A detection-oriented cross-scale fusion YOLOv11 for UAV object detection in hazy environments

UAV object detection under hazy conditions is challenging because atmospheric scattering reduces image contrast, weakens object boundaries, and suppresses fine-grained details that are important for small-object recognition. To address these issues, this paper proposes GRFW-YOLOv11, a YOLOv11-based detection framework with a moderate model size for hazy UAV scenes. The proposed framework incorporates detection-oriented haze-interference suppression, frequency–spatial feature recalibration, content-aware cross-scale feature fusion, and localization-oriented regression optimization. These components are designed to improve the representation and localization of small or degraded targets under haze-related visual degradation. Experiments are conducted on VisDrone2019, Foggy VisDrone2019, RTTS, and HazyDet. On VisDrone2019, GRFW-YOLOv11n achieves 41.3% mAP₅₀ and 24.8% mAP50:95 with 4.09 M parameters, improving YOLOv11n by 9.1 and 6.2 percentage points, respectively. On HazyDet, GRFW-YOLOv11n improves mAP₅₀ from 65.8% to 72.3% and mAP50:95 from 46.0% to 53.3%. The results indicate that the proposed design improves detection accuracy under hazy UAV scenarios, while the increased computational cost suggests that further efficiency optimization is still needed for resource-constrained deployment.