摘要
针对传统PolSAR影像建筑区域提取方法对影像特征利用不充分、自动化程度不高的问题,研究一种基于全卷积网络(fully convolutional networks,FCN)和条件随机场(conditional random field,CRF)相结合的建筑区域提取方法。该方法充分利用FCN网络对影像进行逐像素分类并能自动提取影像高层特征的优势,首先通过制作样本集对FCN网络进行训练;然后利用训练好的模型进行初步的建筑区域提取;最后利用可以联系上下文信息的条件随机场CRF对结果进行优化处理。实验结果表明,该方法可以充分利用影像的语义信息,有效地减少孤立点,提高对细节、轮廓的提取精度,获得较高精度的建筑区域提取结果。
A kind of built-up area extracting method on the basis of integrating the fully convolutional networks(FCN)and the conditional random field(CRF)is researched aiming at the problem that the traditional method for extracting built-up areas from PolSAR images cannot make the best use of the image features and has low automation degree.This method can make the best use of FCN’s advantages of carrying out per-pixel classification to the images and automatically extracting the high-rise characters of such images.Firstly,the FCN is trained by making sample sets.And then,the trained models are utilized for preliminarily extracting the built-up areas.Finally,the CRF that can correlate contextual information is utilized to optimize the results.According to experimental results,this method is capable of making the best use of the semantic information of the images,while effectively reducing the isolated points,improving the extracting precision of the details and outlines as well as obtaining the built-up area extraction results with higher precision.
作者
肖雨彤
张继贤
黄国满
顾海燕
卢丽君
XIAO Yutong;ZHANG Jixian;HUANG Guoman;GU Haiyan;LU Lijun(Chinese Academy of Surveying and Mapping,Beijing 100036,China;National Quality Inspection and Testing Center for Surveying and Mapping Products,Beijing 100036,China)
出处
《遥感信息》
CSCD
北大核心
2020年第3期44-49,共6页
Remote Sensing Information
关键词
POLSAR
建筑区域提取
深度学习
全卷积网络
条件随机场
PolSAR
building area extraction
deep learning
fully convolutional networks
conditional random field
作者简介
肖雨彤(1994—),女,硕士研究生,主要研究方向为极化SAR地物分类,E-mail:514235171@qq.com。
