摘要
为了更好地对极化合成孔径雷达图像进行分类,提出了一种基于神经网络的混合方法.特征集包括图像的5个H/α系数和基于灰度共生矩阵的6个参数.采用主成分分析方法压缩特征维数,利用3层BP神经网络进行分类,并将Levenberg-Marquardt法与共轭梯度算法相结合求解网络权值.利用该算法对San Francisco地面的实测数据进行分类,实验结果显示该算法能有效分辨地形,且性能优于Wishart最大似然估计方法.
In order to classify polarimetric synthetic aperture radar(SAR) images more accurately,a hybrid method based on neural network is proposed.The feature set consists of five H/α parameters and six gray-level co-occurrence matrix parameters.Principle component analysis is used to reduce the dimensions of the feature set.A 3-layer structure is adopted for BP neural network.The Levenberg-Marquardt method and the conjugate gradient method are used to solve the weights and biases.Experimental results for real data of San Francisco demonstrate that the proposed algorithm is effective and the accuracy is higher than that of the Wishart maximum likelihood method.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2009年第S1期294-298,共5页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(60872075)
东南大学优秀博士学位论文基金资助项目(YBJJ0908)
关键词
极化合成孔径雷达
灰度共生矩阵
主成分分析
polarimetric synthetic aperture radar
gray-level co-occurrence matrix
principle component analysis
