高光谱遥感城市植被识别方法研究

Research on Identification Method of Urban Vegetation Based on Hyperspectral Imagery

传统的植被状况调查方式费时、费力,并且更新困难,而高光谱遥感数据图谱合一,能够更精细、准确地进行遥感地物识别和分类,因此采用Hyperion高光谱数据来研究地物混合严重并且呈零星碎片状的城市植被。利用混合像元分解思想改进Gram-Schmidt融合算法,将Hyperion高光谱和ALI全色波段进行融合,提高光谱数据的空间分辨率,来解决城市植被像元混合严重和分布过于零散破碎难题,进而提高植被识别精度。为了避免高光谱植被识别陷入维数灾难,采用主成分分析对融合后的高光谱数据进行数据降维。最后,在地面光谱成像仪获取的纯净像元光谱信息辅助下,选取训练样本进行最小距离分类,完成植被类型识别,总体精度达到84.9%。

Based on EO-1 hyperion imagery, we merged hyperion 30 m and ALI 10 m spatial resolution data by the improved Gram-Schmidt fusion method. To avoid falling into the curse of dimensionality, we used principal component analysis(PCA) to reduce dimensionalities for the merged hyperspectral data. And then, we selected the training sample to do the minimum distance classification. The overall classification accuracy was 84.9%.