“天问一号”着陆区横向风成脊零样本遥感解译与分析  

Zero-shot remote sensing interpretation and analysis of transverse aeolian ridges in the“Tianwen-1”landing area

作　　者：秦楠楠 康志忠[2] 惠建新 陶鹏杰 陈一平 管海燕 童旭东 QIN Nannan;KANG Zhizhong;HUI Jianxin;TAO Pengjie;CHEN Yiping;GUAN Haiyan;TONG Xudong(School of Remote Sensing and Geomatics Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;School of Land Science and Technology,China University of Geosciences,Beijing 100083,China;Purple Mountain Observatory,Chinese Academy of Sciences,Nanjing 210023,China;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;School of Geospatial Engineering and Science,Sun Yat-Sen University,Zhuhai 519082,China)

机构地区：[1]南京信息工程大学遥感与测绘工程学院,南京210044 [2]中国地质大学(北京)土地科学技术学院,北京100083 [3]中国科学院紫金山天文台,南京210023 [4]武汉大学遥感信息工程学院,武汉430079 [5]中山大学测绘科学与技术学院,珠海519082

出　　处：《遥感学报》2025年第2期460-471,共12页NATIONAL REMOTE SENSING BULLETIN

基　　金：国家自然科学基金(编号:42471477);南京信息工程大学人才启动经费(编号:2023r093)。

摘　　要：研究横向风成脊的形貌特征有助于揭示火星稀薄大气层与火星表面之间的相互作用。本文提出了一种基于视觉基础模型的火星横向风成脊零样本遥感解译方法。首先,借助视觉基础模型的通用分割能力,在零样本的情况下实现火星高分辨率影像的精细分割;其次,基于横向风成脊的几何与光谱约束,稳健的从影像所有分割结果中过滤出横向风成脊掩码;最后,利用各个横向风成脊的分割掩码,提取其形貌参数。在公共数据集M-TARset和一幅覆盖“天问一号”着陆点的高分辨率成像相机(HiRIC)正射影像上进行了本文方法的性能测试,结果表明:(1)本文方法在M-TARset上获得了94.33%的精确率、92.32%的召回率以及0.93的F1分数,性能超越了现有其他横向风成脊检测方法;(2)着陆区横向风成脊呈现出总体一致的东西朝向,指示了着陆区主导风向为10°左右的东北风。同时,着陆区横向风成脊的长度、宽度、周长及面积的频率直方图大体服从对数正态分布,这与先前的研究结果一致。本研究可为全球尺度的火星横向风成脊精细解译提供技术支撑,也可为后期火星气候历史重建以及火星全球气候模型完善提供科学依据。Studying the morphological characteristics of transverse aeolian ridges can help reveal the interaction between the Martian thin atmosphere and the Martian surface.In the past few decades,remote sensing exploration has greatly promoted our understanding of Martian aeolian landforms and processes.However,existing research mainly relies on the visual interpretation of remote sensing images,and the substantial amount of manual labor limits its research scope.Although several recent studies have applied deep convolutional neural networks to automatic recognition of transverse aeolian ridges,these methods cannot obtain segmentation masks for individual transverse aeolian ridges and typically rely on a large number of manually annotated training samples.Therefore,developing novel methods is of foremost importance.In response to the above issues,this paper proposes a zero-shot remote sensing interpretation method for Martian transverse aeolian ridges based on the visual foundation model.The overall framework of this method consists of three main steps:First,by leveraging the universal segmentation capability of visual foundation models,fine segmentation of Martian high-resolution images is achieved without training samples.Second,based on the geometric and spectral constraints for transverse aeolian ridges,segmentation masks of transverse aeolian ridges are robustly filtered out from all segmentation masks of the image.Finally,based on the segmentation mask of each transverse aeolian ridge,the corresponding morphological parameters are calculated.On the public dataset M-TARset,this method achieves an accuracy of 94.33%,a recall of 92.32%,and an F1-score of 0.93,and outperforms other state-of-the-art methods for transverse aeolian ridge detection.This method achieves high accuracy and recall on six different regions of test data,and differences in F1-scores obtained in different regions are observed.Compared with flat areas,the overall accuracy of this method is lower in areas with closely distributed transverse aeolian

关 键 词：天问一号 横向风成脊 遥感解译 视觉基础模型 零样本泛化 实例分割 形貌分析 火星气候 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] P2[自动化与计算机技术—控制科学与工程]