The widespread deployment of low-altitude drones in Internet of Things (IoT) applications, such as smart transportation and urban security, demands efficient, real-time visual perception on resource-constrained edge devices. A critical challenge in this domain is the precise detection and localization of small objects against complex backgrounds. Low-altitude UAV imagery is particularly challenging due to blurred small-target edges, weak anti-interference for medium and large targets, and fragmented multi-scale features. Existing convolutional and general-purpose Transformer-based detectors exhibit significant limitations in localization accuracy and robustness under these conditions. To address these issues, we propose E²-Former, a novel object detection model based on the Detection Transformer (DETR) framework, designed for real-time performance on computational-edge devices. Our model integrates three core components: an Edge-Enhanced Backbone that augments sensitivity to contours while preserving deep semantics; a Polarized Dynamic Multi-Feature Fusion (PDMF) Transformer that leverages dual-path polarized attention and frequency-domain modulation to enhance local-global feature modeling and suppress background noise; and an Edge-aware Path Aggregation Network (E-PAN) that uses bidirectional gating and multi-level context interaction to resolve feature fragmentation and promote fine-grained, cross-scale integration. On three challenging low-altitude datasets—VisDrone2019, UAVDT, and CODrone—E²- Former achieves leading AP₅₀ scores of 48.9%, 41.2%, and 33.0%, respectively, outperforming existing mainstream methods in both detection accuracy and robustness. Practical deployment and scene detection experiments further validate its impressive performance in real-world scenarios, establishing a new paradigm for building efficient and reliable low-altitude drone perception systems.

With the rapid advancement of deep learning technology, the problem of object detection in the field of remote sensing has received increasing attention. However, existing methods still face challenges when dealing with remote sensing image detection tasks, such as insufficient accuracy in small object recognition, an imbalance between recognition precision and detection efficiency, and difficulties in handling multiscale variations. Based on the above problems, this letter proposes a lightweight and efficient remote sensing image object detector MERS-Net. First, we propose an enhanced subchannel feature extraction (SCFE) module based on subchannel mixing to improve the feature extraction capability of the model under complex scale transformation. Second, we incorporated the WaveletPool module along with the GSConv and VoVGSCSP modules from Slimneck into the model to reduce computational parameters. Finally, we designed a lightweight detection head MRF-Detect based on the parameter sharing mechanism to improve the recognition capability of the model under limited hardware conditions. We verified the effectiveness of MERS-Net on the public datasets DOTA, AI-TOD, and DIOR. Compared with current mainstream algorithms such as YOLOv11, the average accuracy [mean average precision (mAP)] is improved by 2.6%, 2.5%, and 2.7%, respectively. At the same time, the parameter quantity and giga floating point operations (GFLOPs) are significantly reduced, showing its superior detection performance.

Small object detection in unmanned aerial vehicle (UAV) remote sensing images remains challenging due to large-scale variations, dense object distributions, and complex background interference. Although Transformer-based detectors have improved global context modeling for remote sensing object detection, many existing designs still rely on spatial geometric relationships and conventional cross-level fusion, which may limit the aggregation of spatially scattered small-object features and introduce background interference. To address these issues, this paper proposes a Multi-granularity Prompt and Intensity Guidance Detection Transformer (MPI-DETR), an efficient end-to-end Transformer-based detector for small-object detection in UAV remote sensing images. MPI-DETR consists of three key components: a Dual-Stream Ranked Self-Attention (DRSA) module for intensity-ordered global feature aggregation, a Bilateral Tanh Gating and Cosine Attention Feature Alignment Module (BTC-FAM) for noise-resistant cross-level alignment, and a Prompt-driven Multi-granularity Fusion (PMGF) module for enhancing weak small-object details. Experiments on AI-TOD, DIOR, and NWPU VHR-10 demonstrate that MPI-DETR achieves $A{P}_{50}$ scores of $43.8\%$, $87.5\%$, and $92.5\%$, respectively. Compared with the RT-DETR-R18 baseline, MPI-DETR improves $A{P}_{50}$ by 3.0, 1.0, and 3.9 percentage points on the three datasets, respectively, and increases $A{P}_{\mathrm{S}}$ by 2.9, 2.9, and 11.6 percentage points. It also surpasses the strongest compared models by 1.7, 1.0, and 1.3 percentage points in $A{P}_{50}$ on AI-TOD, DIOR, and NWPU VHR-10, respectively. These results indicate that MPI-DETR provides a robust and efficient solution for small-object perception in complex UAV remote sensing applications, especially in scenes with dense targets, background interference, and limited computational resources.

Recent advances in optical remote-sensing have expanded the use of unmanned aerial vehicles (UAV), yet accurately localising dense, small objects in low-resolution UAV imagery remains challenging. We propose RSRT-DETR, a lightweight DETR derivative tailored for UAV-borne remote-sensing that jointly enhances global context modelling and fine-grained detail preservation. RSRT-DETR couples a Polar Dynamic Spatial Fine-tuning (PDSF) Attention module with a lightweight backbone to amplify fine-grained geometry while suppressing background noise, and embeds a Multi-Scale Hypergraph Feature Network (MS-HFNet) that reasons over high-order spatial relations to disambiguate cluttered targets. These features are hierarchically fused by the HyperSemantic Fusion Network (HSFN), whose progressive edge-enhancement, resolution-adaptation, and multi-granularity aggregation schemes jointly preserve cross-scale context without sacrificing speed. Evaluated on three challenging benchmarks—DOTAv1.0, VisDrone2019, NWPU VHR-10 and DIOR, RSRT-DETR’s AP₅₀ reached 69.7%, 57.1%, 95.4% and 89.8%, respectively, surpassing the latest SOTA methods, and its AP_S also achieved the highest value. Meanwhile, the number of parameters and FLOPs are lower than those of similar DETR variants. The above results prove that RSRT-DETR achieves the optimal coordination of detection accuracy, model scale, and real-time performance in complex backgrounds and multi-scale dense small target scenes, providing a new paradigm for building efficient and reliable UAV remote sensing target detection systems. Our code is avaliable on: https://github.com/zhy1109/RSRT-DETR.

Pansharpening plays a critical role in remote sensing image processing by generating high-resolution multispectral (HRMS) images through the fusion of low-resolution multispectral (LRMS) and high-resolution panchromatic (PAN) images. In recent years, deep learning–based approaches have significantly advanced fusion performance via end-to-end modeling. However, existing methods still face notable limitations, including insufficient modeling of long-range dependencies, tight coupling between spectral and spatial information that may introduce spectral distortion and high computational complexity that restricts practical deployment in high-resolution scenarios. To address these issues, we propose Spectral-Spatial-Scale Mamba for Pansharpening (S³Mamba-Pan), a frequency-aware state space framework that jointly models spectral structure, spatial texture and multi-scale context in a coordinated yet decoupled manner. The method first extracts a global spectral anchor from the LRMS input to guide fusion, then performs wavelet-based frequency decoupling to separate low-frequency spectral components from high-frequency spatial details. A dual-stream Mamba architecture conducts parallel state space modeling for spectral correlation learning and PAN-guided texture enhancement and an adaptive distribution recalibration module aligns channel-wise statistics before reconstruction. Experiments on WV3, QuickBird and GF2 demonstrate consistent improvements over representative CNN, Transformer and SSM-based baselines under reduced-resolution evaluation. Under full-resolution no-reference assessment on WV3, S³Mamba-Pan achieves D_λ 0.0144, D_s 0.0303 and HQNR 0.960, indicating improved spectral fidelity and spatial consistency. Ablation and visualization analyses further corroborate the contribution of each component and provide interpretable evidence for the effectiveness of frequency decoupling and spectral anchoring. Code is available at https://github.com/FreeZS-a/S3Mamba.