Research On The Modeling Method Based On Clustering Algorithms For Spatial Distribution Of The Color Of Tobacco

The recognition and elimination of foreign body in tobacco is a very important part in cigarette production.With the development of science and technology,tobacco online recognizing and eliminating equipment replaces manual operation gradually.The efficiency of elimination is mainly determined by the precision of online recognization algorithm.Therefore,it is of great significance to propose an effective method for foreign body recognition.Tobacco as the research object,a modeling method based on layering clustering algorithm is proposed to establish model elaborately for spatial distribution of the color of tobacco.The model is used as a reference for the foreign recognition,which provides a reliable method to recognize foreign body in tobacco.Firstly,the image of tobacco is processed to devise tobacco and background in order to avoid interference of background pixels and make the model more accurate.Compared with some traditional background segmentation methods,a background removal method based on the characteristic of tobacco image is put forward in this paper.It is demonstrated that this method is effective through experiments.Secondly,Some improvements about traditional K-means algorithm is proposed.This algorithm has two main shortcomings: the value of K is given by user;and it is sensitive to initial centers.An improvement algorithm based on layering clustering is proposed in this paper to eliminate these inadequacies of K-means algorithm.A new layer attribute of data set is established and the data set is divided into high and low layers according to the value of the layer attribute.Clustering from the top to the lower layer,K value and the clustering centers are adjusted dynamicly.Also data is input by the value of layer attribute and the deficiency of the traditional algorithm is improved.Finally,a modeling method based on layering clustering is proposed.The RGB color space of tobacco is divided into several subspaces in which the pixels of tobacco leaf are processed including dimensionality reduction,sorting and projection.Tobacco data set in the B subspace is clustered by the improved algorithm and an effective model of spatial distribution of tobacco color is established.This model provides a feasible method for the tobacco foreign recognition.