Search Results (1,152)

Orbital remote sensing platforms increasingly rely on CNN-based object detection for real-time situational awareness. However, deploying these models on spaceborne edge devices is challenging because of stringent Size, Weight, and Power (SWaP) constraints. In addition, the branch-and-merge topology of conventional single-stage detectors increases on-chip memory usage and introduces pipeline stalls, limiting efficient FPGA implementation. To address these challenges, we proposed RS-YOLO, an object detection algorithm developed through a hardware–software co-design approach. Structural re-parameterization converts heterogeneous branches into a sequential stream of padding-free convolutions, producing a deterministic dataflow and reducing per-state combinational control complexity and data-path multiplexing overhead. To mitigate the high-entropy concentration at the center of the re-parameterized kernels, we further introduce a spatial heterogeneous quantization (SHQ) engine. The SHQ engine assigns 16-bit precision to the central coefficients while preserving vectorized 8-bit computation for peripheral elements, reducing quantization errors for small targets with minimal hardware overhead. Experimental results on the Xilinx Zynq-7020 platform show that the proposed system consumes only 2.24 W while achieving a mean Average Precision (mAP) of 0.887 on the NWPU VHR-10 dataset, representing a 1.4% decrease compared with the FP32 baseline. The system also achieves an energy efficiency of 15.19 GOPS/W, demonstrating an effective balance between hardware efficiency and detection performance for resource-constrained edge platforms such as micro-satellite payloads. Full article

Pb-Zn resources are critical strategic assets for many nations. The Dian-Dongbei (northeastern Yunnan) region in Yunnan Province is a significant production area for these resources in China, boasting considerable prospecting potential. However, conventional exploration methods are increasingly inadequate, as they often fail to rapidly and effectively identify concealed mineralization information. To tackle this challenge, we propose a hybrid GAN-ResNet convolutional neural network methodology. This approach constructs a data-driven prospecting model for Pb-Zn deposits in the Dian-Dongbei region, utilizing multi-source geoscientific data encompassing geology, geophysics, geochemistry, and remote sensing (Geo-Phys-Chem-RS) to conduct quantitative mineral prospectivity mapping. A GAN model was introduced to augment the multi-source geoscientific data based on the concepts of random down-sampling and pseudo-window size. The quality of the generated synthetic samples was evaluated using the Peak Signal-to-Noise Ratio (PSNR) metric. The results show that the synthetic samples achieved an average PSNR value of 33.67 dB, effectively preserving the original features of the geoscientific data. This confirms the feasibility and quality of the data generated by this augmentation method. Furthermore, when applied to train the ResNet model, this augmented data effectively increased the prediction accuracy from 0.765 to 0.842. The results demonstrate that the integrated GAN-ResNet method produces prediction maps with higher accuracy. Moreover, it significantly refines and narrows down the target areas with high mineralization potential. This precision can substantially reduce exploration costs, representing a marked improvement in prediction efficacy. Full article

High-resolution remote sensing visual question answering (RS-VQA) requires models to identify question-relevant regions and reason over fine-grained visual evidence. However, existing vision–language models usually rely on fixed global image inputs, which may lose critical local details in ultra-high-resolution imagery and struggle with sparse informative regions, large object-scale variations, and complex spatial layouts. To address these challenges, this paper proposes a crop-tool-augmented active perception framework with reinforcement learning. The framework introduces structured tokens to explicitly organize the reasoning process into question understanding, cropping decision-making, local evidence acquisition and final answer generation. Based on this design, the model can actively determine whether a cropping operation is needed and select task-relevant regions for further inspection. To enable stable tool-use and multi-turn reasoning in a compact vision–language model, we construct teacher-guided cropping reasoning trajectories from high-resolution images, question–answer pairs, and annotated regions in the LRS-GRO dataset, and use them for cold-start supervised fine-tuning of Qwen2.5-VL-3B. Furthermore, we introduce Group Relative Policy Optimization to refine the model’s active perception policy. A region-aware reward function is designed by integrating output-format constraints, reference-region coverage, answer semantic consistency, and cropping penalties, which encourages compact and informative region selection while reducing redundant tool invocations. Experiments on VRSBench, MME-RealWorld-RS, XLRS-Bench, and LRS-VQA demonstrate that the proposed method achieves competitive overall performance compared with closed-source, open-source, and remote-sensing-specific vision–language models, and obtains the best or comparable results on most benchmarks. Ablation studies further verify the effectiveness of structured supervised fine-tuning, reinforcement learning optimization, and the proposed reward design. Full article

Few-shot remote sensing scene classification aims to accurately recognize unseen scene categories using only a scarce number of labeled samples, which has emerged as a research hotspot in the field of remote sensing image interpretation. However, remote sensing images intrinsically suffer from large intra-class variations, high inter-class similarities, and complex background interferences. Traditional few-shot learning methods typically perform feature metric learning in Euclidean space, making it difficult to capture the non-Euclidean geometric distribution characteristics of remote sensing features, and they often neglect the spatial structural information embedded in feature maps. To address these issues, this paper proposes a novel few-shot remote sensing scene classification method, termed ZSFS-RGBN, which integrates a Zigzag Scanning Feature Sequence with a Riemannian Geometric Barycenter Network. Specifically, ResNet12 is first employed as the backbone to extract deep convolutional feature maps from both the support and query sets. Second, a Zigzag scanning strategy is introduced to reorganize the two-dimensional feature maps into one-dimensional feature sequences, thereby effectively preserving the spatial locality and structural continuity of the features. Third, an autoregressive moving average (ARMA) model is constructed to characterize the spatial dependencies of the feature sequences, and its state parameters are mapped onto a symmetric positive definite (SPD) matrix manifold, enabling the deep semantic representations of remote sensing scenes in a non-Euclidean geometric space. Finally, a Riemannian geometric barycenter network is designed to learn the Riemannian barycenter of each category on the SPD manifold, where a joint loss function is introduced to simultaneously optimize intra-class compactness and inter-class separability. Comprehensive experiments are conducted on three public remote sensing scene datasets: NWPU-RESISC45, UC Merced Land-Use, and WHU-RS19. Experimental results demonstrate that the proposed method consistently outperforms several representative state-of-the-art approaches under both 5-way 1-shot and 5-way 5-shot settings. Furthermore, ablation studies verify the effectiveness of each component within the proposed framework. Full article

Remote sensing ship detection plays a pivotal role in maritime surveillance, safety assurance, and traffic management. However, current detection methods often face significant challenges due to complex sea-surface background noise, large target-scale variations, and edge-hardware limitations. In this paper, we propose RepLite-YOLO, a lightweight detection framework based on YOLOv11n. Specifically, to alleviate irreversible spatial information loss during downsampling, we adopt the ADown module, originally introduced in YOLOv9, to generate spatially complementary features through its two-branch downsampling mechanism. This design helps preserve salient hull-edge responses while suppressing part of the random sea-surface interference, thereby improving feature robustness for small ship targets. To achieve substantial structural streamlining while maintaining competitive representational capacity under strict hardware constraints, we design the C3k2_OREPA_RS module, utilizing online re-parameterization (OREPA) to efficiently reconstruct deep layers without additional re-parameterization-induced inference operations. Furthermore, we construct the ELANFusion_Block by integrating Depthwise Separable Convolutions (DSC) into the ELAN paradigm to alleviate the multi-scale aggregation bottleneck, and tailor the Detect_DWLite head for highly compressed decoupled prediction. Experimental results show that RepLite-YOLO achieves a favorable balance between detection accuracy and computational efficiency. Compared with YOLOv11n, it reduces the number of parameters by 57.4% and GFLOPs by 49.2%, while maintaining competitive detection accuracy with slight mAP@50 improvements of 1.2 and 1.3 percentage points on the Vessel dataset and Ship Detection dataset, respectively. Full article

Remote sensing scene classification remains challenging due to substantial object-scale variations, complex background interference, and high inter-class similarity. To address these issues, a lightweight classification framework, termed MDC-MobileNetV3, is proposed based on the MobileNetV3-Large backbone. The framework integrates a Multi-Scale Feature Extraction (MSFE) module for capturing spatial information at different receptive fields, a Dynamic Feature Weighted Fusion (DFWF) mechanism for adaptive feature recalibration, and the hierarchical CBAM attention strategy to enhance discriminative region representation. The model achieved high classification accuracies of 99.52%, 91.54%, 96.48%, 97.35%, 92.43%, and 99.72% on the UC Merced, WHU-RS19, NWPU-Resisc45, AID, CLRS, and PatternNet benchmark datasets, respectively, validating the effectiveness of the proposed framework, while maintaining a lightweight architecture with approximately 4.35 M parameters. In addition, Grad-CAM visualizations indicate that the model effectively focuses on semantically meaningful regions and suppresses irrelevant background information. The results confirm that the proposed framework provides a favorable trade-off between classification accuracy, model lightweight design, and model interpretability for remote sensing scene understanding. Full article

Remote sensing image super-resolution plays an important role in object recognition, urban monitoring, and fine-grained remote sensing interpretation. This paper studies lightweight single-image remote sensing image super-resolution, in which only one LR observation is available and the model must recover reliable structural details under a limited computational budget. Existing lightweight methods reduce parameter counts and computational complexity, but their limited representation capacity often causes blurred boundaries, broken road structures, and missing high-frequency details in buildings, roads, and texture-rich regions. To address these issues, we propose SFEFeNet, a Structure-Frequency Mutual-Guided Lightweight Network for remote sensing image super-resolution. First, we design a Lightweight Structure-Frequency Block (LSFB) to jointly model local spatial features, structural responses, and frequency responses with low computational overhead. Second, we introduce a Structure-Frequency Mutual Guidance (SFMG) module, where edge responses guide high-frequency component selection, and the selected high-frequency responses further refine edge-aware attention. Finally, we propose a Structure-Frequency Fusion Gate (SFFG) to adaptively integrate lightweight features, local spatial features, frequency-enhanced features, and structure-refined features. Experiments on RSSCN7, DOTA, and WHU-RS19 datasets evaluate SFEFeNet in terms of reconstruction quality, visual performance, and model complexity. Additional analyses further examine structural preservation, complex synthetic degradation, real-image generalization, and statistical stability. Notably, SFEFeNet-Lite contains 0.539 M parameters and 17.07 G FLOPs for $\times 2$, and 0.622 M parameters and 7.12 G FLOPs for $\times 4$, enabling effective structure-frequency feature modeling with lightweight computational cost. Full article

Urban land use intensity (U-LUI) is a widely used term for describing urban development processes, yet its conceptualisation and measurement remain inconsistent. Existing approaches focus on isolated dimensions, such as structural density, functional activity, and socio-economic indicators, resulting in limited comparability and weak integration across scales and data sources. This paper reviews and synthesises current approaches to U-LUI with a focus on remote sensing (RS), in situ data and emerging urban data sources. It analyses definitions, related concepts of urban intensity and existing monitoring frameworks at national, European and global levels, and compares methodological approaches for observing U-LUI. Based on this synthesis, U-LUI is defined as a continuous, multidimensional and spatio-temporally dynamic property of urban systems that reflects the intensity of anthropogenic use. To operationalise this concept, the paper develops an integrative, trait-based framework comprising six indicator families: traits, genesis, structure, taxonomy, function and socio-economics. The proposed framework is illustrated and supported through the synthesis of existing RS approaches, urban monitoring concepts and representative examples from the literature, demonstrating its potential for consistent and scalable U-LUI assessment. These dimensions link physically observable characteristics with functional and contextual aspects of urban systems and provide a basis for more consistent quantification and comparison. The results highlight key challenges for U-LUI monitoring, including limited conceptual harmonisation, incomplete integration of dimensions and the need for improved data integration. The proposed framework supports more coherent and scalable assessments of U-LUI in research, monitoring and planning contexts. Full article

Haze commonly degrades visible-spectrum remote sensing (RS) images by reducing contrast and distorting colors. Existing RS dehazing methods still face two limitations. Prior-driven methods rely on handcrafted assumptions that may become unreliable in complex wide-area scenes without explicit sky regions. Learning-based methods require paired training data, yet real aligned hazy/haze-free RS image pairs are difficult to collect, which limits their real-world generalization. To address these limitations, we propose a method called Remote Sensing Image Dehazing via RGB-Space Physical Constraints (RDPC). The new method revisits the atmospheric scattering model (ASM) from the perspective of RS imaging and builds the restoration process on several physical properties of hazy image formation. For atmospheric light estimation, the RGB-space line-convergence behavior of local regions with similar reflectance and slight depth variations is exploited, allowing atmospheric light to be estimated without explicit sky areas. For transmission estimation, the geometric relation between observed pixels and atmospheric light is used in RGB space, where local perpendicularity provides physically plausible haze-removal guidance and global compensation helps avoid excessive darkening and color degradation. The estimated transmission and albedo guidance are further refined by enforcing ASM consistency and variation sparsity through joint optimization. Experiments on synthetic and real-world RS image dehazing benchmarks demonstrate that RDPC achieves competitive performance against representative prior-based and learning-based methods, including Image Dehazing and Exposure (IDE), Iterative Predictor-Critic (IPC), Curvature-to-Plane Prior (C2P), Adaptive Structure-Texture Awareness (ASTA), Asymmetric U-Net (AU-Net), Efficient Multi-scale Prior Fusion (EMPF), and Lightweight Feature Dehazing (LFD), in terms of peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), learned perceptual image patch similarity (LPIPS), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), neural image assessment (NIMA), and processing time. Full article

Mycotoxin contamination of cereal crops poses escalating food safety risks across the Central and Balkan European (CBE) corridor under climate change, yet no PRISMA 2020-compliant synthesis of remote sensing (RS) and machine learning (ML) evidence for this region exists. We conducted an AI-augmented systematic review applying a four-stage automated pipeline—PICO domain scoring, SBERT semantic deduplication, and Thompson-sampling reinforcement learning—to 36,038 corpus records (2010–2025), yielding 156 included studies (inter-rater κ = 0.81 (95% CI: 0.74–0.88)). Logistic growth modelling identified a 56-fold corpus expansion with inflection at t₀ = 2024.8 (R² = 0.981). Satellite multispectral imaging dominated the literature (91.7% of studies); random forest and gradient boosting models achieved R² = 0.74–0.80 for aflatoxin B1 and deoxynivalenol prediction in CBE maize and wheat when integrating vegetation indices, land surface temperature, and precipitation covariates. Deep learning surpassed classical ML in annual study count from 2021, reaching ~60% relative share by 2025, though the performance advantage narrows at field scale relative to laboratory hyperspectral benchmarks (98–99% accuracy). A five-percentage-point CBE–global performance gap is largely consistent with differences in sample size and multi-toxin design scope rather than algorithmic access. The country × mycotoxin gap matrix identifies zero eligible studies for four CBE nations and for T-2/HT-2 toxins across the Balkan states. Climate-driven satellite mycotoxin prediction emerges as the field’s active research frontier. Full article

Zero-shot remote sensing is attractive for scene classification because new regions, sensors, and label taxonomies often appear before sufficient annotated data are available for supervised adaptation. We present OATS-RS, an inference-centric framework that keeps a remote sensing vision–language model (VLM) backbone frozen and improves zero-shot decisions through ontology-aware prompt construction, hierarchical and contrastive scoring, adaptive multi-view aggregation, unlabeled transductive refinement, ambiguity-aware local re-ranking, and selective prediction. The method targets the common remote sensing regime in which neighboring classes such as annual crop, permanent crop, forest, pasture, herbaceous vegetation, river, and sea or lake overlap strongly in red–green–blue (RGB) appearance, meaning that they require more than a single class-name prompt. On the supplied final EuroSAT RGB evaluation with a GeoRSCLIP Contrastive Language–Image Pre-training (CLIP)-family Vision Transformer Base with 32 × 32-pixel patches (ViT-B-32) backbone, the complete pipeline obtains top-1 accuracy of 0.522, balanced accuracy of 0.522, macro-averaged F1 score (macro-F1) of 0.535, and top-3 accuracy of 0.887. The strongest classes are industrial area, residential area, river, highway, and pasture, whereas the weakest classes remain herbaceous vegetation and several fine-grained vegetation categories. Selective prediction increases accepted-example accuracy to 0.538 at 0.934 coverage, but the expected calibration error (ECE) remains high at 0.384. These results support a qualified conclusion: ontology-guided zero-shot inference can already recover useful semantic shortlists for structured remote-sensing scenes, but fine-grained natural-class disambiguation, calibrated confidence, multi-dataset transfer, component-level ablations, and measured runtime remain essential before dependable deployment claims can be made. Full article

Unmanned aerial vehicle (UAV) visual navigation relies critically on robust image matching between UAV-acquired aerial imagery and pre-existing satellite reference maps. However, extreme cross-domain heterogeneity—encompassing temporal, radiometric, viewpoint, and sensor variations—causes severe performance degradation in existing deep learning-based matchers trained on conventional benchmarks. Furthermore, manual annotation of ground-truth correspondences is prohibitively expensive. This paper proposes a semi-supervised saliency-aware image matching framework with bi-level meta-learning. Our approach comprises two synergistic stages: (1) automated dense correspondence generation via parameterized geometric synthesis, which constructs a large-scale coarse dataset ${\mathcal{D}}_c$ (approximately 50,000 pairs) without dense manual point annotation, serving as the primary training corpus for the feature matching network; (2) expert-validated meta-data curation producing a high-quality meta-dataset ${\mathcal{D}}_m$ (500 pairs) that supervises the training of a Saliency Judgment Network through bi-level meta-optimization, enabling the network to identify and prioritize geometrically reliable correspondences. Experimental results on the proposed RS-Hetero-50K benchmark and cross-domain FuJian-Mountain dataset demonstrate substantial improvements over representative sparse and detector-free matchers, including LoFTR, SuperGlue, and LightGlue. The complete CNN-attention and saliency-aware framework achieves 95.4% matching precision, which is consistent with the best result reported in the experimental section. The plug-and-play experiments further confirm that the proposed saliency module consistently improves representative sparse and detector-free matchers, indicating that the performance gain stems from both stronger feature representation and saliency-guided correspondence selection. The largest terrain-specific gain is observed in gobi scenes, where the AUC@5 px improves by 16.8% relative to the LoFTR baseline, demonstrating improved robustness in weakly textured remote sensing environments. Full article

Monitoring inland surface water quality is essential for water security, ecosystem conservation, public health, and sustainable water resource management. Although in situ measurements remain indispensable, they are often limited by high costs, restricted spatial coverage, low temporal frequency, and discontinuous monitoring networks. This study presents a systematic literature review, guided by the PRISMA 2020 framework, of empirical studies published between 2021 and 2025 on the integration of artificial intelligence (AI) and remote sensing (RS) for inland surface water quality monitoring. Searches were conducted in the Web of Science database, resulting in a final corpus of 367 peer-reviewed articles. Preliminary bibliometric characterization and qualitative content analysis were performed to identify sensors, platforms, AI paradigms, algorithms, estimated parameters, validation strategies, limitations, challenges, trends, and research gaps. The results show rapid growth in the field, with Sentinel-2 and Landsat-8 as the most recurrent sensors and multispectral data as the dominant spectral source. Machine learning approaches, especially Random Forest, Artificial Neural Networks, XGBoost, and Support Vector Machine, predominated, while deep learning, multi-source integration, hybrid models, and Explainable AI emerged as relevant trends. AI–RS integration shows strong potential to complement conventional monitoring, but persistent challenges remain regarding in situ data dependence, limited external and temporal validation, model transferability, generalization, uncertainty reporting, validation robustness, and interpretability. Full article

Remote sensing referring segmentation aims to extract the exact region of an object in an aerial image based on a natural language description, but it remains challenging because remote sensing scenes cover large areas, many objects look similar, and the descriptions are often long and detailed. Existing attention-based models are computationally expensive on large images and may underuse fine-grained language cues, which can lead to inaccurate or incomplete masks. To address this, we present MSMamba, an efficient framework built on a state space model for stable long-range context modeling over large spatial grids. We further strengthen language grounding by identifying descriptive words in the expression and using them to guide visual features from coarse localization to boundary refinement. In addition, we design a scale-aware decoding strategy that fuses multi-scale features with adaptive gating to better handle severe size variation and thin structures. Experiments on four public benchmarks show that MSMamba consistently improves segmentation quality. On RefSegRS, MSMamba improves Pr@0.8 on the test set by 25.53% and increases mIoU by 6.65%. On RRSIS-HR, MSMamba improves Pr@0.8 by 9.09% and increases mIoU by 3.02%. These results suggest that combining a state space model with structured language guidance and scale-aware fusion is a practical alternative to attention-only designs for remote sensing referring segmentation. Full article

Understanding the adaptive dynamics of social-ecological systems (SESs) is critical for regional sustainability as human–environment interactions intensify. However, existing indicator-based research frequently lacks a clear theoretical framework and methodological clarity when analyzing SES adaptation. Using complex adaptive system (CAS) theory as an interpretive lens, this research creates a social-ecological system (SES) adaptability evaluation framework that incorporates the pressure–state–response (PSR) model from a CAS perspective. This study examines the Huaihe River Ecological and Economic Belt (HREEB) as a case study, combining remote sensing (RS) and geographic information system (GIS) data from 28 prefecture-level cities from 2005 to 2020. The entropy-weight approach is used to create a composite adaptability index, and obstacle-degree analysis is used to identify key limiting factors, followed by an examination of spatiotemporal evolution patterns. The study found that: (1) SES adaptability in the HREEB increased steadily (mean annual growth rate: 3.97%), with the social subsystem exhibiting a larger connection with the overall trend and the ecological subsystem displaying greater volatility; (2) there was significant spatial heterogeneity, forming a “high in the east and west, low in the center” pattern (supported by a global Moran’s I = 0.535, p < 0.05); (3) obstacle degree analysis identified per capita afforestation area (ecological response), per capita GDP (social state), and population density (ecological pressure) as persistent key constraints. Full article