SAR图像纹理特征提取与分类研究  

Study on the Extraction of Texture Features and Its Application in Classifying SAR Images

作　　者：胡召玲[1] 李海权[1] 杜培军[2] 

机构地区：[1]徐州师范大学城市与环境学院,江苏徐州221116 [2]中国矿业大学环境与测绘学院,江苏徐州221116

出　　处：《中国矿业大学学报》2009年第3期422-427,共6页Journal of China University of Mining & Technology

基　　金：国家自然科学基金项目(40771143);国家高技术研究发展计划(863)项目(2007AA12Z162)

摘　　要：为了高精度地提取合成孔径雷达(SAR)图像中的有用信息,提出一种基于灰度共生矩阵的纹理特征辅助SAR图像分类方法,该方法选择的是在合适的窗口尺寸下能将各种地物类型区分开的最佳纹理特征组合.采用增强的Frost滤波法对SAR图像进行斑点噪声抑制,通过比较各典型地物基于灰度共生矩阵的纹理特征统计量,确定参与分类的最佳纹理特征组合、计算灰度共生矩阵的最佳窗口尺寸;采用主成分分析法去除各纹理特征之间的相关性,选择信息量大的2个主成分与图像的灰度共同组成3个波段的图像;最后采用最大似然分类法对该组合图像进行分类.结果表明:该方法提取出的纹理特征辅助SAR图像分类,比无纹理信息参与的SAR图像分类,其精度可提高11.20%.To extract useful information with high precision, a method of texture features serving assistance in SAR image classification was proposed on the basis of gray-level co-occurrence matrix. This method made use of the best combination of texture features which can distinguish various types of ground objects in appropriate window size. Firstly, with speckles of a SAR image restrained by the enhanced Frost filter method, statistics of textural features for ground objects based on the co-occurrence matrix were compared to determine the best texture feature combination to be employed in the classification of SAR image and to work out the best window size of gray-level co-occurrence matrix. Then, the principal component analysis method was utilized to remove the correlation among these selected texture features and to select two principal components of the richest information in combination With the gray of the image, thus obtaining an image on three bands. Finally, the new image was classified with maximum likelihood classification method. The results show that the accuracy of SAR image classification assisted by this method is improved by 11.20%, compared with the classification without texture information.

关 键 词：SAR图像 灰度共生矩阵 纹理特征 主成分分析 最大似然分类法 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]