【作者】 周开军；

【导师】 阳春华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2010， 博士

【摘要】 矿物浮选是在特定工艺条件下,在矿浆中加入浮选药剂,并充入空气、然后搅拌产生大量气泡,最后通过回收含矿泡沫来提高原矿品位,以此满足还原冶炼要求的一种选矿方法。浮选泡沫具有数量多、粘连、混杂、形状不规则等特点,泡沫尺寸与形状特征难以定量描述。目前,我国有色金属矿山的浮选过程主要通过人工观察泡沫状态来调整操作,影响浮选过程的优化运行和有用金属回收率。因此,研究泡沫图像形态特征提取方法,并用于指导浮选工业过程操作,对提高矿产资源的利用效率、实现企业的可持续发展,具有非常重要的现实意义。论文在分析研究浮选机理的基础上,依据气泡与矿粒的粘附原理,研究了气泡尺寸、形状特征与浮选工况之间的关系,在此基础上,提出了基于机器视觉的浮选泡沫图像形态特征提取方法,并成功应用于矿物浮选过程泡沫图像监控系统中。论文主要研究工作及创新性成果体现在以下几个方面：(1)针对浮选泡沫图像分割过程中人工选择结构元素存在的问题,详细分析了气泡亮点的统计信息特点,提出了基于几何模式谱的结构元素选择方法。以改进的模糊C均值算法对图像进行聚类,为结构元素的选取提供先验知识,以形态学面积重构开闭运算方法完成图像的除噪。通过定义二值图像的几何模式谱,证明了任意形状属性算子的非递增特性,为将该特性扩展至灰度图像,运用最大树原理对图像进行枝剪,使得算法能够高效地计算尺寸与形状模式谱值。该方法充分利用气泡表面的亮点信息,有效地为泡沫图像分割过程提供结构元素,确保了图像分割准确性,在很大程度上减少了对人工经验的依赖。(2)针对形态学处理过程中的一类结构元素寻优问题,提出了基于粒子群算法的结构元素优化方法,在定义几何模式谱目标函数的基础上,通过连续改变结构元素的尺寸与形状,计算几何模式谱的全局最小值得到最优结构元素。该方法有效避免了局部最优问题,同时保证了算法的实时性。(3)针对气泡混杂粘连的特点,提出了基于分级分水岭算法的泡沫图像自适应分割方法。在选取最优结构元素的基础上,以经典分水岭算法完成对泡沫图像的粗分割。提出了基于模糊纹理谱的泡沫图像识别方法,对模糊纹理谱方法加以改进,提出了非线性模糊纹理谱特征提取方法,以支持向量机完成粗分割区域的识别,根据识别结果采取不同处理策略,其中对欠分割的小泡区域进行细分割,运用图像区域合并方法,对过分割的区域进行区域合并,通过定义图像分割评估机制,完成分割结果的评估。这种图像分割、特征提取与识别协同处理的方法极大地提高了算法的鲁棒性,避免了气泡混杂不均对分割结果的影响,有效地减少了欠分割和过分割区域。(4)针对泡沫尺寸与形状特征难以定量描述问题,提出了泡沫图像的尺寸与形状特征提取方法。在图像分割的基础上,对气泡分割区域像素进行标定,引入样本统计分布的概念,提取了气泡平均尺寸、方差、偏斜度及陡峭度等统计特征。从定性与定量的角度描述了泡沫形状特征,针对定性分析的局限性,提出了基于形态学签名变换的泡沫形状特征提取方法,将气泡复杂形状的特征提取问题转化为从多个签名形状中抽取简单形状特征的问题,在很大程度上简化了复杂形状的描述。该方法通过统计气泡群的特征信息,有效地量化了气泡形态特征,具有良好的实用性。(5)以实际矿物浮选过程为研究对象,设计了泡沫图像获取硬件平台,开发了矿物浮选泡沫图像监控系统,在此基础上,分析泡沫形态特征与矿物回收率的相关性,建立了基于最小二乘支持向量机的矿物回收率预测模型,为提高算法实时性,对模型进行稀疏化处理,实现了浮选生产过程实时监控。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 flotation is a kind of mineral processing methods.During the process, flotation reagents are added to the pulp and air is filled in the slot. Then they are mixed to bring a great deal of air bubbles.Finally ore grade is improved by retrieving froth containing minerals to meet smelting requirements.Usually, froth is large quantity, adherence, hybrid and irregular shape. Flotation process of nonferrous metal mines in China is usually operated by experienced workers through observing froth surface. As a result, it is hard to work in optimized running state and mineral recovery ratio is low. Consequently, researching the morphological characteristics extract method for froth images and applying them into practice are great significance for optimizing flotation process, maximum using resource, reducing resource consume and maintaining enterprise sustainable development.According the adhesion principle of air bubbles and mineral particles, relationship between froth morphological characteristics and flotation operating condition is researched in this paper based on flotation mechanism. Then froth image characteristics extraction scheme based on machine vision is proposed and applied in online monitoring system of mineral flotation process successfully. Main research work and innovative achievements are as follows:(1)Considering the limitation that manual selecting structural element in image segmentation, adaptive selection method for structural element was proposed in this paper based on geometric pattern spectra. Improved fuzzy C-means for image clustering and area reconstruction by open and close operation for de-noise were used to provide prior knowledge for image segmentation.To solve structure elements automatically, morphological geometric pattern spectrum was introduced for binary image. The non-increasing is proved on condition that arbitrary shape operator. In order to extend the characteristic to gray image, a max tree principle is utilized to prune the image. Thus, the proposed algorithm can compute the size and shape pattern spectra value efficiently. This method utilizes the bubble lighting spot information adequately, and provides image segmentation with the structural element. Consequently, it ensured the accuracy of image segmentation, and decreased manual experience to a great extent.(2) Aiming at the structural element optimization problem in morphological processing, a novel structural element optimization method based on particle swarm optimization(PSO) algorithm. On the basis of geometrical pattern spectra objection function, the global mixmization value is obtained by serially changing the size and shape of structural element, which not only avoid local optimum problem, but also ensure the real time property of the proposed algorithm.(3)Considering the problem of the asymmetrical size and irregular shape of bubbles in process of image segmentation, froth image adaptive segmentation based on hierarchical watershed algorithm was presented. On the basis of selection of optimal structure element, coarse segmentation was done by using watershed algorithm. Then texture features of the segmentation regions were extracted by fuzzy texture spectrum. Besides, fine-grained segmentation for under-segmentation vesicle regions was preceded through support vector machines for region recognition. At the same time, image region merging mechanism was introduced to merge the big bubble regions which were over-segmentation. Finally the result of segmentation was evaluated. The cooperation process mechanism combines image segmentation, feature extraction and image recognition. Thus, it enhanced the robustness of algorithm, and avoided the effection of all kinds of segmentation caused by hybrid bubble, and decreased under-segmentation and over-segmentation region.(4) A method for morphological characteristics extraction of froth image was presented. On the basis of image segmentation, pixels in bubble segmentation region were calibrated. At the meantime, concept of sample distribution statistics was introduced. Moreover, some statistical characteristics, such as the average size of bubbles, variance, skewness and abruptness were extracted. The bubble shapes were described from qualitative and quantitative view separately in order to extract bubble shape features. Morphological signature transformation and multi-structural elements were also used for extract bubble morphological characteristics.The feature extraction problem of complex shape is transform into extraction simple shape feature from several signature shapes, which simplify the shaple description method. The experimental results show that the morphological characteristics extracted in this paper have strong practicability.(5)For an acctural mineral flotation process, a hardware platform was designed to obtain froth images and mineral flotation froth video monitoring system was developed. On the basis of this, correlation analysis of froth morphological characteristics and production index was proposed. Sensitivity analysis of the extracted features was researched firstly. Then correlation analysis of forth morphological characteristics and rate of ore recovery was studied. As a result, prediction model of process indexes was established by using least square support vector machines.For the sake of real-timing of algorithm, the model is proceeded with sparse. Because of the using of this method for industrial flotation, real time monitoring of flotation process was realized. The system can give operating mode information and operation suggestion to the worker. Furthermore, labor productivity is increased and operation aimlessness is avoided, which provided the foundation for optimal control of flotation process.更多还原