Research On Defect Detection For Resin Lenses Based On Machine Vision

In recent years,the automation level of China’s lenses production industry has been improved to a certain degree,but the quality inspection of resin lenses in the production process is still at the stage of manual inspection.The biggest problem with manual inspection is that the quality and the qualification criteria of product are completely determined by the experience and level of the inspector.They are susceptible to human factors and individual differences,and the false detection rate is relatively high.In addition,manual inspection exists other problems such as high labor cost,low detection efficiency,and fatigue of human eyes.Therefore,the automatic detection of resin lenses has become a bottleneck in the technological development of lenses production industry.Based on the needs of automatic surface defect detection for resin lenses,this paper carried out research on defect detection for resin lenses based on machine vision.The main contents include:(1)Systematic design of visual inspection system for defects was designed.According to the characteristics of the optical lens from its dark-field image,the selections of CCD camera,camera lens and light source were completed,and resin lens was laterally illuminated to obtain the resin lens image.(2)The camera calibration and image preprocessing algorithms were studied.An improved fast Hough circle detection algorithm was proposed to complete the rapid extraction of the region of interest.The impulse noise and Gaussian noise in the images were analyzed,and an effective hybrid filtering method was designed to eliminate impulse noise and Gaussian noise.(3)Defect segmentation and feature extraction were completed.The methods of object segmentation based on edge and threshold were compared.An improved Canny algorithm based on local window threshold was proposed to divide the window of defects in the image,then the local optimal segmentation threshold was selected and the target defects were segmented from the image.Chain code was used to describe the contour of the defect,and a series of features such as perimeter and area were calculated,and appropriate features were selected to describe the target defects.(4)Target defects were classified and identified target based on SVM classifier.The classifier model of support vector machine was used to classify different types of defects.The classification experiment achieved good results,and the classification recognition rate reached more than 90%.