Extraction Algorithm Of Multiple Features And State Recognition Of Froth Images In Coal Flotation

Coal flotation is operated according to the difference in material in surface wet ability of coal and gangue particles.Thanks to effect in the flotation reagents,the hydrophobic coal particles with better flotability attached to the bubbles and floated in the flotation froth layer,while the hydrophilic gangue particles with poor flotability remained in the coal slurry into the process of flotation tailings.So coal was separated from gangue minerals.Although coal slime flotation process is a complex process affected by many factors,analysis of flotation process through extracting visual features from bubbles of flotation images such as bubble size,bubble texture characteristics and dynamic characteristics,the froth layer,and the parameters of tailings can been made in order to identify the flotation status such as fine coal grade.In view of froth images with large noise,low contrast bubbles get by the industrial CCD camera,a set of improved algorithms of image processing and feature extraction has been proposed and proved superior to others through simulation experiment on the basic of study and analysis in technologies of image processing and characteristic extraction.At the same time the RBF neural network to recognize flotation state has been optimized in design,which can made effective recognition for flotation state by the extracted feature vectors and has good generalization ability.Main research content,used method and conclusion of this paper have been listed as followings:(1)Many kinds of image de-noising methods like spatial de-noising methods,frequency domain de-noising methods and morphological image de-noising methods have been illustrated in detail.By analyzing and comparing image quality after various de-noising methods,low frequency filter based morphological reconstruction by opening and closing was designed in this paper.(2)The improved genetic algorithm was put forward to optimize the structure element of morphology filter through comparing and analyzing of advantages and disadvantages for genetic algorithm.The improved algorithm used similarity criterion by Hamming distance based XOR operation,but giving up the traditional information entropy as the similarity evaluation.It has been proved more time-saving,efficient,accurate;the algorithm adopted the method of adaptive mutation and joined a regulation factor of variation in order to improve traditional method of adaptive mutation;it combined genetic algorithm with immune algorithm and used combining evolutionary strategy;the strategy to retain elite was joined to the improved algorithm in order to remain the outstanding gene individual and to prevent the evolution backwards and to ensure the convergence of the algorithm.(3)Considering objective evaluation methods for image de-noising without reference,in evolutionary algorithm to optimize structure element,this paper presented improved de-noising evaluation index of improved information capacity based on gray level co-occurrence matrix for images without reference and furthermore it was regarded as fitness function of evolutionary algorithm at the first time.(4)In this paper,the image segmentation method was further studied,and an improved algorithm to extract internal markers has been proposed considering the characteristics of flotation bubble mixed inseparable,which is based on fusion of particle swarm optimization algorithm and one-dimensional histogram weighted based on fuzzy C-means clustering method.In the threshold optimization to binarize image based on particle swarm optimization algorithm,the image thresholds were optimized using 2-D maximum entropy as the fitness function based on gray level co-occurrence matrix.The simulation results proved the validity of this kind of segmentation methods,and furthermore demonstrated the accuracy for image segmentation applying this method after comparing this method with other general algorithm of watershed segmentation.(5)In view of the fact that coal flotation is a nonlinear complex process controlled by the coupled multi-variable,multiple features in flotation were extracted as input of prediction model.By correlation analysis of various texture features,size characteristics,flotation froth layer and so on with flotation indexes,some characteristics were selected as feature vector to determine flotation state.These characteristics include contrast,correlation,energy,entropy based on gray level co-occurrence matrix and roughness,fineness based on neighborhood gray level co-occurrence matrix and the bubble size based on the result of image segmentation and thickness characteristics of froth layer,which have been makes a detailed analysis.(6)The parameters of RBF neural network for state identification were optimized using fused algorithm of improved immune algorithm and fuzzy C-means clustering algorithm.Firstly,improved immune algorithm was used to determine the position of the center and the number of hidden layer of neural network.In which the improved algorithm of selection for initial population in immune and divided affinity threshold have been adopted.Furthermore,antibody deletion mechanism,antibody immune mechanism and antibody concentration regulation principle were joined.Secondly,the centers of the RBF hidden layer were further optimized by fuzzy C-means clustering algorithm.Through the sample verification,identification accuracy of optimized by this kind of methods is improved obviously,and RBF network has better generalization ability.