SPOT5影像纹理特征提取与土地利用信息识别方法  

Extraction of Texture Feature and Identification Method of Landuse Information from SPOT 5 Image

作　　者：李金莲[1] 刘晓玫[1] 李恒鹏[1] 

机构地区：[1]中国科学院南京地理与湖泊研究所

出　　处：《遥感学报》2006年第6期926-931,共6页NATIONAL REMOTE SENSING BULLETIN

基　　金：国家自然科学基金项目(编号:40401056);知识创新工程特别重点支持项目(编号:CXNIGLAS-A01-1)联合资助

摘　　要：本文以太湖流域西苕溪上游安吉地区SPOT5影像600像元×600像元为试验区,首先采用主成分分析对SPOT5影像进行数据压缩和几何信息增强,再采用小波分析方法对影像进行滤波和噪音处理,利用灰度共生矩阵对高分辨率图像的纹理信息进行分析,以对比度和熵为统计指标,确定对比度和熵的最佳阈值,进行边界匹配和图像的分割,将此分割结果与NDVI阈值法分类结果进行叠合,得到最终的分类结果。试验结果表明:将纹理分析方法应用于图像分类中可区分光谱混淆的地物,光谱与纹理特征结合得到的分类精度高于单纯依靠光谱特征进行分类和单纯依靠纹理分类的分类精度。The threshold of NDVI classification is one of the most popular methods in the classification of remote sensing image. However, based on spectrum characteristics of objects, it cannot correctly identify objects with the same spectrum characteristics and therefore cannot reach the required accuracy. In this paper, we take an area of the upper part of Xitiaoxi River in Anji County as an example and discuss the method of combining texture of high-resolution images with spectrum to improve the accuracy of extracted information of SPOT5 image. Firstly, principal components are extracted from SPOT5 image, and high-resolution texture information is acquired by means of the algorithms of signal decomposition and reconstruction of Mallet＇s Wavelet. Subsequently textures of the SPOT5 image are analyzed using Gray Level Co-occurrence Matrices and selected statistic index. Then the threshold is selected and an optimal threshold is obtained according to contrast and entropy. Objects with same spectrums, such as residence and water body, are identified using image segmentation in virtue of the optimal threshold. Finally, the final result is compared with the classification results based on single spectrum or texture. The result shows that objects with same spectrum are well identified by using texture analysis in image classification, and higher accuracy is obtained than using single spectrum or texture analysis.

关 键 词：主成分分析 纹理分析 小波变换 灰度共生矩阵 阈值法 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] F301.24[自动化与计算机技术—控制科学与工程]