Study On The Multiple-grade Automatic Detecting Of Glass-bulb's Defects By Machine Vision

Automatic visual inspecting(AVI) technology is based on machine vision and applies image processing and analysis, pattern recognition, artificial intelligence and precision instrument synthetically. It is an emerging detection technology. As a modern inspecting method people pay attention to automatic visual inspecting increasingly, so that visual inspecting that is applied in detection of industrial products on line, especially defect detecting, has extensive market prospect.Bubbles and scratches are the main influential factor to the quality of the glass-bulb in the production of image tube. In the thesis, multiple-grade detecting method by machine vision is studied in order to inspect the defect of glass-bulb automatically.Some main researches are achieved, such as;(1) In the case of refraction defect detecting of glass-bulb by machine vision is presented by means of studying of defect detecting principle. Illuminating mode is lateral. And light source of LED that is appropriate to this system is designed. In the thesis, two grade detection method is presented to resolve the tough problem of detecting the infinitesimal objects in glass-bulb. Calibration based on virtual grid is adopted to locate defects in first grade system, and pixel calibration is presented to measure parameters in second grade system. Second grade system is guided by first grade system to classify and recognize defects.(2) To system noise feature, a novel method based on minimum square error that can eliminate noise is proposed. This method can not only eliminate noise effectively and but also reserve image edge details. So it can improve resolution. The proportion of target defects in glass-bulb image is so low, so that a new method of adaptive and local segmentation is presented. And morphological filter is applied to eliminate isolated dot noise effectively after segementation image.(3) Region labeling is presented and can differentiate multiple target regions. And morphological characteristic parameters are extracted on basis of labeling image. Feature vector that are composed of shape factor, extention and width divided by length of minimum enclosing rectangle can recognize and classify the defects. Last inspecting results are written to data table in order to statistic and query.(4) The method is adopted to evaluate the quality of glass-bulb on the basis of regulations by means of decision tree.