
摘要
随着遥感数据体量的不断增加,海量遥感数据管理与服务系统已经成为必不可少的基础设施。同时,
业务功能的快速增长对遥感数据管理与服务平台的建设带来了新挑战,传统的基于单体架构的遥感
数据管理与服务应用已经无法满足人们的服务需求。为此,本文设计并实现了一站式遥感大数据分
布式管理与模型训练云平台,平台基于分布式存储-计算模型为基础的微服务架构,该架构能够保
证平台的高可用性和易扩展性,为自研的遥感深度学习在线训练服务提供基础。经过测试,本平台
遥感数据处理效率相较于传统框架都有明显提升,加速比大于 2,满足当前遥感数据快速增长对数
据管理和应用的需求。
Abstract
With the continuous increase in the volume of remote sensing data, remote sensing
data management and service platform has become an indispensable infrastructure.
At the same time, functional requirements have evolved rapidly, which brings new
challenges to the construction of remote sensing data management and service
platform. Traditional applications based on monolithic architecture can no longer
meet people's needs for high availability and expandability. A onestop distributed
management and deep learning platform for massive remote sensing data was
designed and implemented based on the above background. Microservice
architecture based on distributed storage-computing model was used. This
architecture ensures high availability and expandability, providing the basis for self-
developed online remote sensing deep learning services. After testing, this
platform's remote sensing data processing efficiency has been significantly improved
compared with the traditional framework. The speedup ratio is greater than 2, which
meets the requirements of data management and application platform in the context
of the rapid growth of remote sensing data。
译
关键词
遥感数据; 深度学习云平台; 分布式存储; 分布式检索; 微服务
Keywords
remote sensing data; deep learning cloud platform; distributed storage; distributed
retrieval; microservices
译
遥感影像数据在各行各业都被广泛应用,发挥着日益重要的作用
[1]
。随着对地观测技术的不断发展,
遥感影像数据量呈现出明显的大数据特征,遥感图像的处理算法也日益多元化,从而对遥感影像服
务能力的要求也不断提升。开发建设高性能的遥感影像时空数据系统,提升影像存储管理能力和影
像解译处理效率对遥感技术在国土、海洋、环境、气象等领域的应用具有重要意义
[2]
。