Most deep-learning-based vision tasks rely heavily on crowd-labeled data, and a deep neural network (DNN) is usually impacted by the laborious and time-consuming labeling paradigm. Recently, foundation models (FMs) have been presented to learn richer features from multi-modal data. Moreover, a single foundation model enables zero-shot predictions on various vision tasks. The above advantages make foundation models better suited for remote sensing images, where image annotations are more sparse. However, the inherent differences between natural images and remote sensing images hinder the applications of the foundation model. In this context, this paper provides a comprehensive review of common foundation models and domain-specific foundation models for remote sensing, and it summarizes the latest advances in vision foundation models, textually prompted foundation models, visually prompted foundation models, and heterogeneous foundation models. Despite the great potential of foundation models for vision tasks, open challenges concerning data, model, and task impact the performance of remote sensing images and make foundation models far from practical applications. To address open challenges and reduce the performance gap between natural images and remote sensing images, this paper discusses open challenges and suggests potential directions for future advancements.

The rapid growth of remote sensing data offers unprecedented opportunities for global environmental monitoring and resource assessment, yet poses significant challenges for efficient selection of large-scale image datasets. Traditional conditional retrieval methods often return extensive sets with substantial spatial redundancy, imposing heavy selection burdens on users. Existing automated selection methods struggle to balance coverage accuracy, redundancy control, and computational efficiency in large-scale scenarios, making efficient and accurate image selection a critical challenge for large-scale applications. To address this, we propose SCOPES (Spatially-Constrained Optimization for Efficient Image Selection), a novel spatial constraint optimization framework. SCOPES operates directly on actual image footprints in continuous space, thereby circumventing the limitations of traditional discretization-based modeling. We design a unit area cost function aimed at balancing image quality with spatial contribution. To ensure computational efficiency and solution optimization, SCOPES adopts a three-stage “preliminary selection-structural optimization-supplementary selection” strategy: employing lazy greedy for efficient initial selection, spatial Boolean overlay for redundancy control, and supplementary selection for coverage gap repair. Experiments conducted in four regions of different scales demonstrate that compared to baseline methods, SCOPES minimizes the number of selected images and maximizes coverage while achieving a near-universally minimal redundancy ratio. Meanwhile, the introduction of the lazy greedy algorithm significantly improves computational efficiency, achieving up to a 229-fold speedup in the large-scale East Asia region. Overall, SCOPES provides an efficient, accurate, and scalable solution for remote sensing data selection, substantially reducing the manual selection workload for platform users.