计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (31): 175-178.

• 图形、图像、模式识别 • 上一篇    下一篇

基于纹理和边缘的SAR图像多维SVM回归

龙海翔1,2,高 鑫1,刘 蓉2   

  1. 1.中国科学院 电子学研究所 航空微波遥感系统部,北京 100080
    2.中国科学院 研究生院,北京 100039
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-11-01 发布日期:2011-11-01

Multi-dimensional SVM regression of SAR image based on texture and edge

LONG Haixiang1,2,GAO Xin1,LIU Rong2   

  1. 1.Institute of Electronics,Chinese Academy of Sciences,Beijing 100080,China
    2.Graduate University of Chinese Academy of Sciences,Beijing 100039,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-11-01 Published:2011-11-01

摘要: 合成孔径雷达(SAR)图像ROI(Region of Interest)区域存在两种情况:只包含单一地物或由混合地物组成。对此,提出一种基于特征映射的SAR图像多维输出回归方法,该方法不仅能够对只包含单一地物的SAR图像进行类别判断,也能对混合地物区域的内容做出估计。首先对SAR图像提取基于灰度共生矩阵的纹理特征,然后构造了一组能够反映SAR图像边缘长度、方向和稀疏程度的边缘特征向量,最后利用纹理特征和边缘特征对SAR图像进行基于近似迭代变权最小二乘法(IRWLS)的多维支持向量机(Support Vector Machine,SVM)回归。实验结果表明,该方法能够对包含不同地物内容的ROI区域进行有效解译,正确率高。

关键词: 合成孔径雷达(SAR)图像, 多维SVM回归, 纹理特征, 灰度共生矩阵, 边缘特征, 图像解译

Abstract: For precisely recognizing and interpreting the content of Region of Interest(ROI) of SAR image,which contains either single or mixed geographical objects,a new multi-dimensional regression analysis method based on features-mapping is developed.It can not only classify SAR images containing single geographical objects,but also interpret the region of mixed geographical objects as well,which shows its practicability.It firstly extracts texture features based on gray level co-occurrence matrix from SAR image,and then constructs a set of vectors which can describe the length,the direction and the density of the edge of SAR image.These texture features are used for multi-dimensional Support Vector Machine(SVM) regression based on Iterative Re-Weight Least Square(IRWLS) at last.The experiment results demonstrate that this approach is effective for interpretation of ROI with various contents with high accuracy.

Key words: Synthetic Aperture Radar(SAR) image, multi-dimensional Support Vector Machine(SVM) regression, texture feature, Gray Level Co-occurrence Matrix(GLCM), edge feature, interpretation