一种基于模糊混合像元分解的高光谱影像分类方法

A Classification Algorithm for Hyperspectral Images Based on Fuzzy Mixed Pixel Decomposition

高光谱遥感影像较低的空间分辨率使得混合像元大量存在于影像中,不仅影响了基于高光谱影像的地物要素识别能力,而且还降低了高光谱影像的分类精度。本文提出了一种基于模糊混合像元分解的高光谱影像分类方法。该方法主要利用约束能量最小化法设计的FIR线性滤波器,使得影像通过滤波器后输出与每类地物类别相关的"丰度图",其维数等于类别数;最后利用类中心匹配分类法实现高光谱影像的分类。实验结果表明,提出的分类方法与直接利用类中心匹配分类法相比,提高了影像的分类精度。

Low spatial resolution of the hyperspectral images leads to the existing of mixed-pixels in hyperspectral images,which not only affects the recognition capability of the objects based on hyperspectral images,but also reduces the classification accuracy.A classification method of hyperspectral images based on the fuzzy mixed-pixel unmixing was proposed in this paper.The method has been implemented mainly based on the constrained energy minimization（CEM）,which is a Finite Impulse Response（FIR） line filter.When the images and endmember spectrums are input in to the filter,the output is the ＂abundance image＂ that are related to each class,and its dimension is equal to the numbers of classes.Finally,the classification for hyperspectral images is taken by using the class-center matching strategy.The results of the experiments on two groups of real hyperspectral images showed that the classification accuracy has been improved fairly by the proposed method comparing with the directly class-center matching classification.