基于泡沫图像特征加权SVM的浮选工况识别  

Froth image feature weighted SVM based working condition recognition for flotation process

作　　者：任会峰[1] 阳春华[1] 周璇[1] 桂卫华[1] 鄢锋[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《浙江大学学报（工学版）》2011年第12期2115-2119,共5页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61071176);国家"863"高技术研究发展计划资助项目(2009AA04Z124);国家杰出青年科学基金资助项目(61025015)

摘　　要：针对浮选泡沫视觉特征的多样性和重要度差异以及浮选工况样本数分布不平衡等问题,提出一种基于在线泡沫视觉表观特征加权支持向量机的浮选工况识别方法.通过色彩空间变换,在CIE-Lab空间计算泡沫颜色,采用多方向融合的空间灰度共生矩阵提取泡沫纹理特征,以视觉特征的信息增益评价该特征的重要度,再利用不同工况的样本数加权策略消除样本数不平衡的影响,采用支持向量机方法实现了浮选工况的自动识别.工业运行数据测试结果表明:该方法能够在线识别浮选工况,自动识别准确率达98%,比人工识别率高6%,比传统灰度共生矩阵方法高2%.Considering the diversity and different importance of visual features and the imbalance of sample distribution, a working condition recognition method for froth flotation process was presented. Firstly, froth color was calculated in CIE-Lab according to color space conversion, and texture were extracted based on four-direction fused space gray co-occurrence matrix. Then, information entropy was introduced to evaluate the feature＇s importance via information gain. Finally, the working condition recognition model was established based on support vector machines using sample weighted strategy to eliminate the imbalance of sample distribution. The experimental results of industrial operation data show that the proposed method can accomplish on-line monitoring of the flotation working condition automatically with identification accuracy of 98%, which is 6% higher than the manual recognition and 2% higher than the traditional Gray level co-occurrence matrix （GLCM） method.

关 键 词：浮选 工况识别 泡沫图像 加权支持向量机 空间融合灰度共生矩阵 

分 类 号：TP274.3[自动化与计算机技术—检测技术与自动化装置]