SAR图像分割的多尺度自回归滑动平均模型方法

给出了合成孔径雷达(synthetic aperture radar简称SAR)图像多尺度自回归滑动平均(multiscale autoregressive moving average简称MARMA)模型建模的一种新方法。研究了基于MARMA模型的SAR图像多尺度随机特征提取的方法，构造了相应的分类器．将这类方法用于实际SAR图像分割，并将MARMA模型与多尺度自回归(multiscale autoregressive简称MAR)模型的分割结果进行比较，说明SAR图像的MARMA模型分割方法优于MAR模型分割方法；最后给出了评价SAR图像分割结果的区域均匀性指标方法，实际应用结果表明该评价方法是有效的。

A New Method for Segmentation of SAR Imagery Based on MARMA(Multiscale Autoregressive Moving Average) Model

Existing algorithms for processing optical images are not suitable for Synthetic Aperture Radar (SAR) imagery because it contains a lot of typical noise called speckles. The conventional prerequisite for segmentation of SAR imagery is despeckling. The side effect of despeckling is that much of detail information (including edge and texture) is lost. To tackle this problem, we present a new method of SAR imagery segmentation based on MARMA model, which can capture the statistical scale-dependency of SAR imagery. We review briefly relevant information about multiresolution framework, which is fundamental for our method. We then explain in some detail the three steps used in establishing our segmentation model. Next we investigate how to extract the multiscale stochastic characteristics of SAR imagery. Our new method utilizes the classifier in Ref.4. Finally we propose a criterion for evaluating the quality of segmentation. We applied our new method to a certain SAR imagery. The segmentation quality, as evaluated by our proposed criterion, is better for our new method based on MARMA model as compared with that for the method based on multiscale autoregressive model.