Research On Machine-Vision Based Online Inspection System Of Glass Bottle

Glass bottle plays an important role in the package industry. The inspection of glass bottle is an important task. In China, the inspection work is mainly executed artificially or half-artificially, witch is not up to the new standard of non-touch producing. Up to now, inspection machines have been produced in China, but are under-developed on global level. More research should be done on the issue of bottle inspection, especially on accurate defect recognizing, positioning, fast processing, etc.Machine vision is the most advanced technology in bottle inspection area. It can be done via grabbing images on-line, classifying defects, selecting the bad products, improving figures for different types of bottles.The object of this article is to design and accomplish a bottle inspection system based on machine vision technique. The main contents of this article are as follows,1. Surveyed and showed how the area of bottle inspection was going on.2. Developed three algorithms in digital image processing. Firstly, it introduced the Integral Images method for mean filtering, and improved it to arbitrary filters. Secondly, it used special filter for dynamic thresholding. The new thresholding could ignore the module lines and screw threads, which may lead to miss-selection of good bottles. Thirdly, it created a new registration algorithm which worked well on bottle images.3. Introduced machine learning method to defect classification. Train and testing experiments were done, showing that classifiers like Bayesian, ANN, and SVM would be helpful to improve the intelligence level of the inspection system.4. Designed and accomplished the machine vision based bottle inspection system, which is now running on-line successfully as expected.