Answers are grounded in this paper and cite their sources.
Foggy weather poses critical challenges to object detection systems due to atmospheric scattering effects: Mie scattering attenuates high-frequency edge information, forward scattering introduces low-frequency noise and contrast compression, and spatially non-uniform fog distribution complicates feature representation. Traditional two-stage dehazing-then-detection pipelines suffer from error accumulation and computational redundancy. This paper proposes a feature-aware detection framework built upon YOLOv11, directly integrating three enhancement modules targeting physical degradation mechanisms. The High-Low Frequency Decoupling (HLFD) module recovers atmospheric scattering-eroded features through frequency-domain decomposition and targeted enhancement. The Content-Adaptive Feature Modulation Fusion (CAFMFusion) module dynamically aligns shallow and deep features through dual-level adaptive gating based on local degradation levels. The Spatial Pyramid Pooling Fast with Cross Stage Partial Connection (SPPFCSPC) module efficiently expands global receptive fields via serial cascaded pooling. Systematic experiments on RTTS and Foggy Cityscapes datasets validate the method's effectiveness. On RTTS, the framework achieves 60.4% mAP and 83.3% mAP@.5, improving 5.4 and 6.6 percentage points over YOLOv11 baseline and outperforming 11 mainstream detectors, while maintaining 95.5 FPS inference speed. The method demonstrates robust performance across varying fog densities, providing reliable technical support for visual perception systems under adverse weather conditions.
No takes yet. Share an insight, caveat, or question.
Jiaxin Zhu (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: