Research On Machine-Vision-based Detection Algorithms And Its Application In Industry

With the higher requirement of product quality, as a newly-developed subject, machine vision has been applied widely in the field of industrial detection. To the system of inspection, lower cost, higher accuracy and the ability for real time and on-line inspection will be the future development directions. In this dissertation, the formation and cause of defect product in presswork and float glass are analyzed and classified. And the algorithms of preprocessing, defect features extraction and classification in quality detection are studied.Firstly, according to detection system particularity the software is studied from data flow, information flow and control flow aspects. The workflow of image preprocessing, feature extraction and defect classification is designed.Through analysis of the fault's types in presswork exterior, a simple template matching algorithm is used to registrate the horizontal and vertical offset. Then it proposes that extract the contour of stand image by Sobel operator, segment the distortion and defect at the first subtraction. And it segments the defect at the second subtraction. These algorithms has been proved available and used in an online quality detecting system. In addition, according to the actual situation of glass production, the finding edge algorithm is designed by computing the average gray level and threshold segmentation. The edge is used to calculate the pressure distance and the actual glass panel's width. Finally it analyzes glass defect images and proposes a positive-negative subtraction segmentation algorithm.According the presswork's defect features the line and block feature is extracted and based-on it a corresponding simple classification is designed. Because the false defect appearances after segmentation, the defect type and causation is studied. The shape feature based-on circular rate and neighborhood feature based-on distortion is proposed to be basis for distinction. The corresponding feature extraction and classification decision tree is designed and the secondary classification standard is established.These algorithms have been applied to the online quality detecting system of high-speed presswork production line in Wuhan Hongjinlong Printing Company and the online detecting system of float glass product line in Wuhan ChangLi Glass Company. It has proved that the algorithms are convenient and effectual, and meets practical demand.