结合目标分割的高光谱城市地物分类

Urban features classification based on objects segmentation and hyperspectral characteristics

结合高光谱影像地物光谱特征与高空间分辨率影像分割获得的目标对象进行地物分类。首先,对Hyperion影像进行坏线和Smile效应去除,经过FLAASH大气校正后,得到研究所用的155个波段;其次,利用地物光谱曲线的特征点确定适合地物识别的光谱分辨率,进行Hyperion影像降维,生成降维后所需的21个宽波段;然后,对IKONOS影像采用小波融合,利用多分辨率分割技术生成高空间分辨率影像目标对象;最后,基于层次分析法对分割后生成的目标对象进行分类,采用模糊隶属函数利用植被红边效应、水体在近红外波段吸收特征进行第1层次分类,再取距离值最大的前10个Hyperion影像波段作为标准最邻近分类的特征波段,完成第2层次分类。分类结果表明,研究区共分出9种地物类型,分类效果明显优于最大似然法分类与光谱角填图法。

Urban features classification is based on hyperspectral characteristics and high-resolution image segmentation objects. After the removal of bad lines and Smile effect,FLAASH atmospheric correction and 155 Hyperion bands were used in this study. Spectrum feature was used to determine objects recognition suitable spectral resolution,and after Hyperion dimensional reduction,21 wide-bands were generated. Utility wavelet fusion was performed,and IKONOS high-resolution objects were generated by multi-resolution segmentation. On the basis of hierarchical analysis classification method for segmentation objects,fuzzy membership function of the vegetation red edge effect and the water absorption characteristics in the near infrared were used to complete first level classification. The larger distance of 10 Hyperion bands was used as feature bands,and the second level classification was completed by standard nearest neighbor classification. 9 types of urban features were separated.The classification results are better than the maximum likelihood classification and spectral angle mapper.