【作者】 何帆；

【导师】 田铮；

【作者基本信息】 西北工业大学 ， 应用数学， 2004， 硕士

【摘要】 合成孔径雷达(Synthetic Aperture Radar简称SAR)图像中的乘性噪声和统计的非正态性对传统的图像分割方法提出了挑战。投影寻踪方法是处理高维非下态数据的有效工具，而投影指标的估计又是投影寻踪方法中的核心问题。本文在研究投影指标的小波估计及其统计性质的基础上，对SAR图像的多尺度投影寻踪分割方法进行了探索性研究。 本文研究了投影指标的小波估计，给出了Cook投影指标族、Friedman投影指标、Hall投影指标和投影寻踪判别分析指标的小波估计。证明了Cook投影指标族、Friedman投影指标和Hall投影指标的小波估计的渐近无偏性和均方收敛性，投影寻踪判别分析指标(Projection Pursuit Dscriminant Analysis简称PPDA)的小波估计的渐近无偏性。 在估计投影指标时需要对密度函数进行估计，本文通过实例计算比较了密度函数的核估计和小波估计。其计算实例结果表明在样本量较大时，密度函数的小波估计较核估计好。 最后，本文给出了多尺度投影特征的提取方法和SAR图像的多尺度投影寻踪有监督分割方法。并分别对单波段机载SAR图像和多波段机载SAR图像作了实例分割。其结果表明该分割方法具有一定的实用性。更多还原

【Abstract】 Nowadays, traditional methods of image segmentation have been challenged by the multiply noise and non-nomal data of SAR image. Projection pursuit is an effective method for dimension reduced. The estimation of projection index is the core of projection pursuit method. Based on the study of the wavelet estimator of projection index, we explore the segmentation of SAR image used multi-scale projection pursuit.In this paper, we utilized the wavelet to estimate Cook’s, Friedman’s and Hall’s projection index, and proved that the wavelet estimator of this projection pursuit indices are MSE convergence and asymptotically unbiased., and the wavelet estimator of the PPDA index is asymptotically unbiased.Estimating the projection index required to estimate the density function of projection data. In this paper, we compared the density function of kernel estimator and the density function of the wavelet estimator through simulation. The result indicated that the wavelet estimator better than the kernel estimator in some conditions.At last, we proposed a new character extraction method, so called Multi-scale Projection character. As well we give a way to segment SAR image by Multi-scale projection pursuit. From the segmentation results of the single-band and the Multi-band SAR image, we can see that this segmentation method is effective.更多还原