Research On Surface Defect Inspection System For Cylindrical Diode

The object of the research is cylindrical diode. The black tube mainly contain seven kinds ofshape and texture defects, including pipe sizes, pipe defects, missing text, sheet printing, polarring missing, polar white plastic ring width inconsistencies and dew white colloid. Using themethod of image processing and pattern recognition to defect the diode, the object of thedefections is to determine whether the object is defective, and which type of the defects.The key to solving the problems lies in two aspects, the first is the design of optical platform;the second is the design of defect detection algorithm.For the design of optical platform, the photo captured by industrial camera is likely to occuruneven illumination, partial reflection, partial image distortion and other unfavorable conditions.Test out the Low-angle scattering front illumination and the placed way of strip light in a specificspatial angle through the optical principle and the structural features of the object.For the design of defect detection operator, using a classic image analysis process: Imagepreprocessing, defect ROI segmentation, defect ROI feature extraction and defect ROIclassification, in which the difficulties is the defect ROI segmentation and texture extraction.For the first difficulty, text may be mixed with the defect ROI, and it’s difficult to removethe interference of text and segment the defect ROI separately because of the similar gray valuebetween the text and defect ROI. Proposing the improvement of SWT by analyzing the textinherent shape characteristic, which is used in the text segmentation, then, segmentation of thetext is perfect and without affecting the segmentation of other defects ROI. So the defect ROI isleft after removing the text.For the second difficulty, we propose histogram of patterned gradient derection(HPG_D)and patterned gradient amplitude(HPG_A) through the analysis of the rule of the internal pixelarrangement, and show the excellent properties of HPG operator through the relative distancebetween the classes and the relative variance within the class, which can descript the transitionaltexture of no significant stripe more accurately; The classification success rate of HPG operator ishigher than other textures operator through the classification results. Finally, extracting three more characteristics including the average gray, compactness andspace edge direction histogram, then, classifing the defects by decision tree classifier, the defectsrecognition rate is close to100%and defects classification success rate reached95.6%, A betterrecognition and classification results were achieved.