一种P ol SAR图像统计模型辨识新方法

摘要

The multi-variant product model is widely applied in the field of PolSAR imagery,whose selection of texture component directly affects the modelwhose accuracy.Aimed at the problem of statistical model recognition for the texture component,an unsupervised method based on covariance matrix log-cumulants (MLC) is proposed.This method colors the second and third MLCs plane,then the PolSAR data are proj ected on the plane,and the statistical model is distinguished by the color of the pixels.The main advantage of the new method is to give a simple and macroscopic result,which can provide important support for the subsequent target detection,identification and classification of PolSAR data.Finally, experiments on the new method are made using simulated data and real PolSAR data and the results show that the new estimator is effective and robust.%多变量乘积模型在极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,PolSAR)图像建模领域中应用广泛,其纹理分量统计模型的选择直接影响到拟合的准确性。针对多变量乘积模型纹理分量分布的选择问题,提出了一种基于矩阵对数累积量(Matrix Log-Cumulant,MLC)的 PolSAR 图像统计模型无监督辨识方法。该方法首先将二阶和三阶 MLC平面进行着色,然后将 PolSAR数据投射到该平面上,根据像素点所在区域的颜色来辨识其对应的统计分布模型。新方法的优点是对全图中各区域的统计模型有简洁、宏观的辨识结果,能为后续的分类识别和目标检测等图像解译手段提供重要支撑。最后,用仿真数据和实测数据对该方法进行了分析。实验结果表明,该方法能实现对图像中不同区域分布模型的有效辨识。