高光谱影像特征子集选择方法  

Research on feature selection method for hyperspectral imagery using PSO and SVM

作　　者：赵星涛 白瑞杰 王强 任长山 

机构地区：[1]北京数维翔图高新技术有限公司,北京100073

出　　处：《测绘科学》2015年第8期22-27,共6页Science of Surveying and Mapping

摘　　要：针对高维遥感数据的降维困难问题,该文提出并构建了一种融合粒子群优化算法全局寻优能力和支持向量机优秀分类性能的高光谱遥感影像特征子集选择与分类方法。通过引入混沌优化搜索技术改进融合粒子群优化算法的全局寻优能力;提出并采用一种基于粒度的网格搜索策略对支持向量机模型参数进行优化;利用二进制融合粒子群优化算法进行特征选择;然后,支持向量机采用该特征子集所对应的训练数据集进行模型参数优化和分类。实验结果表明该方法能有效地提取出用于分类的最佳波段,具有较高的分类精度。为高光谱遥感影像的特征选择与分类探索出了一种可行的方法。Aiming at the problem of dimensionality reduction of hyperspectral remote sensing data, in this paper, a novel feature selection and classification method for hyperspectral image by combining the global optimization ability of particle swarm optimization （PSO）algorithm and superior classification per- formance of support vector machine （SVM）was proposed. Global optimal search performance of PSO was improved by using chaotic optimal search technique. Granularity based grid search strategy was used to op- timize the SVM model parameters. Feature selection was carried out using binary PSO （BPSO）. Parameters optimization and classification of SVM were addressed using the training date corresponding to the feature subset. Experimental results indicated this hybrid method had higher classification accuracy and could ef- fectively extract optimal bands. A feasible approach was provided for feature selection and classification of hyperspectral image data.

关 键 词：高光谱遥感 粒子群优化 支持向量机 特征选择 

分 类 号：P231[天文地球—摄影测量与遥感]