【作者】 任传成；

【导师】 杨建国；

【作者基本信息】 中国矿业大学 ， 矿物加工工程， 2014， 博士

【摘要】 铜精矿品位是硫化铜浮选生产过程参数之一，也是一项重要的浮选产品质量指标。现有的铜精矿品位检测方法存在着主观性强、检测周期长、数据校正复杂、成本较高等不足之处。本文以硫化铜精矿为研究对象，结合彩色图像处理方法，系统研究了硫化铜精矿在固相、固液两相、气固液三相体系中的彩色图像特征，揭示图像特征与铜精矿品位之间的变化规律，建立基于图像特征的精矿品位预测模型，为浮选生产的过程控制提供必要的技术支持。主要研究内容和成果如下：建立了基于硫化铜精矿粉末彩色显微图像颜色特征的品位预测模型。搭建了实验室彩色显微图像采集装置并获取硫化铜微粉图像；提出了色调保持不变的彩色图像增强方法，有效地实现彩色显微图像的降噪和增强处理；采用统计方法提取彩色显微图像的红色、绿色、蓝色、色调平均值、彩色向量角等颜色特征参数，建立了3个基于图像颜色特征的LS-SVR法的品位预测模型；评价模型的预测性能，结果表明，基于色调平均值的铜精矿品位预测模型为最佳。建立了基于硫化铜矿浆彩色图像特征的品位预测模型。针对硫化铜矿浆图像采集问题，设计了一套矿浆彩色图像采集试验装置和方法；提出了矿浆彩色图像的裁剪和增强等预处理方法，采用颜色比率和相对颜色度方法提取矿浆图像颜色特征，首次引入Tamura方法提取图像V分量纹理特征，然后利用相关系数方法对颜色和纹理特征进行降维；依据多元线性回归和GRNN方法研究矿浆图像特征和品位之间的内在关系，结果表明，基于GRNN法的品位预测模型的预测精度均优于基于多元线性回归法的品位预测模型的预测精度，且基于矿浆彩色图像纹理特征的GRNN的品位预测模型为最优。建立了基于硫化铜浮选泡沫彩色图像特征的铜精矿品位软测量模型。搭建了一套浮选泡沫视频图像采集试验装置，实现了硫化铜粗选和精选过程的泡沫彩色图像采集任务。研究了粗选和精选泡沫彩色图像的裁剪、去模糊化、降噪、增强等预处理方法，依据颜色直方图、颜色矩、相对颜色度等方法提取泡沫图像颜色特征，分别采用Tamura方法、WPT结合Tamura方法提取泡沫图像的H、S、V分量纹理特征；提出了一种多层聚类结合Lasso法的图像特征参数降维方法，并采用相关系数硬阈值法择取模型的辅助变量；采用多元线性回归、PLS、LS-SVR方法分别建立基于粗选和精选泡沫图像特征的品位软测量模型，评价这些模型的预测性能，结果表明，在硫化铜粗选过程中，基于粗选泡沫彩色图像颜色和纹理特征组合的LS-SVR的铜精矿品位软测量模型的预测精度为最优；在硫化铜精选过程中，基于精选泡沫彩色图像颜色特征的LS-SVR的铜精矿品位软测量模型的预测精度为最优；利用彩色图像处理方法可对铜精矿品位进行预测。更多还原

【Abstract】 Copper concentrate grade is not only an important parameter in copper sulfideproductive process, but also a significant quality index of flotation product. There are manyshortages in copper concentrate grade detection methods, such as strong subjectivity, lowaccuracy, long detection period, data correction, high cost and so on. Copper sulfide was usedas an object of study in this paper. Based on color image processing methods, color imagefeature of copper sulfide concentrate was systematically studied in solid phase, solid-liquidphase, gas-solid-liquid phase environment. Relations between image feature and copperconcentrate grade was revealed and concentrate grade prediction model based on imagefeature was developed. This research could provide theoretical support for further study aboutcopper concentrate grade detection methods and technical support for process control inflotation. Main research contents and conclusions are as follows:Grade prediction model based on color microscopic image feature of copper sulfideconcentrate powder was developed. For the problems of fine particles in copper sulfide testsamples, a color microscopic image acquisition device was built, then a hue-preserving colorimage enhancement method was proposed to denoise and enhance color microscopic imageeffectively. Color microscopic image features such as color vector angle, average red, green,blue, and hue values were extracted by statistical approach. Furthermore, three concentrateprediction models were developed, base on LS-SVR method and image color feature.Comparative results of three models’ predictive performance indicated that copperconcentrate grate prediction model based on average hue value was an optimal model.Grade prediction model based on color image features of copper sulfide pulp wasconstructed. For image acquisition problem of copper sulfide pulp, a pulp color imageacquisition device and method was designed. Cropping and enhancing preprocess method forpulp image was studied. Color features of pulp image were extracted by color ratio andrelative color degree methods. Tamura method was firstly introduced to extract texturefeatures from V component, then dimensionality reduction for color and texture features wasconducted through correlation coefficient method. Moreover, linear and non-linear relationsbetween pulp image features and copper grade were studied by multiple linear regressionmethod and GRNN method. Study results showed that grade prediction model based onGRNN method had higher prediction accuracy than model based on multiple linear regressionmethod, and grade prediction model, based on GRNN method and texture features of pulpcolor image, was an optimal model. Copper concentrate grade soft-sensor models based on froth color image features incopper sulfide flotation process were developed. A video-image acquisition device forflotation froth was developed in order to accomplish color image task of rougher froth andcleaner froth of copper sulfide. Preprocess methods for rougher and cleaner froth, such ascropping, deblurring, denoising and enhancement, was studied. In addition, color histogram,color moments, relative color degree methods were used to extract color features from frothcolor image. And texture features of H, S and V component were extracted by Tamura methodand WPT combined with Tamura method. Then, dimensionality reduction method, which wasbased on multiple clustering method and Lasso method, was applied to image featureparameters. Combined with hard threshold method of correlation coefficient, secondaryvariable of soft sensor model was selected. Finally, through multiple linear regression method,PLS method and LS-SVR method, soft sensor model of copper concentrate grade weredeveloped, which based on image features of rougher and cleaner froth. Comparative resultsof these models’ predictive performance, results showed: in copper sulfide rougher flotationprocess, the soft sensor model of copper concentrate grade using LS_SVR method was anoptimal model, which was based on color and texture features of rougher froth image; incopper sulfide cleaner flotation process, the soft sensor method of copper concentrate gradeusing LS_SVR method was an optimal model, which was based on color features of cleanerfroth image; it could be concluded the grade of copper concentrate would be predicted byusing color image processing method.更多还原