Flotation Froth Image Classification And Recognition Based On Local Features

It has became a common method to take advantage of digital image processing technology to to obtain flotation performance information in recent years.Different kind of texture,clarity,color and other characteristics of froth image always mean different kind of flotation situation.While much dust,light noise features of froth images make it hard to find an effective algorithm to distinguish different kind of froth image according to its corresponding flotation performance. This paper try to find an effective froth image classification method based on local features.The main research work and innovations achievement of this paper are as follows:(1)The flotation froth image classification methods based on froth bottom characteristics have the problems of low accuracy because of noises. So a flotation froth image classification method based on local features is presented to solve these problems. Referring to the method of text classification,the froth image is firstly divided into blocks,extracted these blocks’bottom characteristics and clustered to build the table of the froth status words. Based on the table,the similarity between the words and the word frequency are calculated,and then the froth image is described with a bag-of-word vector. Finally,the classification and recognition of real time froth image is realized based on VSM method.(2)To take full advantage of expert knowledge,a Bayesian probability model is being used,and then to estimate the probability model parameters with EM algorithm.Based on this,an image classification identification method based on Bayesian probability model is presented. The performance of this method is better than VSM method for takes full advantage of expert knowledge.(3)Considering the dynamic characteristics of flotation froth show some flotation performance information.A local feature matching method based on Scale Invariant Feature Transform (SIFT) is given to obtain dynamic characteristics parameters.(4)After the analysis of the relationship between all types of characteristic parameters and flotation performance of Sulfur,a froth image hierarchical classification and recognition system based on local static and dynamic features is given.Using industrial data for system validation,the experiment results have shown that the established system has good classification accuracy.This classification system can dived froth images into different kinds according to flotation performance. Figures(52),tables(8), references(62).