基于U-Net深度学习方法火星沙丘提取研究  

Information extraction of Mars dunes based on U-Net

作　　者：郭晓征 姚云军 贾坤 张晓通 赵祥 GUO Xiaozheng;YAO Yunjun;JIA Kun;ZHANG Xiaotong;ZHAO Xiang(State Key Laboratory of Remote Sensing Science,Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China)

机构地区：[1]北京师范大学地理科学学部遥感科学国家重点实验室,北京100875

出　　处：《自然资源遥感》2021年第4期130-135,共6页Remote Sensing for Natural Resources

基　　金：北京市科技计划课题资助项目“首次火星探测数据反演关键技术研究”(编号:Z191100004319001);国家重点研发计划(地球观测与导航)重点专项课题“基于国产遥感卫星的典型要素提取技术”(编号:2016YFB0501404)共同资助。

摘　　要：火星沙丘遥感识别对于人类探索火星大气与其表面交互作用具有重要的研究意义。针对传统的机器学习方法自动提取火星沙丘精度低的问题,设计了一种纹理特征提取与深度学习相结合的方法来自动识别火星沙丘。该方法在火星遥感影像纹理特征提取的基础上结合深度学习模型对火星沙丘进行提取,实现火星遥感影像端到端的语义分割。同时将U-Net方法提取结果与传统的随机森林提取方法进行对比,实验结果表明:U-Net方法能够充分利用影像中丰富的纹理信息,提取沙丘的准确率为96.7%,比传统的随机森林方法提高了3.2个百分点;U-Net方法提取的火星沙丘轮廓更为准确清晰,且对破碎程度大的沙丘提取效果较好,U-Net方法可用于火星沙丘的精确自动提取。The remote sensing-based information recognition of Mars dunes has important significance for the exploration of the interactions between the Martian atmosphere and the dune surface.Aiming at the low accuracy of the automatic information extraction of Mars dunes using the traditional machine learning method,this paper designs a method combining texture feature extraction and deep learning to automatically identify the information of Mars dunes.In detail,this method conducts information extraction based on the texture feature extraction of Mars remote sensing images combined with a deep learning model,thus realizing the end-to-end semantic segmentation of the remote sensing images.According to experiment results,the U-Net method can fully utilize the rich texture information in the remote sensing images and the extraction accuracy of dunes of this method was 96.7%,which was 3.2 percentage points higher than that of the traditional random forest method.Furthermore,compared to the traditional random forest method,the U-net method extracted more accurate and clearer contours of Mars dunes and yielded better extraction effects of highly fragmented dunes.Therefore,the U-net method can be used for accurate and automatic information extraction of Mars dunes.

关 键 词：火星沙丘 深度学习 随机森林 纹理特征 自动化提取 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]