The Research On Resin Lens Defect Recognition Method Based On Degenerate YOLO Network

Automated inspection of resin lens defects has always been a problem in the lens manufacturing industry.At present,the testing methods of resin lens manufacturers mainly adopt manual visual inspection methods,which are inefficient and have high false detection rate.Domestic research scholars generally use artificial feature extraction methods to extract lens defect features,and make some efforts to identify lens defect features,but still stay in the research stage.In this paper,a deep learning algorithm is used to provide an effective defect identification method for automatic detection of lens defects.An important problem in lens detection is multi-defect recognition,which needs to solve the authenticity identification and shape discrimination of defects.The identification of authenticity is mainly to solve the difference between the inherent defects of the lens and the dust on the surface of the lens.The morphological distinction is mainly to solve the type and position confirmation of multiple defects.This article will conduct research work on these two types of issues.In view of the above problems,this paper selects the four main defects(bubbles,batt,scratches,pitting)in the resin lens and the dust attached to the resin lens as the research object.Aiming at the problem that the dust on the surface of the lens interferes with the identification of pitting defects,the morphological characteristics of dust and pitting are analyzed.Since the dust is much smaller than the pitting point,it is tried to use the morphological treatment method to study the interference of dust removal.Aiming at the problem of identifying multiple defects in a lens with the whole resin lens image as the research object,this paper proposes a resin lens defect recognition method based on degraded YOLO network,using deep learning to deal with multiple defects in one picture.Identify at the same time.The results show that the morphological opening operation is used to remove the interference of the dust on the surface of the lens on the pitting point discrimination,and the morphological change of the lens defect is not eliminated,and the dust interference is removed in the image preprocessing stage;On the identification problem,the deep learning method of resin lens defect recognition based on degenerate YOLO network proposed in this paper is used to detect and identify multiple defects of a whole picture,so that the recognition accuracy of defects in the lens reaches 95.97%,and the missing detection rate is only 0.69%.