【作者】 牟学民；

【导师】 桂卫华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2012， 博士

【摘要】 自然界所蕴藏的矿产资源绝大多数需经选矿加工后才能使用。在常用的选矿方法中,泡沫浮选是最重要的选矿方法,几乎所有矿石都可用浮选来分选。泡沫浮选是发生在固、液、气三相接口上连续的物理化学过程,其中浮选泡沫层是关键,浮选泡沫表面视觉特征是浮选工况与工艺生产指标的直接指示器。一直以来,浮选过程操作都是依靠人工肉眼观察浮选泡沫表面状态来完成,这种仅凭人工肉眼观察进行浮选生产操作的方式主观性强、误差大、效率低,无法实现浮选泡沫状态的量化描述和生产工况的客观评价,造成浮选生产指标波动频繁、矿物原料流失严重、药剂消耗量大、资源回收率低等情况时有发生。自上世纪90年代以来,一些发达国家将机器视觉引入到浮选过程监控中,随后逐渐得到了各科研单位与选厂的高度重视。研究基于机器视觉的浮选过程监测、建模与浮选过程自动控制方法与技术,对提高矿产资源的利用效率、实现企业的可持续发展具有非常重要理论意义与应用价值。浮选泡沫表面视觉特征的准确提取是实现基于机器视觉的浮选过程自动控制的关键。泡沫表面视觉特征表现在两个方面：一是基于单帧图像的泡沫参数,称为静态特征,比如泡沫的大小、形状、颜色特征、纹理特征等；另一方面是基于图像序列的泡沫图像特征参数,称为动态特征,一般包括泡沫运动速度、泡沫稳定度等。目前,国内外学者主要研究了浮选泡沫图像静态特征的提取方法与技术,对于动态特征的提取并没有详细说明和分析。要实现浮选泡沫的视觉测量与泡沫状态的客观评价,浮选泡沫的动态特征是必不可少的特征参量。本文结合浮选工艺机理,分析浮选生产工况与泡沫表面动、静态视觉特征的内在联系,指出在各种浮选生产工况下的泡沫视觉图像的特点,重点研究浮选泡沫表面动态特征提取算法和量化表述方法及其在矿物浮选过程泡沫图像监控系统中的应用。论文主要研究工作及创新点如下：(1)针对浮选泡沫图像噪声大,浮选泡沫表面视觉特征不能准确提取的难题,提出了一种时空联合的基于泡沫图像序列的多尺度几何变换图像贝叶斯去噪方法。该方法采用Curvelet变换对泡沫图像序列中矿化气泡的几何结构进行稀疏表示；通过分析变换域系数的统计相关性,建立基于单帧图像和时空域多帧泡沫图像Curvelet域统计依赖模型和联合统计分布模型；以所获取图像变换域系数统计分布模型为先验知识,采用贝叶斯准则依据一定的代价测度得到变换域系数的最优估计,获得理想的无噪声污染的泡沫图像最优系数估计；最后进行Curvelet反变换获得去噪的泡沫图像。该方法和其它去噪算法相比能获得较高的峰值信噪比,并在去除噪声的同时较好地保持泡沫图像细节,为后续泡沫特征提取提供了高质量的处理图像。(2)针对矿化气泡在流向刮板过程中不可避免地发生角度旋转与尺度缩放等几何畸变的特性,常用的目标跟踪方法因难以实现对几何形态畸变严重的浮选泡沫的精确跟踪而无法准确获取浮选泡沫表面流速的难题,提出了一种自适应的基于Fourier-Mellin变换与模板匹配相结合的泡沫图像宏块跟踪方法来自动跟踪形变气泡以准确测量浮选泡沫流向刮板的速度。该方法通过自动扫描泡沫图像高亮点,根据气泡大小进行自适应泡沫子块选取；利用Fourier-Mellin变换获取跟踪泡沫子块在相邻帧中的形变系数和子块大致位移估计；然后,根据泡沫子块的形变系数进行几何反变换获取跟踪子块形变前形貌；再利用模板匹配在估计的大致位移局部区域进行细致的子块配准搜索,获取跟踪泡沫子块在相邻帧中的位移,实现形变泡沫子块速度特征的准确测量。该方法解决了泡沫速度特征因浮选气泡形变严重而难以准确提取的问题。(3)针对浮选泡沫在流动过程中气泡破碎率高、坍塌严重,常用的运动估计方法因无法实现对气泡破碎区域的匹配与跟踪,而难以获取破碎严重的浮选泡沫表面流动速度的问题,提出了一种基于泡沫图像灰度SIFT (Scale Invariant Feature Transform)与Kalman滤波相结合的泡沫速度特征提取方法跟踪各种泡沫运动子块。该方法首先通过Kalman滤波预测浮选泡沫子块在相邻帧中对应的大致位移,在预测位移局部区域进行SIFT特征点提取与泡沫子块配准,极大地减少了特征提取与匹配的盲目性,提高了特征匹配速度与准确性。该方法解决了泡沫图像中气泡大量破裂的情况下泡沫子块流速无法准确测量的问题,计算复杂度低,便于工业现场应用。(4)针对泡沫稳定度特征难以定量描述问题,提出基于数字图像处理的浮选泡沫表面形变系数与破碎率特征提取方法。该方法在泡沫速度特征提取的基础上,通过定义相应的稳定度评价准则,实现了泡沫形变系数、泡沫破碎率的量化描述,解决了泡沫图像稳定度特征提取与表征问题,为实现浮选泡沫状态的客观评价提供了数字化参量。(5)以中国铝业中州分公司铝土矿浮选过程为应用对象,在浮选现场设计并搭建了浮选泡沫图像采集硬件平台,开发了相应的矿物浮选泡沫图像视觉监测系统,实时提取了浮选泡沫图像动态特征,初步分析了浮选泡沫速度特征与浮选工况间的关系,实现了浮选生产过程实时测量与客观评价。系统的泡沫特征曲线能够为生产工人提供明确的工况信息,并给出具体的操作建议,避免了工人操作的盲目性,提高了浮选生产效率,为浮选过程优化控制奠定了基础。更多还原

【Abstract】 Mineral beneficiation is generally an indispensable industrial process in the utilization of the most natural mineral resources. Among the commonly used mineral dressing methods, froth flotation is the most important technology and almost all of the different ores can be separated by this method. Flotation process is a complex continuous physicochemical process which occurres in solid-liquid, solid-air and liquid-air three-phase interface. In the flotation process, froth layer is the key factor to production performance. Currently, the flotation operation is mainly adjusted through observing the upper surface of the froth by the experienced operators. The naked-eye observation based flotation process operation fashion is strongly subjective and inefficient. It is deficient in qualitative descriptions of flotation froth states and the objective evaluation of the production performance, which consequently results in frequent fluctuations of production indices and low recovery of raw materials with large consumption of reagent dosage. Since1990s, some developed countries have introduced the machine vision to the flotation process monitoring and control, which incurs increasing interest of the researching communities and plant operators. It is significant to research machine vision based flotation process monitoring, modeling and automatic control method and technique in theory and practical application to improve the efficiency of the usage of the mineral resource and realize the sustainable development of enterprises.It’s essential to extract the accurate visual characteristic parameters of the froth surface to realize the machine vision based automatic control of the flotation process. There are two kinds of visual characteristics of froth surface appearance. One is based on the froth characteristics of single frame image, called static parameters such as bubble size, bubble shape, surface froth color and froth texture characteristic and so on. The other is the characteristic based on image sequence known as dynamic parameters, generally including the bubble velocity, bubble stability and so on. Currently, researchers both at home and abroad mainly focus on extracting and analyzing the static parameters while little attention has been paid to dynamic parameter extraction. It’s worth noticing that the dynamic parameters are also indispensable to realize the objective evaluation of the flotation production statuses. The inherent relation between the flotation production conditions and both the static and dynamic characteristics of froth image are presented in this dissertation based on the analysis of the flotation mechanism. This work mainly focuses on researching the dynamic parameter extraction and characterization method of flotation froth based on digital image processing with the industrial application to a bauxite flotation process monitoring and control system. The main contributions are as follows.(1) A Bayesian denoising method integrated spatial-temporal image information is proposed based on multi-scale geometric analysis, aiming at solving the problem of inaccurate feature extraction of the froth images with serious noise contamination. Curvelet transformation is adopted in advance to express the geometric structures and surface texture of the surface froth image sparsely. Then, the marginal statistical distribution and the joint statistical distribution model of the Curvelet transform coefficients of single-frame image and multi-frame images is constructed after analyzing the statistical characteristic of the image coefficients in the transform domain. According to a specific cost measure, Bayesian inference is used to estimate the ideal uncontaminated coefficients in the transform domain. At last, the inverse Curvelet transformation is applied to get the denoised froth image. Compared to the other image denoising algorithms, this method can achieve higher peak signal noise ratio (PSNR) and keep the froth image details effectively while making noise reduction, which provides high-quality processing signals for the subsequent feature extraction of froth images.(2) An adaptive macroblock tracking method of froth image based on the combination of Fourier-Mellin transformation and gray template matching is proposed to measure the accurate velocity of the flotation froth flowing to the scrapper through automatically tracking the deformed bubbles in the flotation cells. Since the mineralized bubbles in the flotation cells subject to indispensable geometric distortion in scales and orientations, the froth flow velocity cannot be extracted accurately by the traditional object tracking methods, which cannot tracking the non-rigid froth bubbles with serious geometric distortion in the flotation process. This method aims to make a complement of the traditional object tracking methods to measure the froth velocity accurately. The sub-blocks are located and assigned adaptively in advance by scanning the highlights of the froth image automatically and estimating the bubble sizes effectively. The deformation coefficients and the approximate displacement of the sub-blocks in the adjacent image frames are computed by Fourier-Mellin transform, which are used to obtain the non-deformed sub-blocks by inverse geometric transformation. Then, grey template matching is adopted to search the positions of the non-deformed sub-blocks in the local neighbourhood around the first estimated displacement to measure the precise displacement of the sub-blocks in the adjacent froth sequences. This method solves the problem of the froth velocity characteristic cannot be extracted accurately with serious geometric distorted flotation bubbles.(3) A new froth velocity characteristic extraction method based on image gray SIFT(Scale Invariant Feature Transform)with Kalman filtering is proposed. Since the flotation bubbles collapse and burst seriously in the flotation process, the commonly used motion estimation methods cannot match and track the collapsed bubbles areas effectively, which result in inaccurate flow velocity measurement. Firstly, Kalman filter is used to predict the approximate displacement of the sub-blocks in the adjacent frames, then the SIFT features are extracted to registered the sub-block in a narrow neighbor area around the approximate displacement. This method greatly reduces the aimlessness of SIFT features extraction and feature points matching, which greatly improves the efficiency and the accuracy of feature point matching. It computes conveniently with low computational cost and successfully solves the problem of measurement of the froth flow velocity with serious bubble burst and collapse, which is convenient to apply to the real industrial flotation process. (4) A method to extract the deformation coefficient and bursting rate features of the flotation froth is presented based on digital image processing aiming at solving the problem of a qualitative description of the froth stability feature. Based on the froth flow velocity, the corresponding stability evaluation criteria are defined, which results in an effective qualitative description of the deformation coefficient and burst rate of flotation froth. This method is able to extract the froth stability in real time, which provides the digital parameters of the froth images to evaluate the flotation production statuses objectively.(5) A flotation froth image acquisition and processing flat is designed and established in the bauxite flotation plant of Zhongzhou branch of China Aluminum Company with the corresponding visual monitoring system. It extracts the visual characteristic of the froth surface appearance timely and analyses the preliminary relation between the flotation production conditions and the froth velocity characteristics, which results in online measurement of the froth surface and objective evaluation of the flotation production statuses. The froth feature curve can afford clear indication information of the production conditions and provide guidelines for flotation operation, which avoid the blindness of traditional naked-eye observation based flotation operating. The monitoring system improves the flotation efficiency greatly and lays a foundation for the optimal control of the flotation process.更多还原