一种新的高光谱遥感图像纹理特征提取方法研究  

A Novel Texture Feature Extraction of Hyperspectral Remote Sensing Image

作　　者：冯静[1,2] 舒宁[2] 

机构地区：[1]武汉理工大学计算机科学与技术学院,武汉430070 [2]武汉大学遥感信息工程学院,武汉430079

出　　处：《武汉理工大学学报》2009年第3期10-13,17,共5页Journal of Wuhan University of Technology

基　　金：国家自然科学基金(40371079)

摘　　要：提出了一种新的基于自相关的高光谱遥感图像纹理特征提取算法,该算法通过引入核函数技术,将单波段纹理窗口的空间自相关函数,扩展到多波段遥感图像的纹理描述。然后对特征矢量进行无监督C均值聚类实验和有监督RBF神经网络分类实验,在分类实验中确定了最佳窗口尺寸。实验结果表明,该文提出的自相关特征可以有效地描述高光谱遥感图像的纹理。A novel texture feature extraction method of hyperspectral remote sensing image was proposed based on autocorrdation function. The concept of space autocorrelation function was expanded form texture window of single-band image to that of multi-band remote sensing image by introducing kernel function technique. The optimal size of texture window was determined during classification experiments. The results of C-means clustering and RBF neural networks classification experiments show that, the proposed texture feature based on autocorrelation can effectively describe the texture of hyperspectral remote sensing image.

关 键 词：高光谱遥感 纹理 特征提取 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]