一种用于多光谱纹理分类的ICA多尺度算子

An ICA Multi-Scale Operator for Multispectral Texture Classification

针对像元数据为多维的多光谱遥感图像,提出了一种可用于多光谱纹理分类的ICA多尺度算子.首先定义用于描述纹理不同尺度特性的结构单元,然后联合所有波段下的结构单元建立起测量多光谱纹理特性的高维矢量,最后对这些矢量进行ICA分析,将得到的独立成分作为多光谱纹理特征用于分类.通过对Brodatz灰度纹理、VisTex彩色纹理以及实际的盐渍土多光谱图像的分类,试验验证了该算子的有效性.

In light with the multispectal remote sensing images with multidimensional pixel data,an ICA multi-scale operator is presented for multispectral texture classification.Firstly,the tactic patterns with different sizes are defined to describe the structural properties at multi scales for a certain texture.Then the patterns from all the bands are designed to measure the vectors with high dimensions.Finally,the independent component analysis（ICA） is operated on the vectors to obtain the independent components for texture classification.The experiments on Brodatz textures,VisTex color textures and real multispectral images of saline soils demonstrate the validity of the proposed operator.