摘要
高光谱数据具有波段多、光谱范围窄、数据量大等特点,但巨大的数据量给数据处理带来了困难,同时它的高维也容易导致Hughes现象的产生。因此,对其进行降维处理显得非常必要。以Hyperion数据为研究对象,分别利用特征选择和特征提取的方法达到数据降维的目的。结果表明:(1)波段选择之前进行子空间划分,可剔除相关性大的波段,并能减小数据计算量,避免信息的丢失,从而实现高维遥感数据优化处理和高效利用的目的。(2)MNF变换后高光谱数据的有效端元数可为图像的进一步分析和应用提供参考。
Hyperspectral data have more bands, narrow spectral range, large volumes of data, etc. , but a huge amount of data make data processing very difficult, while its high-dimensional phenomenon can easily lead to the generation of Hughes. Therefore, dimensionality reduction process is very necessary. By taking Hyperion data as the research object, using feature selection and feature extraction methods, the purpose of data reduction was achieved. The results show that dividing space before sub-band selection can eliminate the band with bigger correlation, and can reduce the amount of data calculation, to avoid loss of information, thus realizing optimal high-dimensional remote sensing data processing and efficient utilization purposes.
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2011年第11期34-38,共5页
Journal of Central South University of Forestry & Technology
基金
国家自然科学基金项目(30871962)
高等学校博士学科点专项科研基金(200805380001)
国家林业局林业公益项目专题(201104028)
关键词
高光谱数据
降维
特征提取
HYPERION
Hyperspectral data
dimensionality reduction
feature extraction
Hyperion data
