基于深度学习的遥感影像变化检测方法进展综述  

A Review of Development on Changing Detection Methods of Remote Sensing Images Based on Deep Learning

作　　者：姜凯悦 赵竹新 马璐[1] 马聪慧 JIANG Kaiyue;ZHAO Zhuxin;MA Lu;MA Conghui(Beijing Institute of Remote Sensing Information,Beijing 100080,China)

机构地区：[1]北京市遥感信息研究所,北京100080

出　　处：《无线电工程》2025年第2期343-356,共14页Radio Engineering

摘　　要：随着人工智能技术与遥感大数据的紧密结合,深度学习技术在多时相遥感影像变化检测中展现出卓越的性能,推动了传统方法向智能化方向的转变,以更加自动化、精细化和智能化的方式解决多领域的地表时空变化检测问题。旨在综述基于深度学习的遥感影像变化检测方法,探讨智能化技术在变化检测方法中的应用潜力。通过介绍深度学习变化检测的通用框架,分析基于像素、面向对象和面向场景变化检测方法的研究进展,讨论数据质量、模型设计和实用性方面的挑战。研究结果表明,深度学习技术显著提高了变化检测的精度和鲁棒性,但需解决多源异构数据处理、模型泛化能力提升和实际应用效率等问题。未来遥感变化检测技术的发展将聚焦于多源数据融合、超分辨率技术和语义变化检测等关键技术,以实现遥感变化检测技术的智能化和精细化发展。With the seamless integration of artificial intelligence technology and remote sensing big data,deep learning techniques have demonstrated exceptional performance in image changing detection of multi-temporal remote sensing,propelling traditional methods towards an intelligent direction.This shift enables a more automated,refined,and intelligent approach to address surface spatiotemporal changing detection issues across various domains.Deep learning-based remote sensing image changing detection methods are reviewed and the potential applications of intelligent technologies in changing detection methodologies are explored.General framework of deep learning change detection is introduced.Then the research progress of pixel-based,object-oriented,and scene-oriented changing detection methods are analyzed,as well as challenges in data quality,model design,and practicality are discussed.Results indicate that deep learning technology has significantly improved the accuracy and robustness of changing detection,but issues such as multi-source heterogeneous data processing,enhancement of model generalization capabilities,and practical application efficiency need to be addressed.Future development of remote sensing changing detection technologies will focus on key technologies such as multi-source data fusion,super-resolution techniques,and semantic change detection,to achieve intelligent and refined development of remote sensing changing detection technology.

关 键 词：遥感影像 智能化 深度学习 变化检测 综述 

分 类 号：P237[天文地球—摄影测量与遥感]