UAV small object detection improved by hybrid stage-wise deep network
HDMamba-YOLO: Efficient State-Space Perception and Local Spatial Reconstruction for UAV Small Object
Computer Vision and Pattern Recognition
Summary
Detecting tiny objects in images taken by drones is hard because the objects are small, blurry, and often crowded with confusing backgrounds. The authors created a special multi-stage deep-learning system called HDMamba-YOLO that looks at both the big picture and small details at different steps. This approach helps the system better recognize where objects are and what they are, even when the images are tricky. Their system showed good accuracy and efficiency on benchmark drone image datasets.
What this means in practice
- •For drone developers: Improve drone vision systems by integrating this efficient multi-stage network for better detection of small objects in complex aerial images.
- •For surveillance teams: Enhance automated monitoring systems to detect small or distant objects from UAV footage in cluttered environments more accurately.
Authors
Linduo Wei, Junjie Fan, Yijun Mai, Yong Qi
Abstract
Small-object detection in UAV imagery is challenged by weak visual evidence, ambiguous boundaries, dense object distributions, and complex backgrounds. Effective detection therefore requires long-range contextual information for target-background discrimination while preserving explicit local two-dimensional structures for accurate localization. These requirements arise at different stages of the detection pipeline and are not naturally addressed by a uniform feature-processing strategy. We propose Hybrid Dual-domain Mamba-YOLO (HDMamba-YOLO), a stage-wise heterogeneous SSM-CNN detector organized according to a perception-reconstruction-alignment-interaction rationale. EfficientVMamba-based EVSS establishes long-range contextual perception in the backbone, while PhasePatchMerging2D provides phase-aware hierarchical transitions. DST-Wrapper and Native C3k2-ASSAF then perform perception-to-reconstruction transition and repeated local two-dimensional reconstruction during FPN/PAN aggregation. DySample provides content-adaptive cross-scale resampling, while OS-CVTIA introduces macro-micro interaction and task-specific modulation for localization and classification. On VisDrone2019, HDMamba-YOLO-B achieves 42.737% mAP50 and 25.713% mAP50:95 with 10.042M parameters and 29.879 corrected GFLOPs. HDMamba-YOLO-Lite achieves 41.140% mAP50 and 24.741% mAP50:95 with 5.344M parameters. Under the unified AI-TOD evaluation protocol, HDMamba-YOLO-B obtains 21.621% AP and 47.881% AP50. Controlled ablations further support the stage-wise allocation of state-space perception, convolutional reconstruction, dynamic alignment, and task interaction for UAV small-object detection.