.png){kind=icon}
.png){kind=icon}
Return
AEC-YOLO: An Adaptive Edge-Aware Calibration Framework for Robust Small Object Detection in Low-Altitude Remote Sensing Imagery
.png){kind=icon}
PRE
AI
Y
S
C
D
刘
Y
Y
DOI：10.1109/tgrs.2026.3717755.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Small object detection in uncrewed aerial vehicle (UAV) low-altitude remote sensing imagery faces core challenges, specifically edge information degradation and feature misalignment resulting from complex backgrounds and drastic scale variations. To address these issues, this study introduces AEC-YOLO. This framework aims to actively preserve and refine edge information. Based on the YOLO11s architecture, the proposed model achieves performance improvements through synergistic module optimization. First, a parallel edge refinement architecture (PERA) is integrated into the backbone network to actively extract and reinforce high-frequency edge details. Second, an adaptive edge calibration mechanism (AECM) is designed in the neck network. Specifically, a bidirectional attention calibration network (BAC-Net) facilitates the collaborative optimization of deep semantics and shallow edge features. In addition, a P2 detection head is introduced to enhance the response to small objects. Finally, an adaptive weight Focaler-MPDIoU loss function is proposed to balance the sample distribution and localization sensitivity. Experimental results on the VisDrone2019 dataset demonstrate that, compared with the baseline YOLO11s, AEC-YOLO improves mAP50 by 9.43 percentage points and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathrm{mAP}_{50:95}$ </tex-math></inline-formula> by 6.45 percentage points. Furthermore, to verify the generalization capability, the framework was extensively evaluated on the AI-TOD, DOTA-v1.5, and DIOR datasets, achieving mAP50 improvements of 4.9, 4.6, and 4.96 percentage points over the baseline, respectively. Comparative evaluations with state-of-the-art methods further demonstrate the superior performance and robustness of the proposed framework in complex low-altitude scenarios. The results indicate that the preservation of edge information and the multiscale feature collaborative calibration mechanism significantly elevate small object detection accuracy, providing a new technical path for UAV visual perception.
Keywords:
Adaptive loss function
aerial imagery
edge refinement
feature calibration
small object detection
Journal
IF:
8.6
Papers:
2.1W
Citations:
10.7W
Organization
Cited Papers