遥感图像的语义信息提取正成为城市规划利用、土地覆盖勘察、灾害变化检测以及海上态势感知等研究方向的关键技术之一。文中从由“单源”向“多源”发展的遥感图像智能处理需求出发,首先概述并分析了大数据时代和深度学习背景下的遥感图像语义分割发展现状,主要包括单一来源图像语义分割、多源遥感图像融合语义分割和多源(同质/ 异质)遥感图像变化检测。然后在阐述主要方法的基础上,提炼并总结了多源遥感图像语义分割的关键技术,主要有单源遥感图像快速语义分割、语义信息辅助的多源遥感图像精确配准与融合、基于多源遥感图像的语义信息智能提取。最后,针对多源遥感图像在轨处理需求,概括出高分辨率多源遥感图像智能一体化信息提取所面临的技术挑战,并对未来发展趋势进行展望。

Semantic information extraction of remote sensing images is becoming one of the key technologies in urban planning and utilization, land cover survey, disaster change detection and maritime situational awareness. Starting from the intelligent processing requirements of remote sensing images developed from single-source to multi-source, the development status of semantic segmentation of remote sensing images in the era of big data and in the context of deep learning is summarized and analyzed firstly in this paper,including single-source image semantic segmentation, multi-source remote sensing image fusion semantic segmentation and multi-source(homogeneous/ heterogeneous)remote sensing image change detection. Then, on the basis of expounding the main methods, the key technologies of semantic segmentation of multi-source remote sensing images are refined and summarized, including fast semantic segmentation of single-source remote sensing images, accurate registration and fusion of multi-source remote sensing images assisted by semantic information, and intelligent extraction of semantic information based on multi-source remote sensing images. Finally, aiming at the on-orbit processing requirements of multi-source remote sensing images, the technical challenges faced by intelligent integrated information extraction of high-resolution multi-source remote sensing images are summarized,and the future development trend is prospected.

TP751.1

国家重点研发计划资助项目(2021YFA0715202);国家自然科学基金资助项目(62293544,62388102);山东省优秀青年创新团队资助项目(2021KJ005)