基于半监督聚类的煤泥浮选泡沫图像分类方法

Coal slime flotation foam image classification method based on semi-supervised clustering

  • 摘要: 针对选煤厂煤泥浮选过程加药量依靠人工干预存在主观性、滞后性和粗放性的问题,提出了一种基于半监督聚类的煤泥浮选泡沫图像分类方法。首先,采集已知加药比例与未知加药比例下的煤泥浮选泡沫图像样本,并对泡沫图像进行预处理,提取泡沫的气泡个数、气泡面积、气泡周长等形态特征;然后,对已知加药比例下泡沫图像形态特征样本进行标志,对未知加药比例下泡沫图像形态特征样本不做标志,并将已标志泡沫图像形态特征样本与未标志泡沫图像形态特征样本进行混合;最后,利用基于高斯混合模型的半监督聚类方法对混合样本进行聚类后得到各类簇,将各类簇内已标志泡沫图像形态特征样本的信息映射到未标志泡沫图像形态特征样本。应用结果表明,该方法可为煤泥浮选生产过程加药量调整提供指导,降低了药剂消耗量,提高了选煤厂浮选自动化水平和经济效益。

     

    Abstract: In order to solve problems of subjectivity, hysteresis and extensiveness existed in reagent amount addition of coal slime flotation in coal preparation plant depended on manual intervention, a coal slime flotation foam image classification method based on semi-supervised clustering was proposed. Firstly, coal slime flotation foam images under known reagent-addition ratio and unknown reagent-addition ratio are collected, and the foam images are preprocessed to extract morphological characteristics such as bubble number, bubble area and bubble perimeter. Then, foam image morphological characteristic samples under known reagent-addition ratio are marked, while foam image morphological characteristic samples under unknown reagent-addition ratio are not marked, and the marked foam image morphological characteristic samples and the unmarked foam image morphological characteristic samples are mixed. Finally, semi-supervised clustering method based on Gaussian mixture model is used to cluster the mixed samples, so as to get various clusters, and information of the marked foam image morphological characteristic samples is mapped to the unmarked foam image morphological characteristic samples in various clusters. The application results show that the method can provide guidance for adjustment of reagent-addition amount in coal slime flotation production process, reduce consumption of reagent, and improve flotation automation level and economic benefit of coal preparation plant.

     

/

返回文章
返回