论文部分内容阅读
高光谱遥感是现在遥感领域的一个热门领域,已经引起了社会各界越来越多的关注。高光谱图像可以获取远比多光谱图像更丰富的信息,但它的处理难度也随着波段数的增加而提高了,而且这些波段并不一定都是我们所需要的,这样必然要对现有的高光谱图像进行降维处理。波段选择作为高光谱图像降维处理的一个重要方法一直发挥了很大的作用。通过实验比较了三种不同的波段选择方法,第一种是直接按照OIF指数高低进行提取,但往往很难去除相关性,第二种是分段OIF指数方法,可以有效去除相关性,但需要事先将所有波段分成数个波段子集,第三种方法通过迭代由粗到细综合考虑信息量和相关性两个因素,只需最开始迭代的概略初值,可以简便快速地获取最优波段组合。
Hyperspectral remote sensing is now a hot area in the field of remote sensing, which has aroused more and more attention from all walks of life. Hyperspectral images can obtain far richer information than multispectral images, but its processing difficulty increases with the number of bands, and these bands are not always what we need, so it is inevitable that existing Hyperspectral image for dimensionality reduction. Band selection has been instrumental in reducing the dimensionality of hyperspectral images. Three different band selection methods were compared by experiments. The first method was to extract directly according to the OIF index, but it was often difficult to remove the correlation. The second method was the segmented OIF index method, which could effectively remove the correlation but needed All the bands are divided into several waveband subsets in advance, and the third method integrates the information volume and the correlation from coarse to fine iteratively. The initial initial value of the initial iteration can easily and quickly obtain the optimal waveband combination.