免疫克隆优化聚类技术  

Immune clonal optimization clustering technique

作　　者：马文萍[1] 尚荣华[1] 焦李成[1] 

机构地区：[1]西安电子科技大学智能信息处理研究所,陕西西安710071

出　　处：《西安电子科技大学学报》2007年第6期911-915,921,共6页Journal of Xidian University

基　　金：国家"863"项目资助(2006AA01Z107);国家"973"项目资助(2006CB705700);国家自然科学基金资助(60703108)

摘　　要：将人工免疫系统中的克隆选择优化应用于无监督分类与识别问题,提出了一种新的免疫克隆聚类算法.该算法利用克隆算子能够同时在同一父代抗体周围的多个方向进行全局或局部搜索,促使种群中抗体快速进化,从而在特征空间内快速获得聚类问题的全局最优聚类中心,有效克服了经典聚类算法易陷入局部极值的缺点,并从理论上证明了该算法具有全局收敛性.对7个人工数据集的聚类实验和两幅纹理图像的分割实验表明:新算法比常用的K均值算法的平均分类精度高20.9%,比另一种基于遗传算法的聚类方法的平均分类精度高20.3%.A new immune clonal clustering algorithm based on the clonal selection optimization of the artifical immune system is proposed for solving unsupervised classification and recognition problems. The new algorithm can carry out the global search and the local search in many directions around the same antibody simultaneously, and make the most of antibodies in population adequately, in order to search for global optimal cluster centers in the feature space quickly. It avoids the local optimum of the classical clustering algorithm. Theoretical analysis shows that the new algorithm can converge to the global optimum. Experiments on seven benchmark clustering problems of artificial data sets and two texture images segmentation problems show that the average correct rate of the new algorithm is higher than that of the K-means algorithm by 20.9%, and is higher than that of a genetic algorithm based clustering method by 20.3 %.

关 键 词：免疫克隆 聚类 K均值算法 纹理图像分割 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]