煤泥浮选泡沫图像纹理特征的提取及泡沫状态的识别

EXTRACTION OF TEXTURAL FEATURE AND RECOGNITION OF COAL FLOTATION FROTH

用煤泥浮选泡沫数字图像获取系统获取了 5 1幅煤泥精矿泡沫图像 ;引入了空间灰度相关矩阵和邻域灰度相关矩阵来提取泡沫的纹理特性 ,并提取基于这两种算法的一系列特征参数来描述泡沫的结构 ;分析了各泡沫特征参数随浮选时间 (泡沫纹理 )的变化关系 ,定性地指出了各泡沫特征参数与泡沫纹理的相关性 ;并利用自组织神经网络对煤泥浮选泡沫的状态进行了识别 ,分类识别的平均正确率达 76 .5 % .

By conducting a series of coal batch cell flotation experiments, a number of digital froth images are captured. Two algorithms--the spatial gray level dependence matrix (SGLDM) and the neighboring gray level dependence matrix (NGLDM) are introduced to extract the visual textural characteristics of coal froth images. Based on these two matrixes, a series of textural features such as energy ( E ), entropy ( ENTS ), inertia ( I ) of SGLDM and small number emphasis ( Fine ), large number emphasis ( Coarse ), entropy ( ENTN ), second moment ( SM ), number nonuniformity ( NN ) of NGLDM are proposed to describe the coal froth textural characteristics. By using the software developed by the author with DELPHI language, the textural features of flotation froth images captured in laboratory experiments are extracted. The change tendency of each feature with flotation time is analyzed, and the relationship between each feature and froth textural feature is pointed out qualitatively. It is found { E, ENTS, I } of SGLCM and { Fine, Coarse, ENTN } of NGLDM really reveal the variation tendency of image texture of coal froth. However, the { ENTN, SM,NN } of NGLDM have little relationship with image textural characteristics of coal flotation froths. Choosing the reliable textural features { E, ENTS, I, Fine, Coarse, ENTN } as the input to a set of neural network called self-organizing feature mapping, all the images are classified into four patterns, and the average correct rate is 76.5%.