多光谱遥感影像的SAM-SID混合分类技术研究

Classification of Multispectral Images Based on SAM-SID Mixed Measure

以SPOT 5多光谱影像为数据源,通过与SAM、SID以及常规的最大似然法(ML)和最小距离法(MD)的对比,研究了基于SAM-SID混合法的土地覆盖多光谱遥感分类技术。研究结果显示,相比于SAM和SID,SID(TAN)和SID(SIN)两个SAM-SID混合参量对多光谱影像上地物识别的能力更强,尤以SID(SIN)的识别能力最强;基于SID(SIN)的多光谱遥感分类验证精度达78.94%,不但明显高于SAM和SID法,而且也高于常规的MD和ML监督分类方法。这说明SAM-SID混合分类方法不但适用于高光谱遥感分类,同时在多光谱遥感分类中也有很强的适用性。

SAM-SID mixed measue is an improved remote sensing classification method,which is applied by matching spectrum curve based on Spectral Angle Mapper（SAM） and Spectral Information Divergence（SID）,and it has received excellent effect on hyperspectral image classification.In order to evaluate its applicability on multispectral images classification,and taking SPOT 5 images as an example,the classification method of multispectal images based on SAM-SID mixed measure-SID（SIN） and SID（TAN）,was studied by comparison with SAM,SID,Maximum Likelihood（ML） and Minimum Distance（MD）.The results show that the abilities of SID（TAN）and SID（SIN） for recognizing surface features in multispectal images are stronger than that of SAM and SID,in which SID（SIN） is more stronger.The classification accuracy of images based on SID（SIN） reached 78.94%,higher than that of the SAM and SID,also significantly higher than conventional ML and MD methods.This reflects that SAM-SID Mixed Measure is not only suitable for hyperspectral image classification,but also strongly applicable in multispectral image classification.