省级基础地理信息数据遥感解译样本探索与制作  

Exploration and Production of Remote Sensing Interpretation Samples of Provincial Basic Geographic Information Data

作　　者：任光耀 张海 侯恩兵 余永生 汪涵 REN Guangyao;ZHANG Hai;HOU Enbing;YU Yongsheng;WANG Han(The Second Surveying and Mapping Institute of Anhui Province,Hefei 230601,China)

机构地区：[1]安徽省第二测绘院,安徽合肥230601

出　　处：《测绘与空间地理信息》2024年第11期63-66,共4页Geomatics & Spatial Information Technology

基　　金：2024年度安徽省测绘局科技创新项目(2024-KJ-02)资助。

摘　　要：当前空天地一体化遥感数据海量且每日剧增,具备多源、异构、信息量丰富、多尺度等大数据特点,能够为深度学习在遥感信息智能提取中提供大量训练样本,但遥感影像智能解译样本库之间的分类体系、精度、制作方法差异较大,跨数据集用于深度学习模型进行训练时,易导致深度学习效果不佳。本文针对基础地理信息数据更新的特点,提出基于TensorFlow框架的多源卫星遥感影像智能解译样本库设计方案,研究遥感影像智能解译样本库的分类体系、精度要求、制作方法,采用文件数据库管理的方式,通过C#、ArcGISRuntime SDK技术开发样本库管理平台,实现卫星遥感影像智能解译样本库高效生产及管理。At present,the integrated remote sensing data of air,space and earth is massive and increasing daily,with the big data characteristics of multi-source,heterogeneous,information-rich,multi-scale,etc.,which can provide a large number of training samples for deep learning in intelligent extraction of remote sensing information.However,the classification system,accuracy and production methods of intelligent interpretation sample databases of remote sensing images are quite different from each other.It is easy to lead to poor deep learning effect.Aiming at the characteristics of basic geographic information data updating,this paper proposes the design scheme of multi-source satellite remote sensing image intelligent interpretation sample database based on TensorFlow framework,studies the classification system,accuracy requirements and production methods of remote sensing image intelligent interpretation sample databases,and adopts the method of file database management.A sample database management platform was developed by C# and ArcGIS Runtime SDK technology to realize efficient production and management of satellite remote sensing image intelligent interpretation sample database.

关 键 词：遥感智能解译 样本库 样本分类 样本管理 深度学习 

分 类 号：P237[天文地球—摄影测量与遥感]