【摘要】 随着航空航天技术的迅猛发展，可见光、红外、光谱、合成孔径雷达等多模态遥感数据层出不穷，为资源调查、灾害监测、军事侦察等对地观测任务提供了数据支撑。然而，多模态数据在时空和语义上存在巨大差异，导致不同模态数据无法直接融合。多模态对齐将不同模态数据投影至同一空间进行关联，为多模遥感对地观测提供了新的技术途径。本文依据当前遥感技术的发展现状和应用需求，论述了多模态对齐在遥感领域的挑战、现状与机遇，分析了多模态对齐在遥感领域的应用，综述了多模态时空对齐和语义对齐的发展现状，最后结合多模遥感大数据处理中存在的问题和挑战，对多模遥感对齐技术的发展趋势进行了展望。更多还原

【Abstract】 In recent years, the advancement of sensor technology has significantly expanded the application of data from visible light, multispectral, infrared, and synthetic aperture radar sensors across various fields such as precision agriculture, military reconnaissance, and hydrological detection. Multimodal alignment in remote sensing has gained increasing attention, aiming to establish spatiotemporal or semantic relationships between different modal data, thereby achieving efficient data fusion and interoperability. Multimodal alignment is categorized into two main types: semantic and spatiotemporal alignments. Spatiotemporal alignment seeks to synchronize or associate data from different modalities in both time and space, while semantic alignment involves mapping different modal data into a shared high-level semantic space to ensure consistency at the semantic level.Firstly, this study explores the primary application needs of multimodal alignment technology in remote sensing. Key applications include multimodal image registration and fusion, vision-based UAV positioning, and target matching for spatiotemporal alignment, as well as multimodal fusion recognition, image retrieval, and remote sensing change detection for semantic alignment. These applications focus on the spatiotemporal alignment of multimodal remote sensing images to support subsequent image fusion and recognition, and on the semantic alignment of images and text to optimize multimodal information fusion in multi-sensor interactions. Additionally, the study compiles and organizes typical multimodal alignment datasets to support various applications, providing a solid foundation for further research and development in this area.Then, the study reviews the research status of multimodal alignment technology in remote sensing, noting a significant increase in attention over the past decade, particularly in semantic alignment research. It discusses multimodal spatiotemporal alignment from regionbased, feature-based, and deep learning perspectives, highlighting the advantages and disadvantages of each method. It then examines semantic alignment through contrastive learning and generative models, including extensions to image-text-geographic coordinate alignment and addressing high-level issues like logical reasoning, hallucination, and the integration of external structured knowledge in generative models. The study summarizes typical multimodal alignment methods and their successful applications, demonstrating the progress and potential of these technologies.Furthermore, the study details the challenges encountered in real-world remote sensing applications of different multimodal alignment technologies. In spatiotemporal alignment, challenges include large image scale differences, modality specificity, temporal discrepancies, and high computational costs. In semantic alignment, challenges arise from semantic gaps between modalities, poor generality of foundational models, and data scarcity. These challenges underscore the need for continued innovation and refinement in alignment techniques to enhance their applicability and effectiveness in diverse scenarios.Finally, the study forecasts future development directions for multimodal alignment technology in remote sensing. These include researching resource-constrained multi-platform alignment mechanisms, developing foundational models specifically for the remote sensing domain, enriching multimodal alignment datasets, and integrating spatiotemporal with semantic alignment. The study predicts that advancements in these areas will drive innovative applications in environmental monitoring, resource investigation, and national defense security. By addressing current challenges and exploring new frontiers, multimodal alignment technology is poised to significantly impact the field of remote sensing, enabling more accurate, efficient, and comprehensive data analysis and interpretation. This will ultimately lead to enhanced decision-making capabilities and more effective solutions to complex real-world problems.更多还原