基于深度卷积神经网络的PolSAR图像变化检测方法  

Polarimetric SAR image change detection based on deep convolutional neural network

作　　者：王剑 王英华[1,2] 刘宏伟[1,2] 何敬鲁[1,2] WANG Jian;WANG Yinghua;LIU Hongwei;HE Jinglu(National Laboratory of Radar Signal Processing,Xidian University,Xi’an 710071,China;Collaborative Innovation Center of Information Sensing and Understanding,Xidian University,Xi’an 710071,China)

机构地区：[1]西安电子科技大学雷达信号处理国家重点实验室,陕西西安710071 [2]西安电子科技大学信息感知技术协同创新中心,陕西西安710071

出　　处：《系统工程与电子技术》2018年第7期1457-1464,共8页Systems Engineering and Electronics

基　　金：国家自然科学基金(61671354;61771362);国家杰出青年科学基金(61525105);民用航天科研工程先期攻关项目资助课题

摘　　要：针对极化合成孔径雷达(polarimetric synthetic aperture radar,PolSAR)图像变化检测问题,提出了结合区域信息和深度卷积神经网络(deep convolutional neural network,DCNN)的PolSAR图像变化检测算法。在本方法中,用超像素分割算法与超像素合并算法提取图像场景的区域信息,利用区域信息和Wishart似然比得到差异图像;再运用预分类算法以得到训练DCNN的伪训练样本和待分类样本;接着用伪训练样本训练DCNN;最后用训练好的DCNN对待分类样本进行分类得到最终结果。实验结果表明,与多种PolSAR变化检测算法相比,所提算法能够获得更好的结果。In order to solve the polarimetric synthetic aperture radar（PolSAR）image change detection problem,a PolSAR image change detection method is proposed combining the region information with deep convolutional neural network（DCNN）.The superpixel segmentation algorithm and the superpixel combination algorithm are utilized for extracting region information,then a difference image is obtained using region information and Wishart likelihood ratio.Second,apreclassification algorithm is used to obtain the pseudotraining samples and the samples which are ready to be classified.Third,the DCNN is trained using the pseudotraining samples.Finally,the trained DCNN is used to classify the samples that are to be classified to get the final results.Experimental results show that,compared with several existing PolSAR change detection methods,the proposed method can get better results.

关 键 词：极化合成孔径雷达 变化检测 区域合并 深度卷积神经网络 

分 类 号：TN958[电子电信—信号与信息处理]