Search Results (31)

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

Group equivariant convolutional neural networks (G-CNNs) achieve superior sample efficiency by encoding symmetry into network architecture, yet their computational overhead (up to 3.78× slower inference and 4.63× more multiply–accumulate operations) hinders deployment on resource-constrained edge devices. Existing pruning methods cannot be applied directly: arbitrarily removing weights breaks the group representation structure and degrades equivariance. We characterize the complete design space of equivariance-preserving compression, proving that exactly two axes leave a convolutional layer equivariant: irrep-bundle pruning, which reduces irreducible-representation multiplicities, and orbit-wise pruning, which removes complete spatial orbits from kernel supports; via Schur’s lemma, no third structure-preserving axis exists. This completeness result, rather than the use of representation theory itself, is our central contribution. We turn it into practice through direct sub-filter extraction, which yields real convolutional parameter reduction (up to 83%) and 1.4–2.9× measured inference speedup, unlike masking, which gives no real speedup. Across three datasets (MNIST, CIFAR-10, EuroSAT) and three symmetry groups (C₄, D₄, SO(2)), compression is nearly lossless on strongly symmetric data: the 4-layer EuroSAT model drops only 1.07% at 83% reduction. On weakly symmetric data (CIFAR-10), the pruned model can even gain 2.6 points, but our analysis attributes this to relaxing a mismatched equivariance constraint rather than to pruning itself; the value of pruning over from-scratch training scales with the data’s symmetry strength. Full article

Remote sensing provides vast unlabeled imagery at low cost, yet annotation remains expensive, making self-supervised learning (SSL) well suited to this domain. However, existing DINO-style SSL frameworks are not well suited to remote sensing imagery, where object extents vary substantially and standard multi-crop view generation often introduces cross-scale inconsistency. This issue is particularly severe for small objects and elongated structures, whose discriminative features can be lost under scale transformations. To address this limitation, we propose DINO-MS (DINO with multi-scale consistency), a scale-consistent SSL framework for remote sensing imagery. The key idea is to construct feature-aligned cross-scale local views and explicitly enforce prediction-level agreement among them. Specifically, DINO-MS first adopts a co-located multi-scale cropping strategy to sample local views from the same spatial location at different crop scales, and then introduces a local consistency loss that works jointly with the original DINO local-to-global objective. Extensive experiments on land-use classification and change detection benchmarks show that DINO-MS generally improves downstream transfer performance. Notably, on EuroSAT, it improves per-class accuracy from 80.60% to 87.80% for Highway and from 88.00% to 91.60% for River with DINO-MC, confirming its advantage for categories dominated by small objects. Full article

To address statistical heterogeneity and update-level privacy risks in federated learning for geospatial data, this paper proposes a hierarchically decoupled collaborative framework that integrates client-side privacy perturbation with server-side consistency-aware aggregation, while incorporating governance as a system-level support module. Under strong non-IID conditions, the proposed soft-weight aggregation strategy mitigates update mismatch and improves convergence stability without hard filtering legitimate but distributionally shifted client contributions. Meanwhile, the risk-aware perturbation mechanism adaptively adjusts clipping and noise strength across clients to better balance privacy protection and model utility. An on-chain governance and off-chain training coordination mechanism is further introduced to support auditable and traceable collaboration without interfering with the main optimization process. Experimental results on EuroSAT_RGB with ResNet-18 show that the proposed design achieves more stable training and better overall performance than the compared baselines, especially under severe heterogeneity. These findings highlight the value of jointly considering privacy-aware perturbation and consistency-aware aggregation for improving training stability and preserving utility in geospatial federated learning under statistically heterogeneous settings. Full article

Cross-domain few-shot learning (CD-FSL) remains challenging in medical imaging, where labeled data are scarce and source–target domain gaps are often large due to modality differences. In particular, existing few-shot learning methods rely on source–target domain similarity, which limits their effectiveness in cross-modality settings such as MRI-to-CT transfer. To address this problem, this paper proposes an adapter-based Vision Transformer framework for cross-domain few-shot brain tumor classification. Lightweight adapter modules are inserted into a pretrained Vision Transformer to enable parameter-efficient domain adaptation without fine-tuning the entire backbone. In addition, a Prototypical Network is employed to construct class prototypes from limited labeled samples, while a prototype-level Maximum Mean Discrepancy (MMD) loss is introduced to align feature distributions across domains. Unlike prior approaches, the proposed framework introduces a unified prototype-level alignment strategy within an episodic learning paradigm, enabling direct class-wise cross-modal alignment. This design improves generalization under large modality gaps and limited labeled data by jointly optimizing representation learning and domain adaptation. The proposed framework is evaluated on MRI-to-CT brain tumor classification as well as several heterogeneous cross-domain benchmarks, including Chest X-ray, ISIC, CropDisease, and EuroSAT. Experimental results demonstrate that the proposed method achieves competitive performance compared to existing few-shot learning baselines, showing strong robustness under significant domain shifts. Full article

Land cover classification (LCC) is a fundamental task in remote sensing, which enables effective environmental monitoring, agricultural planning, and disaster management. The existing approaches often rely on fine-tuning pre-trained models, which are not specifically designed for LCC, which lead to suboptimal performance in complex scenarios. To address these limitations, we propose the Cross-Module Attention Land Cover Network (CALCNet), a novel architecture developed from scratch. CALCNet follows a contracting and restoration backbone, where the contracting path extracts progressively abstract semantic features while reducing spatial resolution, and the restoration path recovers fine-grained spatial details through upsampling and skip connections. In addition, CALCNet integrates a cross-module attention mechanism that combines spatial attention and multi-scale feature selection to enhance feature representation. Furthermore, we applied a differential evolution-based neuron pruning strategy to create a compressed CALCNet variant, which retains high classification performance while reducing computational cost. The CALCNet is evaluated on four benchmark LCC datasets, AID, UCMerced_LandUse, NWPU_RESISC45, and EuroSAT, demonstrating strong performance across all benchmarks. Specifically, the model achieves classification accuracies of 98.09%, 99.47%, 99.19%, and 99.19%, respectively. The compressed CALCNet variant reduces computational cost to 78.55 million floating point operations (FLOPs) with a model size of 43 MB, while achieving improved inference speeds (38.32 frames/sec on CPU and 118.3 frames/sec on GPU), representing approximately 45–50% reduction in FLOPs and model storage. These results highlight that CALCNet is both highly accurate and computationally efficient, making it well suited for real-world LCC applications. Full article

Accurate remote sensing scene classification is essential for large-scale Earth observation but remains challenging due to significant inter-class similarity and complex spatial layouts in medium- and low-resolution imagery. Conventional convolutional neural networks (CNNs) effectively capture local structural patterns but struggle to model long-range semantic dependencies, whereas Vision Transformers excel at global context modeling yet often show reduced sensitivity to fine-grained spatial structures. To address these limitations, we propose CrtNet, a structure-aware Cross-Model Residual Transformer Network that establishes a dual-stream collaborative architecture integrating convolutional structural representations with Transformer-based semantic modeling through gated residual cross-model interactions. In this framework, a convolutional branch first extracts stable local structural features with strong spatial inductive biases. These features are continuously injected into the Transformer encoding process via residual cross-model connections, enabling persistent structural guidance during global attention modeling. In addition, a sample-adaptive dynamic gating mechanism is introduced to flexibly balance structural and semantic features during prediction. Extensive experiments conducted on two public remote sensing benchmarks, EuroSAT and UCM, demonstrate that CrtNet consistently outperforms representative CNN-based, Transformer-based, and hybrid state-of-the-art models, particularly in visually ambiguous scene categories. Full article

This paper presents a lightweight hyper-network architecture for cross-scale multi-task object classification, addressing the critical challenge of gradient interference in joint learning scenarios. We propose a HyperConv module integrated into a slim ResNet-12 backbone, which dynamically generates task-adaptive 3 × 3 convolutional kernels from compact two-dimensional latent vectors. This design allows explicit control over gradient flows for different tasks with minimal parameter overhead (only 3.2% additional parameters). Our framework incorporates adversarial regularization via a Gradient Reversal Layer (GRL) and dynamic task-weight scheduling to mitigate gradient conflicts across domains. Experiments on both natural image datasets (Mini-ImageNet and CIFAR-100) and remote sensing benchmarks (EuroSat and UCMerced_LandUse) demonstrate statistically significant improvements over conventional shared-parameter baselines. The proposed method effectively reduces negative transfer, enhances feature representation, and offers a practical solution for on-device multi-task learning in resource-constrained remote sensing applications such as UAVs and edge satellites. Full article

This systematic review investigates recent advances and persistent challenges in Land Use and Land Cover (LULC) classification using Sentinel-2 imagery, emphasizing the gap between benchmark results and operational performance. Following PRISMA guidelines, we analyzed 89 peer-reviewed studies published between 2020–2025 to address the discrepancy between academic benchmarks and real-world deployment. While benchmark datasets such as EuroSAT routinely achieve accuracies above 98%, operational systems deployed at regional or global scales typically reach only 75–85%. Through systematic analysis and meta-analysis of reported results, we identify three main factors: (i) methodological issues, particularly the inflation of reported accuracies caused by spatial autocorrelation; (ii) domain adaptation limitations, where geographic and temporal transferability reduce accuracy by 15–25%; (iii) training data constraints, where geographic diversity proves more important than sample size. Multi-spectral approaches provide modest 5–8% gains over RGB at significantly higher computational costs. Foundation models (e.g., Prithvi, Sky Sense) and self-supervised learning show promise for reducing data requirements while maintaining performance. Comparisons with operational products such as ESA WorldCover and Google Dynamic World confirm the more modest performance achievable under real-world conditions. The findings emphasize the need for rigorous spatial validation protocols, standardized evaluation frameworks, and closer integration between research and operational development. Full article

Ensuring global food security requires accurate and robust solutions for crop health monitoring, weed detection, and large-scale land-cover classification. To this end, we propose AgroVisionNet, a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) for local feature extraction, Vision Transformers (ViTs) for capturing long-range global dependencies, and Graph Neural Networks (GNNs) for modeling spatial relationships between image regions. The framework was evaluated on five diverse benchmark datasets—PlantVillage (leaf-level disease detection), Agriculture-Vision (field-scale anomaly segmentation), BigEarthNet (satellite-based land-cover classification), UAV Crop and Weed (weed segmentation), and EuroSAT (multi-class land-cover recognition). Across these datasets, AgroVisionNet consistently outperformed strong baselines including ResNet-50, EfficientNet-B0, ViT, and Mask R-CNN. For example, it achieved 97.8% accuracy and 95.6% IoU on PlantVillage, 94.5% accuracy on Agriculture-Vision, 92.3% accuracy on BigEarthNet, 91.5% accuracy on UAV Crop and Weed, and 96.4% accuracy on EuroSAT. These results demonstrate the framework’s robustness across tasks ranging from fine-grained disease detection to large-scale anomaly mapping. The proposed hybrid approach addresses persistent challenges in agricultural imaging, including class imbalance, image quality variability, and the need for multi-scale feature integration. By combining complementary architectural strengths, AgroVisionNet establishes a new benchmark for deep learning applications in precision agriculture. Full article

By using Geographic Information Systems, satellite imagery from remote sensing techniques provides quantitative and qualitative data about Earth’s natural and human elements. However, the direct use of raw imagery may prevent the accurate identification of the spectral and temporal characteristics of the target objects. To obtain meaningful results from these data, the object and surface features in the image must be classified correctly. In this context, this study develops a new deep learning approach that includes hyperparameter optimization that considers uncertainty factors when classifying satellite imagery. In the proposed approach, a hybrid architecture called CNN-Pythagorean Fuzzy Deep Neural Network (PFDNN) is developed by combining the ability of convolutional neural networks (CNN) to reveal expressive features with the ability of Pythagorean fuzzy set (PFS) theory to predict uncertainty. In addition, to further improve the model’s success, the hyperparameters are automatically optimized using Optuna. In the experiments conducted on the EuroSAT RGB dataset, the CNN+PFDNN+Optuna model achieved 0.9696 ± 0.0037 accuracy and a macro-AUC value of 0.9983, outperforming other methods such as DNN, FDNN, PFDNN and VGG16+PFDNN. Including the Pythagorean fuzzy layer in the system provided about 13.05% higher accuracy than conventional fuzzy systems. These results show that integrating the Pythagorean fuzzy set approach into deep learning models contributes to more effective management of uncertainties in remote sensing data and that hyperparameter optimization significantly impacts model performance. Full article

Differential privacy (DP) has become a cornerstone for privacy-preserving machine learning, yet its application to high-resolution satellite imagery remains underexplored. Existing DP algorithms, such as DP-SGD, often rely on static noise levels and global clipping thresholds, which lead to slow convergence and poor utility in deep neural networks. In this paper, we propose ADP-SIR, an Adaptive Differential Privacy framework for Satellite Image Recognition with provable convergence guarantees. ADP-SIR introduces two novel components: Convergence-Guided Noise Scaling (CGNS), which dynamically adjusts the noise multiplier based on training stability, and Layerwise Sensitivity Profiling (LSP), which enables fine-grained clipping at the layer level. We provide theoretical analysis showing that ADP-SIR achieves good convergence in non-convex settings under Rényi differential privacy. Empirically, we evaluate ADP-SIR on EuroSAT and RESISC45, demonstrating significant improvements over DP-SGD and AdaClip-DP in terms of accuracy, convergence speed, and per-class fairness. Our framework bridges the gap between practical performance and rigorous privacy for remote sensing applications. Full article

Unmanned platforms such as unmanned aerial vehicles, unmanned ground vehicles, and autonomous underwater vehicles often face challenges of data, device, and model heterogeneity when performing collaborative data processing tasks. Existing research does not simultaneously address issues from these three aspects. To address this issue, this study designs an unmanned platform cluster architecture inspired by the cloud-edge-end model. This architecture integrates federated learning for privacy protection, leverages the advantages of distributed model training, and utilizes edge computing’s near-source data processing capabilities. Additionally, this paper proposes a federated edge intelligence method (DSIA-FEI), which comprises two key components. Based on traditional federated learning, a data sharing mechanism is introduced, in which data is extracted from edge-side platforms and placed into a data sharing platform to form a public dataset. At the beginning of model training, random sampling is conducted from the public dataset and distributed to each unmanned platform, so as to mitigate the impact of data distribution heterogeneity and class imbalance during collaborative data processing in unmanned platforms. Moreover, an intelligent model aggregation strategy based on similarity measurement and loss gradient is developed. This strategy maps heterogeneous model parameters to a unified space via hierarchical parameter alignment, and evaluates the similarity between local and global models of edge devices in real-time, along with the loss gradient, to select the optimal model for global aggregation, reducing the influence of device and model heterogeneity on cooperative learning of unmanned platform swarms. This study carried out extensive validation on multiple datasets, and the experimental results showed that the accuracy of the DSIA-FEI proposed in this paper reaches 0.91, 0.91, 0.88, and 0.87 on the FEMNIST, FEAIR, EuroSAT, and RSSCN7 datasets, respectively, which is more than 10% higher than the baseline method. In addition, the number of communication rounds is reduced by more than 40%, which is better than the existing mainstream methods, and the effectiveness of the proposed method is verified. Full article

Multispectral images contain richer spectral signatures than easily available RGB images, for which they are promising to contribute to information perception. However, the relatively high cost of multispectral sensors and lower spatial resolution limit the widespread application of multispectral data, and existing reconstruction algorithms suffer from a lack of diverse training datasets and insufficient reconstruction accuracy. In response to these issues, this paper proposes a novel and robust multispectral reconstruction network from low-cost natural color RGB images based on free available satellite images with various land cover types. First, to supplement paired natural color RGB and multispectral images, the Houston hyperspectral dataset was used to train a convolutional neural network Model-TN for generating natural color RGB images from true color images combining CIE standard colorimetric system theory. Then, the EuroSAT multispectral satellite images for eight land cover types were selected to produce natural RGB using Model-TN as training image pairs, which were input into a residual network integrating channel attention mechanisms to train the multispectral images reconstruction model, Model-NM. Finally, the feasibility of the reconstructed multispectral images is verified through image classification and target detection. There is a small mean relative absolute error value of 0.0081 for generating natural color RGB images, which is 0.0397 for reconstructing multispectral images. Compared to RGB images, the accuracies of classification and detection using reconstructed multispectral images have improved by 16.67% and 3.09%, respectively. This study further reveals the potential of multispectral image reconstruction from natural color RGB images and its effectiveness in target detection, which promotes low-cost visual perception of intelligent unmanned systems. Full article

The classification of multiband images captured by advanced sensors, such as satellite-mounted imaging systems, is a critical task in remote sensing and environmental monitoring. These sensors provide high-dimensional data that encapsulate a wealth of spectral and spatial information, enabling detailed analyses of the Earth’s surface features. However, the complexity of these data poses significant challenges for accurate and efficient classification. Our study describes and highlights methods for creating ensembles of neural networks for handling multiband images. Two applications are illustrated in this work: (1) satellite image classification tested on the EuroSAT and LCZ42 datasets and (2) a species-level identification of planktic foraminifera. Multichannel images are fed into an ensemble of Convolutional Neural Networks (CNNs) (ResNet50, MobileNetV2, and DenseNet201), where each network is trained using three channels obtained from the multichannel images, and two custom networks (one based on ResNet50 and the other one based on attention) where the input is a multiband image. The ensemble learning framework harnesses these variations to improve classification accuracy, surpassing other advanced methods. The proposed system, implemented in MATLAB 2024b and PyTorch 2.6, is shown to achieve higher classification accuracy than those of human experts for species-level identification of planktic foraminifera (>92% vs. 83%) and state-of-the-art performance on the tested planktic foraminifera, the EuroSAT and LCZ42 datasets. Full article