基于支持向量机的高光谱影像分类研究

Research of hyperspectral image classification based on support vector machine

高光谱遥感技术,将反映目标辐射属性的光谱信息与反映目标空间几何关系的图像信息有机地结合在一起。高光谱影像丰富的光谱信息使其较全色遥感、多光谱遥感能够更好的进行地面目标的分类识别。本文综合利用支持向量机分类的若干关键技术,包括序列最小优化训练算法、多类支持向量机构造方法、核函数及其参数选择的交叉验证"网格搜索",给出了高光谱影像分类流程,进行了遥感数据试验分析。

Hyperspectral remote sensing technology organically combines the radiation information which relate to the targets＇ attribute, and the space information which relate to the targets＇ position and shape. The spectrum information, which the hyperspectral image enriches are better to carry on the ground target classification, compare with panchromatic remote sensing image and multispectral remote sensing image. This paper introduces the key technologies of hyperspectral image classification based on support vector machine, including sequential minimal optimization, kernel function＇s parameter and regularization constant selection cross-validation via parallel grid search, and multi-classes classification problem. Then the method is applied to the hyperspectral image classification test.