端元可变非线性混合像元分解模型

Nonlinear Spectral Unmixing for Optimizing Per-pixel Endmember Sets

遥感影像中混合像元普遍存在。端元固定的情况下对混合像元进行分解,很难高精度地识别影像地物。本文基于支持向量机,提出了端元可变的非线性混合像元分解模型。首先,通过构建多个支持向量机获取每个像元的优化端元集,在优化端元集的基础上运用支持向量机与两两配对方法相结合的算法获取像元组分。试验结果表明,本文提出的方法效果优于传统的多端元光谱分解法。

For a given pixel, fractional abundances predicted by spectral mixture analysis （SMA） are most accurate when only the endmembers that comprise it are used. This paper presents a support vector machines （SVM） method to achieve land use/land cover fractions of remote sensing image using two steps, ①defining the optimal per-pixel endmember set, which removes endmembers based on negative fractional abundances generated by SVM method; ②using SVM extended with pairwise coupling （PWC） to output probabilities as the abundance of landscape fractions. The performances of the proposed method were evaluated with the multiple endmember spectral mixture analysis （MESMA） method, which has been widely applied to map land cover for the goodness of the model fitting. The results obtained in this study were validated by real fractions generated from SPOT high resolution geometric （HRG） image. The best classification results were obtained by the proposed method indicated by the lower total mean absolute error, the higher overall accuracy, and the higher kappa. From this study, the proposed method is proved to be effective in obtaining abundance fractions that are physically realistic （sum close to one and nonnegative）, and providing valuable application in selectincl endmembers that occur within a pixel.