论文部分内容阅读
为挖掘高光谱遥感数据内在的非线性结构特性,采用全局化流形学习算法等距特征映射(ISOMAP)对高光谱遥感数据进行非线性降维,并取得了优于常用的最小噪声分离(MNF)变换方法的结果,具有更好的数据压缩性能。将光谱角相似性度量方法用于ISOMAP算法,取得良好的降维效果。通过把ISOMAP降维算法和k-最邻近分类器相结合对降维后子空间特征进行分类,实验表明:ISOMAP利用较少的特征维数获得比MNF更高的分类精度,并达到较高稳定的分类精度,尤其对难以区分、光谱相似的两类别问题,ISOMAP的特征维数能够有效的提高两类别的可分性。
In order to exploit the inherent nonlinear structural characteristics of hyperspectral remote sensing data, ISOMAP was used to nonlinearly reduce the hyperspectral remote sensing data and to achieve better than the commonly used minimum noise separation (MNF ) Transform method results, with better data compression performance. The spectral angle similarity measure method is used in ISOMAP algorithm to obtain a good dimension reduction effect. By combining the ISOMAP dimensionality reduction algorithm and the k-nearest neighbor classifier to classify the subspace features after dimensionality reduction, experiments show that ISOMAP achieves higher classification accuracy with fewer feature dimensions and higher stability than MNF The classification accuracy of ISOMAP can effectively improve the separability of two categories, especially for two types of problems that are indistinguishable and spectrally similar.