Research On Automatic Classification And Recognition Technology Of Large Size Plate Microdefects

Currently,the domestic display panel capacity has reached the first in the world.However,the development of domestic high-end equipment for large-size plate defect detection required by the production line,especially the development of review systems with defect classification and recognition functions,is relatively lagging behind,still in a state of foreign technology monopoly.And the type of defect is mainly judged manually.Therefore,there is an urgent need for review systems for micro-defect identification and classification in China.The currently published literature shows that the research on defect classification mainly focuses on the classification of point defects,linear defects according to the defect shape and other information under periodic background texture,or fatal and non-fatal defects based on the location and size of the defect Classification,can not achieve the required re-inspection system to identify the type of defects of micro-defects.Therefore,in order to solve the problem of defect identification of large size plate,this paper studies the review system.This paper designs the review system,studies the defect recognition algorithm under non-periodic background texture,and performs accuracy analysis.The main research contents of this topic are as follows:(1)Design the optical system and mechanical structure of the review system.According to the analysis of the requirements of the review system,through the analysis of the advantages and disadvantages of several possible methods of imaging,illumination,and focusing,it is finally determined that the infinite microscopic imaging system is adopted.The illumination method uses critical coaxial illumination,and focusing method uses laser displacement sensors for ranging and focusing.Finally,This paper also completes the selection of the devices required by the system,and designs the mechanical structure of the connection between the devices according to the selected devices.(2)This paper proposes a method based on the combination of singular value decomposition and reconstruction method and hole-based feature detection method for defect extraction of review images with non-periodic complex background texture.For the singular value decomposition and reconstruction algorithm,the residual background texture noise filtering method and the reconstruction boundary point automatic extraction method are designed.In order to overcome its limitation on the area where the defect is located,a combination of hole-based feature detection method is proposed to extract the defects of the non-horizontal and vertical synaptic texture areas.(3)This paper proposes a defect classification and recognition method based on color space.For the defect classification object of this subject,the HSV color model is finally selected.Adopt color rule-based definition and probability classification recognizer to identify the adhesive film defects?exfoliated film defects and foreign object defects.In order to improve the classification and recognition accuracy of foreign object defects,the classification method was improved.The foreign object recognition method combined with the color features of the contour position is used to realize foreign object recognition,which effectively improves the type recognition of foreign object defects at the edge of the circuit texture...