PET Bottle Neck Defect Detection Based On Sobel And Least Square Circle Fitting

PET,called Polyethylene terephthalate,is widely used in daily life in all its aspects because of its excellent physical-mechanical properties?PET bottle is one of the important links and mainly used in the food and beverage industry?Blowing method is used to process PET bottles and bottle areas prone to bumps,scratches and other defects,have an impact on the appearance and quality of the product,so necessary for product testing.At present,the international package including Cognex,Keyence,most manufacturers of products are bottle defect detection on glass bottles,even a bottle of PET products,but also because of the complexity of the algorithm and the height of the customization can not be applied to different scenarios.Domestic visual industry on the PET bottle mouth detection products are few,technically not mature enough.The research team is mainly responsible for the two aspects of the illumination design and algorithm of the visual system.The illumination design is mainly to ensure that the features of the bottle neck are highlighted.At present,the detection efficiency of the whole system has reached 18000 bottles/hour,and the detection accuracy is as high as 98%.The main innovative features of the paper lie is:?The detailed analysis of the characteristics of PET materials and production process,summarizes the possible causes of defects,and according to the optical properties,the blue ring light with angle as a light source,while highlighting the defects brought by burr surface to reduce interference.?We propose a supervised least squares circle fitting method based on improved Gauss filter,edge extraction,after sampling operation,re fitting,has greatly improved in accuracy and efficiency.?The traditional Sobel operator is improved by introducing attenuation coefficients and directional information,so that Sobel operator can be applied to circular edge detection.According to the gradient transformation mouth region,can effectively distinguish between the defect and the burrs generated by the shaded area,compared with the traditional detection methods,the detection error rate is greatly reduced.?By comparing and analyzing different classification methods,the defects are merged and separated.The traditional algorithms usually judge the defect size directly,but ignore the situation that some defects are scattered into different defects due to uneven illumination.The application of the check set improves the robustness and precision of the algorithm.