Research On Tobacco Foreign Body Recognition Algorithm Based On Color Model

With the progress and development of high-performance and high-power computers in recent years,solving the problems in reality based on computer vision has become the focus of some enterprises.In the cigarette manufacturing enterprises also have a broad application,it can improve the level of production automation and reduce labor costs.This paper uses computational vision to study tobacco body recognition algorithm based on tobacco leaf color model,aim to provide tobacco companies with solutions for foreign body identification.Since the recognition algorithm is based on the color model,making the color model better describe the color information of tobacco leaf is very important.First of all,we need to calibrate the tobacco image using statistical histogram to get the qualified tobacco image.Then segmented the image,including Otsu algorithm,region growing method and clustering algorithm.We also propose the segmentation method is suitable for the tobacco images and effectively resolves the over-segmentation phenomenon and the change of image quality.Then,using the hierarchical clustering method to create a color model of tobacco leaves.The research of foreign body recognition algorithm is an important part of this study.We proposed two foreign body recognition algorithm,a class-based standard deviation algorithm,SVDD and look-up table algorithm based on color model.The former extracts the one-dimensional distance feature from the color model,and then uses the distance to compare the standard deviation of the corresponding class to judge the pixel.The latter algorithm extracts another feature data,and training the SVDD classifier.In order to improve the recognition speed of algorithm,we established a look-up table.Then taking the image unit division,through the appropriate threshold for each unit set to determine the image foreign body area and mark it.Finally,the algorithm is tested by foreign body image.Experiments show that the recognition rate of the algorithm is higher than 95% and the feasibility and reliability of the algorithm is verified.At the same time.A tobacco foreign body identification system is designed.The function of each processing module of the system and the user interface are also introduced.