TL;DR
Compared to existing state-of-the-art methods, LightS-DETR delivers superior detection accuracy and robust real-time performance while significantly reducing computational demands, offering an optimal balance for UAV deployment.
Small-object detection in unmanned aerial vehicle (UAV) imagery is challenging because targets are often tiny, densely distributed, and embedded in cluttered backgrounds, while edge platforms impose strict computational constraints. This paper presents a lightweight end-to-end detector based on RT-DETR to improve detection accuracy, localisation quality, and deployment efficiency. The proposed method includes three components: a detail-enhanced backbone for preserving fine-grained texture and edge cues, rotary positional encoding combined with a content-aware bidirectional feature pyramid for improved cross-scale alignment, and a difficulty-aware composite regression loss for more stable bounding-box optimisation. Experiments on VisDrone2019 and DOTA show that the proposed method improves mAP50 by 3.5% and 1.6%, respectively, over the RT-DETR baseline, while reducing the number of parameters by 28.6%. Ablation studies confirm the complementary contributions of the three components, and robustness experiments demonstrate reliable performance under challenging aerial conditions. Deployment on a Jetson Orin NX using a self-built UAV dataset further validates the practicality of the method for edge-side UAV monitoring.
Detecting small infrared targets from Uncrewed Aerial Vehicles (UAVs) at night presents a significant challenge, requiring a delicate balance between high accuracy and low computational cost for resource-constrained edge deployment. To address this, we introduce a lightweight, high-precision framework extending the YOLOv11 architecture. First, a DP-PSA block, integrating Progressive Channel-wise Self-Attention (PCSA) and Dynamic Tanh (D-Tanh), is proposed to amplify small-object features within the C2PSA module. Second, a Bottleneck-FPD module, combining Frequency Dynamic Convolution (FDConv) with Partial Convolution (PConv), is designed to drastically cut computational redundancy while preserving feature richness. Finally, coupled with the Lightweight Single Channel Detection (LSCD) head, our model surpasses the YOLOv11 baseline. Experimental results demonstrate that it achieves an mAP@[.50:.95] of 52.0% and an inference speed of 382 FPS, while simultaneously reducing GFLOPs by 27% and parameters by 23%. This work provides a practical and efficient solution for real-time aerial surveillance at night.
Unmanned aerial vehicle (UAV) imagery supports intelligent surveillance, environmental monitoring, traffic management, and infrastructure inspection. Yet aerial detection is difficult when objects are small, crowded, and observed at markedly different scales. Background clutter and illumination changes further weaken target cues and impair localization. We therefore propose HD-YOLO, a lightweight multiscale detector for small objects in UAV imagery. Its Multi-Dilation Shared Convolution Kernel (DSCK) extracts local texture and contextual information with shared dilated kernels. The Hybrid Dilated Bidirectional Feature Pyramid Network (HDFPN) reconstructs global and local cues before bidirectional aggregation, enabling high-resolution evidence to reach the prediction layers. The Efficient and Slim Head (ES-Head) combines shared operations with differential convolution to reduce cost and strengthen boundary-sensitive features. A joint ShapeIoU and Normalized Wasserstein Distance loss improves regression for small, irregular objects. Together, these components reduce missed detections in dense, cluttered scenes without relying on large model capacity. On VisDrone2019, HD-YOLO improves precision, recall, mAP50, and mAP50:95 over YOLOv8n by 6.9%, 7.2%, 8.2%, and 5.2%, respectively, while reducing parameters from 3.0 M to 0.9 M. Evaluations on TinyPerson and HIT-UAV also support its utility for tiny pedestrians and infrared aerial targets. HD-YOLO therefore improves small-object detection with a compact parameter footprint, while direct hardware benchmarks remain necessary to establish deployment efficiency.
To tackle the challenge of detecting small targets in UAV imagery, this paper proposes PF-DETR, an enhanced object detection model based on RT-DETR, designed to improve detection accuracy in complex scenes. The improvements are primarily reflected in the following aspects. First, a P2 detection head is added to extend the feature pyramid to finer scales, thereby enhancing the ability of shallow features to detect small targets. Second, a Pyramidal Hierarchical Frequency-Domain Fusion (PHF) module is introduced. By combining wavelet pooling with high- and low-frequency attention, the module effectively extracts and fuses multi-scale features, reduces information loss, and improves detection accuracy for small targets. Finally, the backbone network is restructured through the design of a lightweight BasicBlock_FasterNet_Rep module, which integrates FasterNet and RepVGG-style re-parameterization. This restructuring significantly reduces model complexity and parameters while strengthening multi-scale feature extraction. Experimental results on the VisDrone2019 dataset show that the improved PF-DETR achieves a notable performance boost: compared to the original model, mAP@0.5 increases by 5.4%, while the number of parameters is reduced by 25.0%. The computational cost increases by about 35.0%, but this comes with higher accuracy, resulting in a favorable balance between detection performance and model efficiency. Overall, these improvements enhance the model’s robustness and accuracy in detecting multi-scale and small targets in complex and cluttered scenes.
The detection of small objects in aerial traffic scenes is vital for advancing intelligent transportation systems (ITS), yet traditional methods often suffer from scale variation, background complexity, and insufficient context utilization. To address these challenges, we propose SOD-MSC, a YOLOv8-based small object detector designed for remote sensing images. SOD-MSC integrates a global multi-scale (GMS) module to enhance global perception and multi-scale fusion, a lightweight fusion feature pyramid network (LFFPN) to strengthen spatial correlations and suppress background noise, and an adaptive RT-DETR-based detection head with Inner_SIoU loss to improve localization accuracy and convergence. Experiments on VisDrone2019-DET and HazyDet demonstrate that SOD-MSC achieves 39.9% and 49% on mAP@0.5, with only 5.22 M parameters and 11.2 GFLOPs, surpassing several baselines and state-of-the-art models in balancing accuracy and efficiency. These results highlight the effectiveness of SOD-MSC for dense small object detection in complex traffic environments, with promising applications in congestion monitoring, vehicle counting, and related ITS tasks.