基于图像特征提取的浮选关键参数智能预测算法  

Intelligent prediction algorithm for floatation key parameters based on image features extraction

作　　者：周开军[1] 阳春华[1] 牟学民[1] 桂卫华[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《控制与决策》2009年第9期1300-1305,共6页Control and Decision

基　　金：国家自然科学基金重点项目(60634020);国家自然科学基金项目(60874069;60804037)

摘　　要：针对矿物浮选过程中回收率参数难以在线检测的问题,提出了一种智能预测算法.首先采用相对红色分量提取泡沫颜色特征,采用改进面积重构变换与分水岭方法分割泡沫图像并提取尺寸特征;然后在此基础上,通过斯密特正交化对最小二乘支持向量机(LSSVM)核矩阵进行简约,利用核偏最小二乘方法进行回归计算,得到具有稀疏性的LSSVM预测模型.实验结果表明,该预测算法能有效地对矿物回收率进行预测.For the problem that the mineral recovery is hard to online measure in mineral floatation process, an intelligent prediction algorithm is proposed. The froth color feature is extracted by means of relative redness component. Meanwhile, the froth image is segmented by using an improved area-reconstruction and watershed transform, and the bubble size feature is extracted. On the basis, the kernel matrix of least mean support vector machine （LSSVM） is reduced by Schmidt orthogonalization, and the kernel partial least squares regression ealculation is conducted to obtain the sparse LSSVM. The experimental results show that the prediction algorithm can predict the mineral recovery effectively.

关 键 词：矿物浮选 泡沫图像 特征提取 预测模型 最小二乘支持向量机 

分 类 号：TP273.5[自动化与计算机技术—检测技术与自动化装置]