摘要
为应对高光谱图像分类中的特征高维度问题,提出一种基于多分类器融合的高光谱图像分类方法.利用高光谱数据相邻波段的高相关性,通过自适应子空间分解产生多个特征子空间,进而训练生成子分类器;利用ReliefF-S算法,对各特征子空间进行评价并生成各子分类器的权重,最终通过加权表决融合实现分类决策.实验表明,所提方法可有效规避高维特征问题并提升分类性能.
A novel approach for hyperspectral image classification is proposed based on fusion of multiple classifiers to deal with the high dimension in applications of hyperspectral image classification.High correlation of neighboring bands of hyperspectral image data is used to generate feature subsets through adaptive subspace decomposition.A modified ReliefF algorithm（ReliefF-S） is proposed to evaluate feature subsets and to generate their corresponding weight values.Member classifiers are trained based on resulting feature subspaces and their weighted majority voting,and then the fusion of multiple classifiers is accomplished.Experimental results show that the proposed approach reduces the dimension of features effectively,and improves the classification performance.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2010年第8期20-24,共5页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(60605009)
国家重点基础研究发展规划资助项目(2007CB311006)
陕西省电子信息系统综合集成重点实验室资助项目(200910A)
关键词
高光谱图像
多分类器融合
自适应子空间分解
加权表决
hyperspectral image
multiple classifier fusion
adaptive subspace decomposition
weighted voting.
