Research And Simulation On Image Recognition Of Tobacco Impurities

Solving the problem of foreign matter in tobacco is a key point to reduce the harmfulness of smoking and improve the cigarette quality.Currently,the mainstream system of tobacco foreign matter removal system is based on machine vision which is not very effective for the accuracy to handle the similar color foreign matter and can be greatly improved.In this thesis,an airflow machine based on computer vision for foreign matter removal has been designed.In the matter of machine vision core module,two different image processing algorithms are designed.One is based on color classification,and the other is based gradient energy and also the combination of two of them.Several difficult problems such as bad image quality,uneven illumination,dark foreign matter region are considered in the algorithms.Experimental results show that the impurity-removing efficiency of color classification can reach 99% and the impurity-removing efficiency of gradient energy can reach about 85%.The combination of two algorithms can reach 100% theoretically and can remove impurity effectively,however,algorithms will be redesigned when the target and environment has changed.The thesis then introduced a machine learning algorithm based on Cascade Adaboost.For the multiple windows problem,a window fusion module is designed.For the feature selection,both HOG and LBP are compared with Cascade Adaboost training and prediction.The experiment results show that HOG feature is more suitable for distinguishing between tobacco foreign matter and non-tobacco foreign matter,the impurity-removing efficiency of which can reach 99% and the machine learning method is equivalent to the image processing methods,but it can better cope with environmental and foreign matter changes and more convenient for product system maintenance.