The Oven Flame Sequence Image Color Feature Extraction Method

Steelmaking process is a complex multiphase physical and chemical process,different molten steel composition and temperature of different wavelength andfrequency spectrum, in order to realize effective control of the extracted features morecomprehensive, more representative, is to improve the robustness of control, in orderto make the Image feature extraction, the time it takes to meet the requirements ofreal-time control, often need to convert three-dimensional color Image to atwo-dimensional digital Image, shortening the time of Image feature extraction, iscurrently the most widely used Multivariate Image Analysis (Multivariate ImageAnalysis, MIA) method, the core of this paper, using the method of MIA PrincipalComponent transformation (Principal-Component Analysis, PCA) of the3d Imagedata can be converted into2d, in2d plane design multiple fixed size of windows,usearea clustering analysis method to determine the size of the window, direct extractionflame logo color features. This is the innovation of this paper, the selected color in thewindow of a single pure color, which is different from the past, the entire image isextracted random mix of colors, so you can better reflect the variation of color flamesmelting process.Adopting different furnace smelting of the same type steel gradewere analyzed to illustrate the effectiveness of the method. In terms of frequencyfeature extraction, the method of using wavelet packet analysis can be a very goodfrequency transformation characteristics of the flame.In this paper, the work includes the following aspects:(1) Selection of consistency characteristic vector. Feature vector is determinedby the covariance matrix. Before the end of the study of all images, oxygen blowing, athird of the image, flame color clear image and symbolic color images of four ways todetermine the covariance matrix and its corresponding feature vector, and based onthe principal component t1, t2value scope of disparity consistency to select featurevector.(2) The flame image color feature extraction. Of the sequence image byconsistency characteristic vector MPCA principal component transformation (t1, t2)corresponding to disperse point figure (256x256histogram). Clustering method is used to determine the color window size, by color window in figure the scatteredpoint extraction sequence image color pixels, quantity characteristics, thetransformation of the time before and after the color transforms from within the classand class spacing, that color clustering performance before and after imagetransformation has consistency. Experiments show that the extracted by the method ofclustering analysis, color features can well reflect the physical chemistry of convertersteelmaking process.(3) The fire frequency feature extraction. Through the selection of waveletpacket decomposition, the parameters of the wavelet packet, the decomposition ofeach color of refactoring and calculate the reconstruction of node energy, finally getthe frequency characteristics of the flame front.The system detailed steelmaking end-point control of each link in the process ofthe flame image feature extraction. Results showed that the three dimensional colorimage data after PCA, consistent color clustering performance before and after itstransformation, the transformation of2data extraction respectively symbol colorfeatures is feasible; Using cluster analysis method to determine logo color window,through the color window to extract logo color features can reflect the smeltingprocess of the flame; Different logo color features reflect different physical andchemical processes in the process of smelting, respectively than MIA logo colorfeature extracting method to extract the comprehensive characteristic has moreabundant information, is helpful to use the end of the flame image control ofsteelmaking.