基于半监督学习的遥感影像变化检测研究综述

A review of remote sensing image change detection research based on semi⁃supervised learning

近年来,在人工智能技术与遥感大数据的深度融合下,基于深度学习的全监督变化检测框架通过大量标注数据训练,表现出优异的性能。然而,变化检测数据标注需要逐像素比对两影像间差异,这将消耗大量的人力、时间成本。为解决数据标注的局限性,基于半监督学习变化检测框架逐渐成为变化检测研究热点,该框架能充分利用大量的无标注数据提高变化检测方法的鲁棒性,减少模型对标注数据的依赖。

In recent years,under the deep integration of artificial intelligence technology and remote sensing big data,change detection frameworks based on deep learning have demonstrated excellent performance through extensive training with annotated data.However,the annotation of change detection data requires pixel-to-pixel comparison of differences between two images,which incurs significant human and time costs.To address the limitations of data annotation,semi-supervised learning-based change detection frameworks have gradually become a research hotspot.This framework can fully leverage a large amount of unlabeled data to enhance the robustness of change detection methods and reduce the model’s dependence on annotated data.