信息熵多属性约简的煤粉尘图像特性机理

(西安科技大学 电气与控制工程学院,陕西 西安 710054)

安全科学与工程; 图像灰度特征; 信息熵; 模糊类别隶属度; 多属性约简

Mechanism of coal dust imagery characteristics based on information entropy multi-attribute reduction
WANG Zheng,WANG Mei

(College of Electrical and Control Engineering,Xi'an University of Science and Technology,Xi'an 710054,China)

safety science and engineering; image grey feature; information entropy; fuzzy membership; multi-attribute reduction

DOI: 10.13800/j.cnki.xakjdxxb.2019.0421

备注

为研究无明确特征模式的煤尘颗粒图像特性,以某煤矿煤样为研究对象,按国标标准运用粉尘采样器对粉尘溢散源处颗粒物进行多点采样。采用多决策属性约简模糊粗糙集3个阶段即提出隶属度模型、实现属性约简、确定最大信息熵阈值分割对颗粒形态特征机理进行分析。首先建立粉尘图像各像素点对应的模糊类别隶属度模型,利用多分段函数确定隶属度; 分析煤粉尘图像灰度特征并将其作为条件属性,确定条件属性的模糊依赖度,获取最优值并提取模糊属性约简,进行目标及背景区域的模糊下近似和模糊上近似划分; 最后建立煤粉尘颗粒的信息熵模型,存储信息熵并实现对分割阈值的提取。结果 表明:依据模糊属性约简的互异重要度可实现多属性约简; 并确定煤粉尘图像模块区域的最大信息熵分割阈值。所建立模型可删除冗余属性,选择出对分类更为重要的属性,并通过属性约简完成特征选择分类。

To investigate imagery characteristics of coal dust particles without clear characteristic mode,coal samples from a coal mine were taken as research objects,and the dust sampler was used to conduct particles multi-point sampling at dust spill source according to the international standard.The multi-decision attribute reduction fuzzy rough set,including three stages of the membership model,realizing attribute reduction,and determinating maximum information entropy threshold segmentation,are adopted to analyze the particle morphology characteristics.The corresponding fuzzy degree membership model was established for dust image pixels and meanwhile the membership coefficient was determined by multi-segment function.In additon,the gray feature of coal dust image was analyzed and used as conditional attribute so that the fuzzy dependence of conditional attribute can be determined to obtain its maximum value and extract the fuzzy attribute reduction.At the same time the fuzzy lower approximation and the upper approximation in the target and its background regions were divided.Finally the information entropy model of coal dust particles was established with the information entropy stored and its corresponding segmentation threshold extracted.The results show that multi-attribute reduction can be realized according to the mutual importance of fuzzy attribute reduction; and the maximum information entropy segmentation threshold of coal dust image module area is determined.The established model,therefore,can delete the redundant attributes,select more important classification attributes,and complete the feature selection classification through attribute reduction.