基于核主成分支持向量机的火成岩QAPF分类——以青海格尔木地区为例

利用核主成分(KPCA)较强的非线性特征提取能力对Hyperion高光谱数据进行降维及光谱特征提取,将特征信息作为支持向量机(SVM)建模样本的观测数据,建立KPCA-SVM回归模型,利用该模型进行研究区岩石氧化物百分含量反演。同时,依据国际地质科学联合会提出的QAPF火成岩分类方案对区内火成岩进行了岩性划分。研究结果表明:KPCA降维后的高光谱数据反演氧化物含量的效果良好;而基于QAPF模型的火成岩划分结果也十分理想,分类结果对已有地质图进行了有效的补充。KPCA-SVM理论模型为利用高光谱遥感数据进行岩性分类提供了一种快速可行的方法。

The Classification of Igneous Rocks with QAPF Based on Kernel Principal Component SVM: A Case Study of Golmud Area in Qinghai Province

In this paper, the non-linear feature extraction capability of KPCA was used to reduce dimensionality and extract spectral features of Hyperion hyperspectral data. The extracted feature information was employed as the sample data and the KPCA-SVM regression model was established. According to this model, the percentage of rock oxide in the study area was retrieved. The QAPF igneous rock classification scheme proposed by IUGS was utilized to classify the igneous rocks. The oxide content retrieved from the hyperspectral data became more reasonable by using KCPA for dimension reduction. In accordance with the QAPF model, the igneous rock classification results were most satisfactory, and the classification results became an effective complement of the existing geological map. It is proved that the KPCA-SVM method is a fast and feasible means for lithologic classification based on hyperspectral remote sensing data.