Optical satellites are the most popular observation platforms for humans viewing Earth. Driven by rapidly developing multi-source optical remote sensing technology, content-based remote sensing image retrieval (CBRSIR), which aims to retrieve images of interest using extracted visual features, faces new challenges derived from large data volumes, complex feature information and various spatio-temporal resolutions. Most previous works delve into optical image representation and transformation to the semantic space of retrieval via supervised or unsupervised learning. These retrieval methods fail to fully leverage geospatial information, especially spatio-temporal features, which can improve the accuracy and efficiency to some extent. In this article, we propose a cross-modal contrastive learning method (CCLS2T) to maximize the mutual information of multi-source remote sensing platforms for correlation-aware retrieval. Specifically, we develop an asymmetric dual-encoder architecture with a vision encoder that operates on multi-scale visual inputs, and a lightweight text encoder that reconstructs spatio-temporal embeddings and adopts an intermediate contrastive objective on representations from unimodal encoders. Then, we add a hash layer to transform the deep fusion features into compact hash index codes. Additionally, CCLS2T exploits the prompt template (R2STFT) for multi-source remote sensing retrieval to address the text heterogeneity of metadata files and the hierarchical semantic tree (RSHST) to address the feature sparsification of semantic-aware indexing structures. The experimental results on three optical remote sensing datasets substantiate that the proposed CCLS2T can improve retrieval performance by 11.64% and 9.91% compared with many existing hash learning methods and server-side retrieval engines, respectively, in typical optical remote sensing retrieval scenarios.

With continuous improvement of earth observation technology, source, and volume of remote sensing data are gradually enriched. It is critical to realize unified organization and to form data sharing service capabilities for massive remote sensing data effectively. We design a hierarchical multi-dimensional hybrid indexing model (HMDH), to address the problems in underlying organization and management, and improve query efficiency. Firstly, we establish remote sensing data grid as the smallest unit carrying and processing spatio-temporal information. We implement the construction of the HMDH in two steps, data classification based on fuzzy clustering algorithm, and classification optimization based on recursive neighborhood search algorithm. Then, we construct a hierarchical “cube” structure, filled with continuous space filling curves, to complete the coding of the HMDH. The HMDH reduces the amount of data to 6–17% and improves the accuracy to more than eight times than traditional grid model. Moreover, it can reduce the query time to 25% in some query scenarios than algorithms selected as the baseline in this paper. The HMDH model proposed can be used to solve the efficiency problems of fast and joint retrieval of remote sensing data. It extends the pattens of data sharing service and has a high application value.

It is extremely important to extract valuable information and achieve efficient integration of remote sensing data. The multi-source and heterogeneous nature of remote sensing data leads to the increasing complexity of these relationships, and means that the processing mode based on data ontology cannot meet requirements any more. On the other hand, the multi-dimensional features of remote sensing data bring more difficulties in data query and analysis, especially for datasets with a lot of noise. Therefore, data quality has become the bottleneck of data value discovery, and a single batch query is not enough to support the optimal combination of global data resources. In this paper, we propose a spatio-temporal local association query algorithm for remote sensing data (STLAQ). Firstly, we design a spatio-temporal data model and a bottom-up spatio-temporal correlation network. Then, we use the method of partition-based clustering and the method of spectral clustering to measure the correlation between spatio-temporal correlation networks. Finally, we construct a spatio-temporal index to provide joint query capabilities. We carry out local association query efficiency experiments to verify the feasibility of STLAQ on multi-scale datasets. The results show that the STLAQ weakens the barriers between remote sensing data, and improves their application value effectively.

Optical remote sensing (RS) images serve as a pivotal source of geographic information. Owing to the continuous development of deep learning technology, the evolving demands for multisource optical RS of the public shifted from recognition and acquisition of explicit features to comprehension and application of the fine-grained semantics and relationships implied in images. To address this challenge, we propose a semantic-augmented approach integrated multiview knowledge graph for a comprehensive understanding of optical RS images (RSMVKF). The RSMVKF delves into the structured representations of external knowledge from different human-like cognitive views and further explores the discovery ability of high-level features on the basis of multiple modalities and granularities. Specifically, the RSMVKF consists of two stages. First, we guide a large language model to condense relevant knowledge from lengthy external knowledge passages and generate a view-level knowledge graph (RS-VKG). Then, an asymmetric multimodal contrastive network model (RS-M2CL) is designed to investigate efficient semantic augmentation. In this way, two types of contrastive loss functions, cross-modal and cross-granularity, are adopted to improve the understanding of implicit knowledge. The experimental results demonstrate that the RSMVKF greatly improves several perception tasks and reasoning tasks with rich features in optical RS imagery. In particular, in perception tasks such as fine-grained object detection and k-nearest neighbor retrieval, the RSMVKF yields enhancements of 6.7% and 8.1%, respectively. In addition, in knowledge-driven reasoning tasks such as RS image captioning, RS visual grounding and RS visual question answering, the RSMVKF demonstrates superior performance with margins of 8.9%, 5.3%, and 11.4%, respectively.

Recent advances in Earth observations have enabled the acquisition of high-precision optical remote sensing imagery (ORSI) with sub-meter spatial resolution and hourly revisit periods. While early neural network-based methods have demonstrated notable limitations in visual perception, more recent multimodal fusion approaches still struggle to effectively recognize and reason over geospatial object relations. To fully promote the collaborative analysis of very-high-resolution (VHR) dual-temporal ORSIs, this paper presents a novel framework named R2C-MGR2L that is designed to enhance multi-granularity relational representation in dual-temporal optical remote sensing images through LLM-based scene graph generation (SGG). This framework presents a two-stage learning paradigm—knowledge alignment and cross-modal fusion—that incorporates structured knowledge representations and a hierarchical attention mechanism to improve the refinement and reasoning of fine-grained relational semantics. First, we investigate the types of geospatial object relations and assess the feasibility of enhancing relational semantic representation through multimodal knowledge extraction and embedding. We subsequently employ a pre-trained code large language model (Code-LLM) to transform multigranularity relation triplets from extracted optical remote sensing (ORS) scene graphs into semantically equivalent programs, thereby alleviating the scarcity of domain-specific ORSI knowledge. Moreover, a hybrid-structured asymmetric contrastive learning network is developed to further refine the semantic representations derived from complex scenes. By hierarchically fusing dual-temporal ORSIs and corresponding scene graphs, an integrated representation of visual targets and structural knowledge across multiple depths is achieved. We evaluate our proposed framework on three relation-aware ORS scene understanding tasks. The experimental results suggest that the R2C-MGR2L outperforms the state-of-the-art generic methods, resulting in consistent improvements in accuracy, completeness and timeliness for relational semantic recognition.