基于信息熵和组合纹理特征的熟料状态检测

Clinker status recognition based on information entropy and hybrid texture features

基于信息熵和组合纹理特征提出了回转窑熟料烧结状态检测方法。利用信息熵对窑头图像序列进行分析以消除粉尘干扰导致的坏帧;对提取出的熟料感兴趣区计算了隶属于纹理统计学方法的灰度直方图法、灰度共生矩阵、灰度游程矩阵、灰度梯度法共32个统计学纹理描述子,结合MI互信息参数确定了10个区分度高的纹理特征待选集;基于K-NN分类器实现了过烧、稍过烧、正烧、稍欠烧和欠烧5种不同烧结状态的熟料检测。实验结果表明该方法仅用灰度共生矩阵的SA和灰度游程矩阵的Vert_LRE这2个纹理参数熟料状态动态识别率达到86.5%,对实现回转窑熟料烧结状态检测具有较好的实用价值。

A method based on information entropy and hybrid texture features is proposed to predict the sintering status of the clinker in rotary kiln. Aiming at smoothing out images which are blurred by heavy dust, the image sequence analysis is implemented by information entropy. 32 statistic texture features of gray-level histogram, gray-level co-occurrence matrix, gray-level run-length matrix and gray-level gradient are calculated for clinker ROI. Then the top 10 discriminable features are extracted by MI as candidates for K-NN to classify clinker samples into different categories. The experiment results demonstrate that only two parameters, SA of gray-level co-occurrence matrix and Vert_LRE of gray-level run-length matrix, are needed for the status dynamic recognition of clinker and the accuracy is 86.5%. This method is significant for the sintering status recognition of clinker in rotary kiln.