PET Bottle Recognition And Sorting Based On Machine Vision

As energy conservation,environmental protection measure,post-consumer PET bottle recycling is attracting more and more attention.Sorting by color is a common way for bottle recycling.Manual operation is instable and time consuming,So automatic recycling facility based on machine vision is designed to replace human.However,due to the low accuracy of bottle sorting,the automatic recycling facility needs to be improved.In response to the above problem,this article propose an improved PET bottle recycling method.The first chapter introduces the hardware structure of a common vision inspection system,and summarizes the related researches in shape matching and color classification.And we give the research content,and introduce the organization structure of this thesis.The second chapter proposes an efficient approach for ROI extraction based on double-threshold segmentation.The background color is modeled to determine the low-threshold and high-threshold.Double-threshold segmentation is applied for the ROI extraction.The third chapter proposes a shape matching method for overlapped bottle recognition.The distribution of points on bottle contour under polar coordinate is used for extract shape descriptor,and the SVDD one-class classifier is trained for outlier detection.The fourth chapter proposes a color classification method for bottle sorting based on color.The K-means cluster is used for getting the standard color of the different kinds of bottles.The color of pixels of the bottle's region is replaced by the standard colors,and the proportion of the standard colors is used as the color descriptor.The fifth chapter describes the hardware structure and workflow of the PET bottle sorting facility.Then introduces the experiment result of overlapped bottle recognition and color classification.The sixth chapter summarizes the achievements of this dissertation,and discusses the area which needs to be improved and further research.