Geospatial Data Cubes (GDCs) raise as a new foundational infrastructure for managing multidimensional heterogeneous spatial datasets. Existing GDC efforts mainly focus on improving computational efficiency, ignoring the optimization of heterogeneous data query due to the lack of semantic connections between them. In this paper, we propose Geospatial Knowledge Cube (GKC) to address the above problem. First, we build a semantic description layer grounded in ontology languages and align it with the schema of GDC. In this way, entities, relationships, and attributes are modeled to enable semantic mappings between heterogeneous datasets. Second, we combine large language models (LLMs) with ontology to transfer natural language queries into machine-readable queries, which significantly improves query efficiency and applicability. Experimental results demonstrate that our method achieves 28.4% higher accuracy than text-to-SQL methods on complex factual queries, 36.3% greater precision for analytical queries, and 22.9% overall accuracy improvement with significantly reduced error rates and enhanced robustness. Furthermore, we applied the GKC framework to flood warning scenarios in the Rhine-Meuse Delta in the case study, which validates the capabilities of GKC.

Existing building annotation methods require significant human resources or other costs, making it challenging to achieve both low cost and high efficiency simultaneously. Crowdsourced OpenStreetMap (OSM) data, with its extensive volume and openness, is widely used for annotation purposes. However, issues such as missing quality information and poor data completeness have hindered its potential to generate deep-learning samples. In this context, our research developed an automated method for generating high-quality building samples based on OSM and deep learning. To address the impact of poor OSM data completeness, we designed a Region-Of-Interest (ROI) generation algorithm to alleviate the negative impact of missing annotations during model training. Leveraging the superior performance of models specialized in the building extraction domain, we devised a method for selecting high-quality samples. Experimental results on the open-source simulation datasets WHU-SIM, MASS-SIM, and real environments in the San Angelo and Washington regions demonstrated the effectiveness of this method. We produced high-quality building samples with a resolution of 0.3 m for the San Angelo and Washington areas, enriching the available data in the building extraction field. This research advances the application of OSM in the remote sensing domain and provides comprehensive insights into its potential for automated sample generation in deep learning.

Remote sensing video (RSV) provides continuous, high spatiotemporal earth observations that are increasingly important for environmental monitoring, disaster response, infrastructure inspection and urban management. Despite this potential, operational use of video streams is hindered by very large data volumes, heterogeneous acquisition platforms, inconsistent preprocessing practices, and the absence of standardized formats that deliver data ready for immediate analysis. These shortcomings force repeated low-level computation, complicate semantic extraction, and limit reproducibility and cross-sensor integration. This manuscript presents a principled multi-level analysis-ready data (ARD) model for remote sensing video, named VideoARD, along with VideoCube, a spatiotemporal management and query infrastructure that implements and operationalizes the model. VideoARD formalizes semantic abstraction at scene, object, and event levels and defines minimum and optimal readiness configurations for each level. The proposed pipeline applies stabilization, georeferencing, key frame selection, object detection, trajectory tracking, event inference, and entity materialization. VideoCube places the resulting entities into a five-dimensional structure indexed by spatial, temporal, product, quality, and semantic dimension, and supports earth observation OLAP-style operations to enable efficient slicing, aggregation, and drill down. Benchmark experiments and three application studies, covering vessel speed monitoring, wildfire detection, and near-real-time three-dimensional reconstruction, quantify system performance and operational utility. Results show that the proposed approach achieves multi-gigabyte-per-second ingestion under parallel feeds, sub-second scene retrieval for typical queries, and second-scale trajectory reconstruction for short tracks. Case studies demonstrate faster alert generation, improved detection consistency, and substantial reductions in preprocessing and manual selection work compared with on-demand baselines. The principal trade-off is an upfront cost for materialization and storage that becomes economical when queries are repeated or entities are reused. The contribution of this work lies in extending the analysis-ready data concept from static imagery to continuous video streams and in delivering a practical, scalable architecture that links semantic abstraction to high-performance spatiotemporal management, thereby improving responsiveness, reproducibility, and cross-sensor analysis for Earth observation.

Fire detection from remote sensing images plays a critical role in early fire warning, fire risk reduction, and post-fire damage assessment. Though deep learning-based approaches have achieved huge successes in fire detection and locating, their data-driven nature requires substantial human resources and time for meticulously labeling pixel-level annotations. The challenge of limited well-annotated labels is still hampering its application. In this paper, we propose a novel label evolution method, Box2Pix, that is to automatically transform box-level labels into pixel-levels labels, alleviating the heavy labor and time cost of pixel-wise fire annotation. Box2Pix leverages box-level annotated object detection datasets with the pre-trained large segmentation models to generate pixel-level annotated semantic segmentation datasets to further boost fire detection performance. Moreover, we introduce innovative point prompt generation methods by leveraging superpixel segmentation and saliency detection to identify potential target regions for generating random points, augmenting the semantic segmentation capability of Segment Anything Model (SAM). Box2Pix eliminates the need for extensive human involvement, avoids reliance on large-scale datasets for training or fine-tuning, enhancing the speed and quality of pixel-level label generation while reducing annotation difficulty and costs. The experiment on the highly generalized Flame And Smoke Detection Dataset (FASDD) demonstrates its superiority over the existing state-of-the-art methods, with improvements of 52.51% in mIoU and 41.38% in overall accuracy. Based on Box2Pix, the largest pixel-level fire detection dataset, Flame And Smoke Segmentation Dataset (FASSD) is generated. The newly created dataset is accessible on: https://github.com/OyamingO/Box2Pix.

Buildings are core carriers of urbanization and socio-economic development. Accurate extraction of buildings from remote sensing imagery is crucial for urban planning, population statistics, and economic assessment. However, existing convolutional neural network (CNN)-based methods struggle to effectively model the complex global contextual information of buildings, and vision Transformers (ViT) face computational constraints that limit their applicability to large-scale very high-resolution (VHR) imagery. To address these challenges, we introduce Mamba - a recently proposed state-space model with linear complexity - into the domain of building extraction. Motivated by the directional alignment and structural symmetry commonly observed in building layouts, we propose the Multi-Directional Adaptive Remote Sensing Mamba (MDA-RSM). Specifically, a Multi-Directional Scanning (MDS) module is designed to flexibly configure or extend Mamba's scanning directions based on task-specific requirements, thereby enhancing the model's adaptability to diverse remote sensing scenarios. On this basis, a Multi-Directional Attention Block (MDA) block is developed to dynamically model the contributions of different scanning directions, reinforce critical directional features, and suppress redundant or noisy information. Extensive experiments on the WHU, Inria, and Massachusetts building datasets demonstrate the effectiveness of the proposed MDA-RSM, achieving state-of-the-art performance under simple training strategies and validating the potential of directional modeling and the Mamba backbone in remote sensing tasks. The implementation will be available at https://github.com/openrsgis/MDA-RSM.

Reducing the reliance of remote sensing semantic segmentation models on labeled training data is essential for practical model deployment. Self-supervised pre-training methods, which learn representations from unlabeled data by designing pretext tasks, provide an approach to address this requirement. One inconvenience of the currently contrastive learning-based and masked image modeling-based self-supervised methods is the difficulty in evaluating the quality of the pre-trained model without fine-tuning for semantic segmentation task. Hence, this paper proposes a pixel-level clustering-based self-supervised learning method, named SDCluster, which allows for a qualitative evaluation of the pre-trained model through visualizing the clustering results. Specifically, SDCluster extends the self-distillation framework to the pixel-level by incorporating the clustering assignment module. Then, clustering constraint modules, including prototype constraint module and semantic consistency constraint module, are designed to eliminate ineffective cluster prototypes and preserve the semantic information of ground objects. Benefiting from the correlation between pixel-level clustering and per-pixel classification of semantic segmentation, experimental results indicate that SDCluster exhibits competitive fine-tuning accuracy and robust few-shot segmentation capabilities when compared to prevalent self-supervised methods. Large-scale pre-training experiment and practical application experiment also prove the generalization ability and extensibility of the proposed method. The code and the dataset for practical application experiment are available at https://github.com/openrsgis/SDCluster.

Building extraction has become a cornerstone for accurately assessing climate change, urban development, and human activities. The substantial variability in imaging conditions and building appearances poses a significant challenge to precise building extraction. While recent work has attempted to integrate foundational models into remote sensing tasks, most approaches focus on supervised methods or fine-tuning streams. There has been limited exploration of unsupervised pipelines, particularly for large-scale building extraction from very high-resolution remote sensing images. In this work, we propose a two-stage unsupervised building extraction method driven by multi-modality foundation models. First, we introduce a zero-shot pseudo-label generation method, guided by the integration of the Segment Anything Model (SAM) and the CLIP model. To address the misclassification of fragmented objects, we design a zoom-out strategy to restore broken segments. Next, we present a hybrid feature fusion network that combines CLIP patch tokens with task-specific features, achieving high data adaptability while maintaining text-related visual features. Extensive experiments demonstrate that our proposed method achieves F1 scores of 56.00% and 62.53% on the Manhattan and WHU Building datasets, respectively, outperforming or matching supervised methods that require 700 training samples. Notably, when tested on two small-scale datasets, our method exhibits superior robustness compared to existing unsupervised domain adaptation approaches, showing 15%-25% less performance variation and demonstrating high adaptability to remote sensing datasets of varying scales.

Change detection (CD) based on multitemporal remote sensing imagery is a crucial step for various Earth observation applications. While deep learning (DL) has revolutionized CD, its data-driven nature demands substantial labeled images for supervised model training, which is costly and time-consuming. This article addresses the challenge of limited training samples by proposing a novel object-based change augmentation (OCA) method. Unlike conventional image-level augmentation methods that can introduce irrelevant contextual dependencies, OCA decomposes the augmentation process into few-shot object classification and foreground-background pasting, thereby generating in-distribution synthetic images with increased change diversity. An object-based training strategy is developed to create a high-confidence binary classifier for pseudosemantic segmentation, facilitating the copy-paste operation. Experimental results on the very-high-resolution remote sensing images demonstrate the superior performance of OCA compared to existing augmentation- and generation-based methods. A comprehensive analysis of parameter sensitivity, adaptability to varying training data volumes, and compatibility with diverse CD methods validates its robustness. This approach provides a practical and effective solution for few-shot CD scenarios, advancing the applicability of DL-based CD methods in training data-limited environments. Codes and data are available: https://github.com/openrsgis/OCA

Earth observation (EO) video analysis emerges as an attractive solution for various applications, including moving vehicle tracking, disaster monitoring, and environmental change detection. Conventional studies focus on single-source video analysis, which limits observations to a narrow temporal window. By extending from single-source to multisource approaches, cross-platform EO video analysis offers a promising solution for continuous object tracking over extended time periods. It should be noted that multisource EO videos refer to videos captured by different platforms or sensors at different locations or times. However, cross EO video tracking presents several unexplored challenges, including spatiotemporal and color misalignments between different sources, the inherent difficulties of detecting small objects from aerial perspectives, and the high error rates encountered in cross EO video object reidentification. Accordingly, in this article, we propose a new framework targeting for cross EO video tiny object tracking (CVTOT) that comprises three core components: 1) a multisource EO alignment module for spatiotemporal and color registration; 2) a modified YOLO framework for tiny EO object detection; and 3) an enhanced motion matrix with a geo-awareness matching strategy to realize robust object reidentification across videos. In addition, we introduce the CVOT benchmark (CVOT21 and CVOT230), the first dataset tailored for CVTOT, with explicit annotations of object handover (enter/exit) events and cross EO video identity links. Extensive experiments on this benchmark demonstrate that the proposed CVTOT pipeline significantly enhances tracking consistency (e.g., IDF1 21.4% up arrow ) and retrieval accuracy (e.g., CVC 74.3% up arrow ) under realistic multisource domain shifts.

Next point-of-interest (POI) recommendation has been applied by many internet companies to enhance the user travel experience. Recent research advocates deep-learning methods to model long-term check-in sequences and mine mobility patterns of people to improve recommendation performance. Existing approaches model general user preferences based on historical check-ins and can be termed as preference pattern models. The preference pattern is different from the intention pattern, in that it does not emphasize the user mobility pattern of revisiting POIs, which is a common behavior and kind of intention for users. An effective module is needed to predict when and where users will repeat visits. In this paper, we propose a Spatio-Temporal Intention Learning Self-Attention Network (STILSAN) for next POI recommendation. STILSAN employs a preference-intention module to capture the user's long-term preference and recognizes the user's intention to revisit some specific POIs at a specific time. Meanwhile, we design a spatial encoder module as a pretrained model for learning POI spatial feature by simulating the spatial clustering phenomenon and the spatial proximity of the POIs. Experiments are conducted on two real-world check-in datasets. The experimental results demonstrate that all the proposed modules can effectively improve recommendation accuracy and STILSAN yields outstanding improvements over the state-of-the-art models.

Earth observation (EO) videos are undergoing rapid expansion due to the swift advancements in aerial, spaceborne, and ground remote sensing technologies enabling the continuous capture of imagery of the Earth's surface. Compared with traditional image-based EO data, EO videos offer persistent EO, rendering a promising observation resource across diverse applications, including climate monitoring and hazard assessment. The continuous observation capability introduces a challenge to the community, i.e., how to effectively manage and fully harness the value of the substantial volume of EO videos. In this article, we propose a novel approach leveraging a spatiotemporal data cube with EO video management to facilitate analysis. It suggests an analysis ready data (ARD) for EO videos, termed as analysis ready video data, which is incorporated into an Earth video cube. The ARD includes semantics at frame, object/trajectory, and event levels. This article presents the cube data organization and query processing for EO videos. A prototype system is implemented to demonstrate the applicability of the approach.

Most deep-learning methods that achieve high segmentation accuracy require deep network architectures that are too heavy and complex to run on embedded devices with limited storage and memory space. To address this issue, this letter proposes an efficient generative adversarial transformer (GATrans) for achieving high-precision semantic segmentation while maintaining an extremely efficient size. The framework utilizes a global transformer network (GTNet) as the generator, efficiently extracting multilevel features through residual connections. GTNet employs global transformer blocks with progressively linear computational complexity to reassign global features based on a learnable similarity function. To focus on object- and pixel-level information, the GATrans optimizes the objective function by combining structural similarity losses. We validate the effectiveness of our approach through extensive experiments on the Vaihingen dataset, achieving an average $F1$ score of 90.17% and an overall accuracy (OA) of 91.92%. Codes are available at https://github.com/qiuluyi/GATrans .

The agriculture sector is highly vulnerable to natural disasters and climate change, leading to severe impacts on food security, economic stability, and rural livelihoods. The use of geospatial information and technology has been recognized as a valuable tool to help farmers reduce the adverse impacts of natural disasters on agriculture. Remote sensing and GIS are gaining traction as ways to improve agricultural disaster response due to recent advancements in spatial resolution, accessibility, and affordability. This paper presents a comprehensive overview of the FAIR agricultural disaster services. It holistically introduces the current status, case studies, technologies, and challenges, and it provides a big picture of exploring geospatial applications for agricultural disaster "from farm to space". The review begins with an overview of the governments and organizations worldwide. We present the major international and national initiatives relevant to the agricultural disaster context. The second part of this review illustrates recent research on remote sensing-based agricultural disaster monitoring, with a special focus on drought and flood events. Traditional, integrative, and machine learning-based methods are highlighted in this section. We then examine the role of spatial data infrastructure and research on agricultural disaster services and systems. The generic lifecycle of agricultural disasters is briefly introduced. Eventually, we discuss the grand challenges and emerging opportunities that range from analysis-ready data to decision-ready services, providing guidance on the foreseeable future.

3D point cloud is an important geometric data structure. In recent years, deep learning for 3D point cloud has attracted more and more attention and has been widely applied in autonomous driving, robotics and some other fields. Existing 3D point cloud deep learning models can guarantee a relatively high accuracy during training while sacrificing the inference speed due to the large amount of model parameters, which poses a challenge for real-time semantic segmentation of 3D point clouds. To accelerate the inference process, alternative solutions optimize the model through model pruning (reducing the amount of parameters and calculations of the model) or model quantization (compressing the amount of data size), which will inevitably lead to loss of accuracy. Therefore, a natural thought is to balance inference speed and training accuracy. In the paper, we propose a plug-and-play dynamic multi-branch neural network module, which can improve the training accuracy without pruning the size of the inference model or hampering the speed of inference. This module refers to the idea of structural re-parameterization, which we named it Rep-PointNet Module. In general, Rep-PointNet Module can efficiently improve the training accuracy of the model while performing lossless compression of the model. The experiment results show that our method beats the state-of-the-art on the ShapeNet dataset with rapid inference speed. The gains in the overall accuracy metric brought by the Rep-PointNet Module are about 1.5% and 2% on the classification task and the segmentation task, respectively.

Using virtual constellations to combine data from multiple satellites has become increasingly important for continuous monitoring of the Earth’s surface and environment. However, there is still a lack of a mechanism to effectively collect and promote these virtual constellations, limiting the applicability and reutilization of remote sensing data sources. This paper proposes an ontology-based approach including providing domain knowledge representation, data standardization, and semantic integration to manage and improve utilization of virtual constellations. Specifically, an ontology for virtual constellation recommendation is presented to represent and reason virtual constellation knowledge, including the composition of virtual constellations, related sensors, satellites, and remote sensing products. The ontology is developed using Protégé and validated through an application example of evapotranspiration. The model can recommend appropriate virtual constellations in specific application scenarios, supporting the reutilization, integration, and sharing of knowledge in the virtual constellation domain.

Recently, deep learning has been widely used in the segmentation tasks of remote sensing images. However, the existing deep learning method most focus on local contextual information and has limited field of perception, which makes it difficult to capture the long-range contextual feature of objects at large scales form very-high-resolution (VHR) images. In this paper, we present a novel Local–global Framework consisting of the dual-source fusion network and local–global transformer modules, which efficiently utilize features extracted from multiple sources and fully capture features of local and global regions. The dual-source fusion network is an encoder designed to extract features from multiple sources such as spectra, synthetic aperture radar, and elevations, which selective fuse features from multiple sources and reduce the interference of redundant features. The local–global transformer module is proposed to capture fine-grained local features and coarse-grained global features, which enables the framework to focus on recognizing multiple-scale objects from the local and global regions. Moreover, we propose a pixelwise contrastive loss, which could encourage that the prediction is pulled closer to the ground truth. The Local–global Framework achieves state-of-the-art performance with 90.45% mean f1 score on the ISPRS Vaihingen dataset and 93.20% mean f1 score on the ISPRS Potsdam dataset.

Road extraction from remote sensing (RS) images in very high resolution is important for autonomous driving and road planning. Compared with large-scale objects, roads are smaller, winding, and likely to be covered by buildings’ shadows, causing deep convolutional neural networks (DCNNs) to be difficult to identify roads. The letter proposes a semantics-geometry framework (SGNet) with a two-branch backbone, i.e., semantics-dominant branch and geometry-dominant branch. The semantics-dominant branch inputs images to predict dense semantic features, and the geometry-dominant branch takes images to generate sparse boundary features. Then, dense semantic features and boundary details generated by two branches are adaptively fused. Further, by utilizing affinity between neighborhood pixels, a feature refinement module (FRM) is proposed to refine textures and road details. We evaluate the SGNet on the Ottawa road dataset. Experiments show that the SGNet outperforms other competitors on the road extraction task. Codes is available at https://github.com/qiuluyi/SGNet.

Plant diseases and pest infections are major factors that undermine the growth of plants along with their life cycle. Optical image-based plant disease detection provides an efficient and low cost way for real-time plant growth monitoring and management. In recent years, the thriving development of deep learning techniques in a variety of communities has validated its great performance in image interpretation and understanding. Existing deep learning-based methods for plant disease classification mostly adopt convolutional neural networks (CNNs) that have been originally developed for general image classification purposes. These CNN architectures consist of a very large volume of training parameters, which severely hinders its applicability under scenarios requiring fast and flexible deployment on compact devices with limited computation powers. In this paper, an ultra-lightweight efficient network (ULEN) is proposed targeting image-based plant disease and pest infection detection. The proposed network consists of two parts, a deep feature extraction module that adopts residual depth-wise convolution and a classification module receiving multi-scale features enhanced by a spatial pyramid pooling layer. The network is constructed in a very compact design with approximately only 100 000 parameters, which greatly favors the demand for a lightweight model for practical needs. Two publicly available plant datasets collected at the indoor and outdoor environments were tested on two compact devices to validate its applicability under different scenarios. Compared with the state-of-the-art architectures, the proposed network showed superior performance with the least computation complexity and compelling classification accuracy.

Artificial Intelligence Machine Learning (AI/ML), in particular Deep Learning (DL), is reorienting and transforming Earth Observation (EO). A consistent data model for delivery of training data will support the FAIR data principles (findable, accessible, interoperable, reusable) and enable Web-based use of training data in a spatial data infrastructure (SDI). Existing training datasets, including open source benchmark datasets, are usually packaged into public or personal repositories and lack discoverability and accessibility. Moreover, there is no unified method to describe the training data. Here we propose a training data model for AI in EO to allow documentation, storage, and sharing of geospatial training data in a distributed infrastructure. We present design rationales, information models, and an encoding method. Several scenarios illustrate the intended uses and benefits for EO DL applications in an open Web environment. The relationship with Open Geospatial Consortium (OGC) standards is also discussed, as is the impact on an AI-ready SDI.

Heterogeneous remote sensing source-based change detection with optical and SAR data and their combined all-time and all-weather observation capability provides a reliable and promising solution for a wide range of applications. State-of-the-art supervised methods typically take a two-stage strategy that suffers from the loss of original image features and the introduction of noise on the transferred images. This paper proposes a domain adaptation-based multi-source change detection network (DA-MSCDNet) suitable to process heterogeneous optical and SAR images. DA-MSCDNet employs feature-level transformation to align inconsistent deep feature spaces in heterogeneous data. Feature space transformation and change detection are bridged within the network to encourage task communication. Experiments are conducted on two public datasets based on Sentinel-1A and Landsat-8 imagery acquired over the Sacramento, Yuba, and Sutter Counties (California, USA), and QuickBird-2 and TerraSAR-X imagery over Gloucester (UK), as well as one new large-scale dataset of Sentinel-2 and COSMO-SkyMed imagery over Wuhan (China). Compared with other six supervised and unsupervised approaches, the proposed method achieves the highest performance with an average precision of 80.81%, recall of 84.39%, mIOU of 73.67% and F1 score of 82.58%, beating the state-of-the-art method with 5.42% improvements on F1 score and 10 times efficiency on training time cost on the large-scale change detection task.