基于多层形状特征提取与融合的城市高光谱影像解译

以前的研究往往从像素光谱的角度来解译高光谱影像,忽略了像素间的空间上下文关系。本文提出一种基于像素和对象层形状特征提取与融合的方法,把多层形状特征和光谱信息用支持向量机(SVM)输出函数方法进行融合,用于提取城市高光谱影像的形状特性,利用影像的空间关系。实验用HydICE-DC航空高光谱数据对提出的方法进行了验证,结果表明:像素级形状指数能够提供比对象级形状指数更优的结果,但像素—对象级形状特征的融合,能够给出更高的精度。

Multilevel shape feature extraction and fusion for classification of urban hyperspectral imagery

Traditional methods for hyperspectral image interpretation only focused on spectral information, without considering the spatial context. In this paper, we proposed a multilevel shape feature fusion approach combing both pixel- and object-level shape features. The extracted multilevel shape information was integrated with the spectral features using the output of SVM-based discriminant function. Experiments were conducted on the HydICE DC Mall hyperspectral dataset, and results showed that the pixel-level shape index outperformed the object-based index, while the multilevel shape features gave the highest accuracies.