Visual Inspection Method For Glass Bottles In Beverage Intelligent Production Line

The beverage production scale of China is the first in the world.And beverage packaging materials are different,including plastic bottles,glass bottles,metal cans and paperplastic composite materials.Glass bottles are widely used packaging materials,especially,in the beer packaging industry,because there are many advantages such as good sealing performance,reusability and low price.Foreign objects and dirt may lie on the surface of bottles.Moreover,some broken may exist for the finish,body and bottom of bottles because glass bottles are easily broken by the collision and friction during transportation and production.However,if glass bottles with dirt,foreign objects or damages are used for packaged that may lead to the production line malfunction which affects production efficiency.Worst of all,it may even endanger consumers' lives and health.Therefore,before using glass bottles for packaging,the quality of glass bottles must be strictly inspected.Studying the visual inspection technology in the beverage production line,guaranteeing the quality of beverage products,speeding up the transformation and upgrading of the beverage automation production line to the intelligent production line,realizing the intelligent manufacturing of beverages and meeting the needs of beverage manufacturers,is an important part of ?Made in China 2025?.This paper focuses on the practical questions of quality inspection of glass bottles in beverage production lines,and the main contribution is given as follows:(1)The structural characteristics of beverage intelligent production line and glass bottles are analyzed,the advantages and disadvantages of various imaging mechanisms summarized,and the glass bottle imaging scheme is designed.Secondly,the electrical control system and the visual inspection software system for glass bottles are developed.Finally,the entire system of glass bottle machine vision inspection platform is assembled and test.(2)A bottle mouth localization algorithm for multiple random circle detection and round fitness evaluation is proposed.The edge points are obtained by threshold segmentation,centroid method and radial scanning.A circle is determined by three points randomly sampled from the edge points.The ratio of the number of edge points whose distance from each edge point to the circle is less than a given threshold to the total number of edge points is defined as the circle fitting degree.Circle fitting degree is taken as evaluation criterion for searching the optimal location result.Many circle fitting results and circle fitting degree are obtained by the multiple random circle detection.The circle fitting result with the maximum circle fitting degree is taken as the bottle mouth center.To further improve the positioning accuracy,a new single circle detection method based on polar coordinate space model fitting and least squares circle detection is proposed.The mathematical model of circle in the polar coordinate space is established and noise points are filtered by this model.Then,the final circle detection result is got by least squares circle detection.Experiments prove that both of the proposed methods can achieve high-speed and high-precision positioning of bottle mouth when the bottle mouth are seriously damaged.The former is faster and the latter is more accurate.(3)A defect detection method for bottle mouth based on residual analysis dynamic threshold segmentation and global threshold segmentation is proposed.All kinds of defects features are analysed.Three-circle positioning method based on random evaluation for determining the accurate position of the interest for region of bottle mouth is proposed to improve the anti-interference ability and positioning accuracy.And a novel defects detection method that combined residual analysis and dynamic segmentation with global threshold segmentation is proposed to overcome the effects of image gray change and bottle mouth broken.Experiments prove that,compared with five traditional approaches,the proposed method improves the accuracy of defect detection,and can achieve rapid and accurate detection of defects in bottle mouth images with strong interference and large defects.(4)A surface defect detection framework with improved geodesic distance transformation and template matching is proposed.The Hough transform circle detection is combined with the prior size of the bottle bottom to realize the positioning of the bottle bottom,and then the bottle bottom is divided into three detection regions: a center panel region,an annular panel and an annular texture region.To highlight the differences between defects and background regions,and to overcome the influence on defect detection from the different bottle bottom image gray scale transformation range,an improved geodesic distance transformation saliency detection method is proposed to realize defect detection in the center plane of bottle bottom.Multiscale mean filtering is used to detect defects in annular panel.Template matching and multiscale mean filtering are combined to realize defect detection in circular texture region.Three image test datasets of bottle bottoms are constructed to evaluate the performances of algorithms.Experiments prove that the proposed method can accurately detect defects with small size and low contrast.(5)A surface defect detection framework based on saliency detection and wavelet transform for bottle bottom is proposed.The input image is first downsampled to reduce computational complexity.And an improved random circle detection algorithm combined with least squares circle detection and entropy rate segmentation to achieve high-speed and high-precision bottle bottom location is proposed.According to the bottom structural characteristics,bottle bottom is divided into two detection regions: central panel region and circular texture region.A defect detection method,which integrates Frequency-Tuned saliency detection,anisotropic diffusion and improved superpixel segmentation,is proposed.Defects is highlighted by saliency detection and anisotropic diffusion.Each defect region is clustered as a whole as possible through superpixel segmentation.The defects in the central panel are recognized by the features of areas and saliency values.To further suppress the texture effect and improve the robustness to location errors,a method based on wavelet transform and multi-scale filtering algorithm is proposed to detect defects in annular texture region.Experiments prove that the proposed method overcomes the influence of texture interference and location error of the bottle bottom,and further improves the accuracy.(6)A fast and robust bottle wall detection method with binary template matching is proposed.Considering the structure information is the most valid information for the template matching.However,the resolution reduction has only little effect on the structural characteristics of the detected bottle wall and template.Therefore,to speed-up the algorithm,the input image is down-sampled,and the bottle wall or bottleneck is taken as the template.The downsampled image is binarized,and then the binary template matching is performed to obtain the midline position of the bottle wall.To highlight the edge area on the bottle mouth,a novel filtering kernel function is proposed to filter the image of the bottle wall.The coordinates of the top edge point of the bottle mouth is obtained by segmentation and scanning.Two test data are constructed.The proposed method compares with many other methods.Experiments prove that the proposed method effectively overcomes the problem of inaccurate location when there are multiple bottle walls in the same view image,and meets the on-line detection requirements.In summary,this paper takes the actual demands and problems of visual inspection of glass bottle quality in beverage intelligent production line as guide.A visual inspection system of glass bottle is first developed.Meanwhile,many new visual inspection methods for locating and surface defects are proposed.Moreover,many experiments(test data and algorithm codes are available)are implemented on the images acquired by the designed system.Experiments prove that the proposed methods can overcome some problems in the visual inspection of beverages.The research results of this study are important in theoretical significance and engineering application in the relevant fields.