东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

基于Gabor小波的浮选泡沫图像纹理特征提取

期刊名称：仪器仪表学报
时间：2010.8.8
页码：1769-1775
分类：TP391.41[自动化与计算机技术—计算机应用技术;自动化与计算机技术—计算机科学与技术]
作者机构：[1]中南大学信息科学与工程学院,长沙410083
相关基金：国家自然科学基金重点项目（60634020）资助
相关项目：矿物浮选泡沫视觉图像处理方法研究

关键词：浮选泡沫图像, 纹理特征, GABOR小波, GAMMA分布, 聚类分析, flotation froth image, texture feature, Gabor wavelet, Gamma distribution, clustering analysis

中文摘要：

矿物浮选泡沫表面纹理是浮选工艺指标高低的重要指示器,为了获取泡沫图像纹理的细微差别进而对浮选生产状态进行机器识别,提出一种基于Gabor滤波的泡沫图像纹理特征提取方法。首先利用Gabor小波提取了泡沫图像多尺度与多方向上的泡沫纹理幅度谱（GMTR）与相位谱（GPTR）;然后根据GMTR与GPTR的分布特征,通过估计GMTR的Gamma分布参数和计算GPTR的熵作为泡沫纹理特征参量;最后,利用所提取的泡沫纹理特征对浮选工业生产状态进行无监督的模糊聚类分析与有监督的生产状态识别。实验结果表明,该纹理特征提取方法能有效地获取各种浮选状态下泡沫表面纹理的细微差别,基于该纹理特征参量的浮选状态识别准确率高于90%。

英文摘要：

The surface texture of flotation froth is an important indicator of flotation production performance,a method for froth image texture extraction based on Gabor filter is presented to obtain the distinctive representation of flotation froth images texture and then try to identify the flotation production states using the extracted texture parameters.First,two Gabor convolution features,i.e.Gabor magnitude texture representation （GMTR） and Gabor phase-based texture representation （GPTR） are captured using the convolution between multi-scale and multidirectional Gabor wavelet function and the froth images And then the distribution parameter of GMTR is characterized using Gamma density function and the entropy of GPTR is calculated;the distributions of GMTR and GPTR in all subbands are considered,and both of them are treated as texture feature parameters.The froth images captured from industrial field were used in experiments;the texture variables were extracted using this method and the industrial production states were clustered using unsupervised fuzzy clustering analysis and recognized using artificial neural network.Experiment results demonstrate that this method can extract the distinctive froth texture in various flotation states and the flotation production states recognition rate is grater than 90% using these texture parameters.