高光谱图像分类方法综述

在过去数十年中，高光谱图像的研究与应用已经完成了从无到有、从差到优的跨越式发展.在对其研究的众多方面中，高光谱图像分类已经成为了一个最热的研究主题.研究表明空间光谱联合的分类方法可以取得比仅依赖光谱信息的逐像素分类方法更好的分类效果.本文将对众多的空间光谱联合分类方法进行归类和分析.首先介绍高光谱图像中相邻像素间的两类空间依赖性关系，因而可将现有的空谱联合分类方法分为依赖固定邻域和自适应邻域两类；此外，还可以依据是否同时利用两类依赖关系将现有方法进一步分为单依赖和双依赖两类.另外，还可以依据空谱信息融合的不同阶段将现有的分类方法划分为预处理方法、一体化方法及后处理方法三类.最后展示几种具有代表性的空间光谱联合分类方法在真实高光谱数据集上的分类结果.

Overview of hyperspectral image classification methods

Over the past few decades,the research into and application of hyperspectral images has made significant progress,with interest levels progressing from very little to intense.Among the many aspects of this research,hyperspectral image classification has become one of the most-studied topics,with experiments showing that the spatial spectral joint classification method can achieve better classification results than the pixel-by-pixel classification method that relies on spectral information alone.Here,we classified and analyzed a number of spatial spectral joint classification methods.First,we introduced two kinds of spatial dependence relations between adjacent pixels in hyperspectral images;using these,we were able to classify existing spatial-spectral classification methods into fixed-dependent neighborhood-based and adaptive neighborhood-based types.In addition,we were able to further divide existing methods into two types,single-dependency and double-dependency,based on whether or not we used two types of dependencies at the same time.We were also able to divide existing classification methods into three types:preprocessing,integration,and post-processing,according to the different fusion stages of the spatial and spectral information.Finally,we showed the classification results achieved from the application of several representative spatial-spectral classification methods to real hyperspectral datasets.