Abstract: To address challenges of detecting multi-scale objects under adverse weather conditions such as fog, including severe occlusion, low image visibility, and frequent false positives and missed detections, a foggy weather object detection algorithm based on YOLOv11, named MFA-YOLO (Multi-Scale Frequency-Aware Feature Alignment YOLO), was proposed. Firstly, the C3k2_Multi-Scale Edge Feature Enhancement (C3k2_MSEFE) module was designed to replace original C3k2 module in YOLOv11, enhanced network’s ability to extract features from multi-scale objects. Secondly, the Lightweight Frequency-Aware Feature Pyramid Network (LFFPN) was designed to address the issue of inaccurate bounding box localization caused by low contrast in foggy images, thereby improved both accuracy and robustness of detection. Third, the Dynamic Feature Alignment Detection Head (DFADHead) was introduced to strengthen model’s capability to detect overlapping and occluded objects through mechanisms such as dynamic convolution and task decomposition, and improved detection performance in dense scenes. Experimental results demonstrate that on the RTTS dataset, the proposed algorithm achieves a mAP@50 of 76.5% and a mAP@50:95 of 52.6%, representing improvements of 2.7 and 1.9 percentage points over the baseline model, respectively. Meanwhile, the number of parameters and model size are reduced by 23.4% and 15.3%, respectively. Furthermore, evaluations on two foggy datasets with different fog densities synthesized from Cityscapes dataset show that the proposed algorithm achieves mAP@50 scores of 48.5% and 46.7% which are 2.4 and 3.5 percentage points higher than YOLOv11s, respectively. These results further validate adaptability and effectiveness of the proposed algorithm across various foggy weather scenarios.

摘要： 针对雾天等恶劣天气下多尺度目标难以检测、目标间重叠遮挡严重，以及图像能见度低导致误检、漏检频发等问题，提出一种基于YOLOv11的雾天目标检测算法(MFA-YOLO)。首先，设计多尺度边缘特征增强模块C3k2_MSEFE (C3k2_Multi-Scale Edge Feature Enhancement)替换YOLOv11网络中的C3k2模块，增强网络对多尺度目标的特征提取能力；其次，设计轻量化频率感知特征金字塔网络LFFPN(Lightweight Frequency-Aware Feature Pyramid Network)，以应对因雾天图像对比度低所导致的边界框定位不准问题，提升检测的准确性和鲁棒性；最后，设计动态特征对齐检测头DFADHead(Dynamic Feature Alignment Detection Head)，通过动态卷积和任务分解等机制，提升密集场景下重叠遮挡目标的识别能力。实验结果表明，在RTTS数据集上，改进算法的mAP@50和mAP@50:95分别达到76.5%和52.6%，与基础模型相比，分别提升2.7和1.9个百分点，同时参数量和模型大小分别下降了23.4%和15.3%。此外，在Cityscapes数据集合成的两种不同雾浓度的雾天图像数据集上进行验证，改进算法的mAP@50分别为48.5%和46.7%，与YOLOv11s相比，分别提升2.4和3.5个百分点，进一步证明了改进算法在多种雾天场景下的适应性和有效性。