual copper,burr and notch.This algorithm is insensitive to the changes of environmental illumination and angle of object to be inspected,with strong robustness,it can satisfy the requirement of PCB defects detection and also has a good universality.