Feature Abstract And Accurate Discern Study Of PCB Solder Joint Defects Based On Image Processing

Inspection of solder joint defects is a key step during PCB manufacture.As circuit is trending more sophisticated,traditional examination cannot meet our requirements and has been replaced gradually by AOI that is now used widely on SMT line.It is the feature extraction of PCB solder joint defect that decides whether or not the objective of accurate identification can be achieved.So it was studied in this thesis,and the main work completed was as follows:The color image of PCB solder joint is first captured by optical CCD,then transformed to required gray image with kinds of noise,narrow contrast and blurry edge owning to disruption stemming from table vibration or ambient light,etc.Given PCB image character,a series of image-preprocessing methods,including space denoising,enhancement,thresholding and morphological filtering,were proposed to improve image condition,the treated image will become clean,sharp and feature highlighted so as to be ready for future feature extraction.Result comparison of pre-post images by different treatment ways was emphasized when analyzing.It is required to complete image alignment before feature extraction.Here excellent Canny edge detector was selected to search edge after comparing performance of several edge detection operators belonging to first and second order.Making use of Hough transform to detect direct line and circle was discussed,cross correlation coefficient with the ability to reflect image alignment degree among other image features was proposed.Finally,the image mosaic technique was studied,here affinetransformation and gray interpolation by backward mapping of nearest neighbor,bilinear and high order were analyzed.Basic features of solder joint such as shape,topology,centroid,gray histogram and its statistical characteristic were introduced based on feature selection principle,solder joint feature extraction should follow pattern recognition mode like linear discriminant and tree classifier.The detection program applied the hybrid method combining reference method and non-reference method,then how to discern each defect of these 5 common severe defects selected in chapter one was analyzed.The work completed contains different features extraction according to what the defect is,for solder-missing and short,extracting shape features,for solder insufficient and solder excess,picking up R component gray image and color gradient for further analysis,and for shift,centroid coordination was used to calculate the shift coefficient,listing main detection steps,summarizing extracted features to draw inspection flow chart.Finally,in order to test accuracy of defect identification,certain number of solder joints with those defects were experimented for examination by means of ways mentioned above,the outcome showed it can get a satisfactory result.