混合核特征加权SVM遥感湿地空间分类  

Feature weighted-mixed kernel SVM for remote sensing wetland classification

作　　者：刘冰[1] 吴超 林怡[3] 

机构地区：[1]河南省航空物探遥感中心,河南郑州450053 [2]河南省电力勘测设计院,河南郑州450007 [3]同济大学测绘与地理信息学院,上海200092

出　　处：《测绘工程》2016年第7期13-17,共5页Engineering of Surveying and Mapping

摘　　要：针对湿地空间信息的复杂性和SVM的分类性能,设计一种基于混合核函数的特征加权SVM分类模型,综合利用多种特征信息,避免被弱相关特征所支配,从而提供更佳的映射性能和泛化能力。实验结果表明,该分类模型兼具良好的外推和内推能力,能够有效地融合不同信息源特征,得到更完整和准确的分类结果,在总体精度、Kappa系数等多项指标上都表现出更高的水平。In view of complexity of wetland information extraction and classification performance of the SVM classifier,this paper proposes a feature weighted-mixed kernel SVM classifier,which can utilize many kinds of feature information and avoid being dominated by weak correlation characteristics.Then better mapping performance and generalization ability can be realized.The experimental results show the classification model with good ability of extrapolation and push can effectively integrate different information source characteristics,which also achieve more complete and accurate classification results,with a higher classification accuracy on the index such as overall accuracy and Kappa coefficient.

关 键 词：混合核 特征加权 支持向量机 湿地 分类 

分 类 号：P237[天文地球—摄影测量与遥感]