基于地标空间信息的高光谱遥感图像分类

摘要

A hyperspectral remote sensing image has the characteristics of spectral information and spatial information,it can effectively improve the classification resuhs by making full use of spectral and spatial information.In the hyperspectral image,the difference of the neighbor region spatial information of the same kind of pixel points will affect the classification effect.This paper proposes a new method of spatial information utilization,which is written as landmark spatial information (Lspatial).This paper combines the spectral information with the landmark spatial information,then extracts the feature by deep belief network.Experiments show that landmark spatial information can effectively improve the classification performance of deep belief network (DBN) and contrast method.%高光谱遥感图像同时具有光谱与空间特征信息,充分利用空间特征能够有效提高分类结果.高光谱图像上同类像素点的邻域空间信息差异较大会影响分类效果,针对这个问题提出了一种新的空间信息利用方法,记作地标空间信息.将光谱信息与地标空间信息结合使用,通过深度信念网络提取特征.实验表明地标空间信息能够有效提高深度信念网络以及对比方法的分类结果.