《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (8): 2630-2639.DOI: 10.11772/j.issn.1001-9081.2025070921
• 多媒体计算与计算机仿真 • 上一篇
收稿日期:2025-08-11
修回日期:2025-09-26
接受日期:2025-09-29
发布日期:2025-11-05
出版日期:2026-08-10
通讯作者:
张健宇
作者简介:汤莉(1979—),女,河北衡水人,副教授,博士,CCF会员,主要研究方向:人工智能、数据挖掘基金资助:
Li TANG, Jianyu ZHANG(
), Rui YAO
Received:2025-08-11
Revised:2025-09-26
Accepted:2025-09-29
Online:2025-11-05
Published:2026-08-10
Contact:
Jianyu ZHANG
About author:TANG Li, born in 1979, Ph. D., associate professor. Her research interests include artificial intelligence, data mining.Supported by:摘要:
针对雾天等恶劣天气下多尺度目标难以检测,包括目标间重叠遮挡严重以及图像能见度低,导致误检和漏检频发等问题,提出一种基于改进YOLOv11的雾天目标检测算法MFA-YOLO (Multi-scale frequency-aware Feature Alignment 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,以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)应对因雾天图像对比度低所导致的边界框定位不准问题,以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)以通过动态卷积和任务分解等机制提升对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,所提算法的mAP@50和mAP@50:95分别达到76.5%和52.6%,与基础模型相比,分别提升了2.7和1.9个百分点,同时参数量和模型大小分别下降了15.3%和23.4%。此外,在Cityscapes数据集合成的两种不同雾浓度的雾天图像上的验证结果表明,所提算法的mAP@50分别为48.5%和46.7%,与YOLOv11s相比,分别提升了2.4和3.5个百分点,进一步证明了所提算法在多种雾天场景下的适应性和有效性。
中图分类号:
汤莉, 张健宇, 姚睿. 基于改进YOLOv11的雾天目标检测算法[J]. 计算机应用, 2026, 46(8): 2630-2639.
Li TANG, Jianyu ZHANG, Rui YAO. Object detection algorithm based on improved YOLOv11 in foggy weather[J]. Journal of Computer Applications, 2026, 46(8): 2630-2639.
| 参数 | 设定值 |
|---|---|
| 训练轮数(epoch) | 200 |
| 批量大小(batch size) | 16 |
| 进程数(workers) | 8 |
| 关闭Mosaic增强epoch数(close_mosaic) | 10 |
| 图片尺寸(imgsz) | 640×640 |
| 初始学习率(lr0) | 0.01 |
| 权重衰减系数(weight_decays) | 0.000 5 |
| 学习率动量(momentum) | 0.937 |
| 优化器(optimizer) | SGD |
| 自动混合精度(amp) | True |
表1 实验参数设计
Tab. 1 Experimental parameter design
| 参数 | 设定值 |
|---|---|
| 训练轮数(epoch) | 200 |
| 批量大小(batch size) | 16 |
| 进程数(workers) | 8 |
| 关闭Mosaic增强epoch数(close_mosaic) | 10 |
| 图片尺寸(imgsz) | 640×640 |
| 初始学习率(lr0) | 0.01 |
| 权重衰减系数(weight_decays) | 0.000 5 |
| 学习率动量(momentum) | 0.937 |
| 优化器(optimizer) | SGD |
| 自动混合精度(amp) | True |
| 实验序号 | 改进策略 | mAP@50/% | mAP@50:95/% | R/% | 参数量/106 | 浮点运算量/GFLOPs | 模型大小/MB | |||
|---|---|---|---|---|---|---|---|---|---|---|
| 基线 | M1 | M2 | M3 | |||||||
| 1 | √ | 73.8 | 50.7 | 66.3 | 9.4 | 21.3 | 18.3 | |||
| 2 | √ | √ | 74.8 | 50.3 | 64.5 | 9.7 | 22.7 | 19.5 | ||
| 3 | √ | √ | 74.2 | 50.3 | 68.1 | 7.6 | 16.9 | 15.7 | ||
| 4 | √ | √ | 75.0 | 51.7 | 66.6 | 8.7 | 30.0 | 17.7 | ||
| 5 | √ | √ | √ | 75.8 | 50.7 | 67.9 | 7.6 | 18.0 | 15.5 | |
| 6 | √ | √ | √ | 75.5 | 51.3 | 68.2 | 9.1 | 32.4 | 19.3 | |
| 7 | √ | √ | √ | 76.0 | 52.5 | 67.1 | 6.9 | 24.8 | 14.2 | |
| 8 | √ | √ | √ | √ | 76.5 | 52.6 | 68.5 | 7.2 | 27.0 | 15.5 |
表2 消融实验结果
Tab. 2 Ablation experiment results
| 实验序号 | 改进策略 | mAP@50/% | mAP@50:95/% | R/% | 参数量/106 | 浮点运算量/GFLOPs | 模型大小/MB | |||
|---|---|---|---|---|---|---|---|---|---|---|
| 基线 | M1 | M2 | M3 | |||||||
| 1 | √ | 73.8 | 50.7 | 66.3 | 9.4 | 21.3 | 18.3 | |||
| 2 | √ | √ | 74.8 | 50.3 | 64.5 | 9.7 | 22.7 | 19.5 | ||
| 3 | √ | √ | 74.2 | 50.3 | 68.1 | 7.6 | 16.9 | 15.7 | ||
| 4 | √ | √ | 75.0 | 51.7 | 66.6 | 8.7 | 30.0 | 17.7 | ||
| 5 | √ | √ | √ | 75.8 | 50.7 | 67.9 | 7.6 | 18.0 | 15.5 | |
| 6 | √ | √ | √ | 75.5 | 51.3 | 68.2 | 9.1 | 32.4 | 19.3 | |
| 7 | √ | √ | √ | 76.0 | 52.5 | 67.1 | 6.9 | 24.8 | 14.2 | |
| 8 | √ | √ | √ | √ | 76.5 | 52.6 | 68.5 | 7.2 | 27.0 | 15.5 |
| 实验序号 | 改进策略 | AP@50 | mAP@50 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 基线 | M1 | M2 | M3 | bicycle | bus | car | motorbike | person | ||
| 1 | √ | 64.1 | 65.4 | 88.5 | 69.0 | 81.7 | 73.8 | |||
| 2 | √ | √ | 65.1 | 66.6 | 88.6 | 71.1 | 82.7 | 74.8 | ||
| 3 | √ | √ | √ | 66.3 | 68.7 | 89.0 | 71.9 | 82.9 | 75.8 | |
| 4 | √ | √ | √ | √ | 66.6 | 67.6 | 89.2 | 75.9 | 83.1 | 76.5 |
表3 不同改进算法对各个类别的精度评估实验结果 (%)
Tab. 3 Precision evaluation experiment results of different improvement algorithms on various categories
| 实验序号 | 改进策略 | AP@50 | mAP@50 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 基线 | M1 | M2 | M3 | bicycle | bus | car | motorbike | person | ||
| 1 | √ | 64.1 | 65.4 | 88.5 | 69.0 | 81.7 | 73.8 | |||
| 2 | √ | √ | 65.1 | 66.6 | 88.6 | 71.1 | 82.7 | 74.8 | ||
| 3 | √ | √ | √ | 66.3 | 68.7 | 89.0 | 71.9 | 82.9 | 75.8 | |
| 4 | √ | √ | √ | √ | 66.6 | 67.6 | 89.2 | 75.9 | 83.1 | 76.5 |
| 算法 | mAP@50/% | mAP@50:95/% | 参数量/106 |
|---|---|---|---|
| Faster_rcnn_r101 | 60.6 | 32.9 | 60.1 |
| Retinanet_r101 | 67.6 | 41.9 | 55.2 |
| SSD | 72.4 | 45.2 | 25.0 |
| YOLOv3 | 72.0 | 40.4 | 61.6 |
| YOLOv4s | 72.4 | 44.7 | 9.1 |
| YOLOv5s | 70.3 | 43.3 | 7.2 |
| YOLOv5s-Fog[ | 73.6 | 45.7 | 26.9 |
| YOLOv6s | 70.4 | 44.0 | 17.2 |
| YOLOv7s-tiny | 71.6 | 42.6 | 6.0 |
| YOLOv8s | 73.7 | 50.1 | 11.1 |
| YOLOv10s | 71.8 | 48.9 | 8.0 |
| YOLOv11s | 73.8 | 50.7 | 9.4 |
| Gold-YOLOs[ | 73.0 | 48.5 | 21.5 |
| ASF-YOLOs[ | 74.2 | 47.6 | 7.3 |
| AOD+YOLOv11s | 73.6 | 49.2 | 9.4 |
| GridDehazeNet+YOLOv11s | 74.6 | 50.4 | 10.4 |
| IA-YOLO | 54.8 | 61.7 | |
| MFA-YOLO | 76.5 | 52.6 | 7.2 |
表4 RTTS数据集上不同算法的对比实验结果
Tab. 4 Comparative experiment results of different algorithms on RTTS dataset
| 算法 | mAP@50/% | mAP@50:95/% | 参数量/106 |
|---|---|---|---|
| Faster_rcnn_r101 | 60.6 | 32.9 | 60.1 |
| Retinanet_r101 | 67.6 | 41.9 | 55.2 |
| SSD | 72.4 | 45.2 | 25.0 |
| YOLOv3 | 72.0 | 40.4 | 61.6 |
| YOLOv4s | 72.4 | 44.7 | 9.1 |
| YOLOv5s | 70.3 | 43.3 | 7.2 |
| YOLOv5s-Fog[ | 73.6 | 45.7 | 26.9 |
| YOLOv6s | 70.4 | 44.0 | 17.2 |
| YOLOv7s-tiny | 71.6 | 42.6 | 6.0 |
| YOLOv8s | 73.7 | 50.1 | 11.1 |
| YOLOv10s | 71.8 | 48.9 | 8.0 |
| YOLOv11s | 73.8 | 50.7 | 9.4 |
| Gold-YOLOs[ | 73.0 | 48.5 | 21.5 |
| ASF-YOLOs[ | 74.2 | 47.6 | 7.3 |
| AOD+YOLOv11s | 73.6 | 49.2 | 9.4 |
| GridDehazeNet+YOLOv11s | 74.6 | 50.4 | 10.4 |
| IA-YOLO | 54.8 | 61.7 | |
| MFA-YOLO | 76.5 | 52.6 | 7.2 |
| 算法 | 浓度为0.01 | 浓度为0.02 | ||||
|---|---|---|---|---|---|---|
| mAP@50/% | mAP@50:95/% | 参数量/106 | mAP@50/% | mAP@50:95/% | 参数量/106 | |
| YOLOv5s | 46.8 | 27.0 | 7.2 | 43.8 | 25.3 | 7.2 |
| YOLOv8s | 46.0 | 27.0 | 11.1 | 44.4 | 26.0 | 11.1 |
| YOLOv10s | 45.2 | 25.8 | 8.0 | 44.0 | 25.2 | 8.0 |
| YOLOv11s | 46.1 | 26.9 | 9.4 | 43.2 | 25.3 | 9.4 |
| Gold-YOLOs | 43.2 | 24.3 | 21.5 | 41.4 | 23.2 | 21.5 |
| ASF-YOLOs | 42.3 | 22.7 | 7.3 | 39.3 | 21.3 | 7.3 |
| MFA-YOLO | 48.5 | 28.5 | 7.2 | 46.7 | 26.9 | 7.2 |
表5 浓度0.01和0.02的Foggy Cityscapes数据集上的泛化实验结果
Tab. 5 Generalization experiment results on Foggy Cityscapes dataset with concentration of 0.01 and 0.02
| 算法 | 浓度为0.01 | 浓度为0.02 | ||||
|---|---|---|---|---|---|---|
| mAP@50/% | mAP@50:95/% | 参数量/106 | mAP@50/% | mAP@50:95/% | 参数量/106 | |
| YOLOv5s | 46.8 | 27.0 | 7.2 | 43.8 | 25.3 | 7.2 |
| YOLOv8s | 46.0 | 27.0 | 11.1 | 44.4 | 26.0 | 11.1 |
| YOLOv10s | 45.2 | 25.8 | 8.0 | 44.0 | 25.2 | 8.0 |
| YOLOv11s | 46.1 | 26.9 | 9.4 | 43.2 | 25.3 | 9.4 |
| Gold-YOLOs | 43.2 | 24.3 | 21.5 | 41.4 | 23.2 | 21.5 |
| ASF-YOLOs | 42.3 | 22.7 | 7.3 | 39.3 | 21.3 | 7.3 |
| MFA-YOLO | 48.5 | 28.5 | 7.2 | 46.7 | 26.9 | 7.2 |
| [1] | 梁中豪,彭德巍,金彦旭,等. 基于交通场景区域增强的单幅图像去雾方法[J]. 计算机应用, 2018, 38(5): 1420-1426. |
| Liang Zhonghao, Peng Dewei, Jin Yanxu, et al. Single-image dehazing method based on traffic scene region enhancement[J]. Journal of Computer Applications, 2018, 38(5): 1420-1426. | |
| [2] | 郭璠,蔡自兴,谢斌,等. 图像去雾技术研究综述与展望[J]. 计算机应用, 2010, 30(9): 2417-2421. |
| Guo Fan, Cai Zixing, Xie Bin, et al. Review and prospect of image dehazing techniques[J]. Journal of Computer Applications, 2010, 30(9): 2417-2421. | |
| [3] | Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]// CVPR 2005. Piscataway: IEEE, 2005: 886-893. |
| [4] | Lowe D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110. |
| [5] | Hornyák O, Iantovics L B. AdaBoost algorithm could lead to weak results for data with certain characteristics[J]. Mathematics, 2023, 11(8): No.1801. |
| [6] | Valero-Carreras D, Alcaraz J, Landete M. Comparing two SVM models through different metrics based on the confusion matrix[J]. Computers and Operations Research, 2023, 152: No.106131. |
| [7] | Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// CVPR 2014. Piscataway: IEEE, 2014: 580-587. |
| [8] | Girshick R. Fast R-CNN[C]// ICCV 2015. Piscataway: IEEE, 2015: 1440-1448. |
| [9] | Ren S, He K, Girshick R, et al. Faster R-CNN: towards real time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. |
| [10] | Hussain M. YOLOv1 to v8: unveiling each variant: a comprehensive review of YOLO[J]. IEEE Access, 2024, 12: 42816-42833. |
| [11] | Liu W, Anguelov D, Erhan D, et al. SSD: single shot MultiBox detector[C]// ECCV 2016, LNCS 9905. Cham: Springer, 2016: 21-37. |
| [12] | Zhang L J, Fang J J, Liu Y X, et al. CR-YOLOv8: Multiscale object detection in traffic sign images[J]. IEEE Access, 2024, 12: 219-228. |
| [13] | Yue S, Zhang Z, Shi Y, et al. WGS-YOLO: a real-time object detector based on YOLO framework for autonomous driving[J]. Computer Vision and Image Understanding, 2024, 249: No.104200. |
| [14] | Niu Y, Zhang J. YOLOP-MVF: a multi-task autonomous driving perception detection method based on multi-scale feature weighted fusion[J]. IEEE Access, 2025, 13: 91374-91383. |
| [15] | Ren H, Jing F, Li S. DCW-YOLO: road object detection algorithms for autonomous driving[J]. IEEE Access, 2025, 13: 125676-125688. |
| [16] | Cheng S, Chen L, Yang K. DGSS-YOLOv8s: a real-time model for small and complex object detection in autonomous vehicles[J]. Algorithms, 2025, 18(6): No.358. |
| [17] | Ma H, Zhao W, Liu B, et al. Multi-scale target detection in autonomous driving scenarios based on YOLOv5-AFAM[J]. Applied Sciences, 2024, 14(11): No.4633. |
| [18] | Liu W, Ren G, Yu R, et al. Image-adaptive YOLO for object detection in adverse weather conditions[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36(2): 1792-1800. |
| [19] | Yuan Y, Wei Y, Guo Y, et al. CME-YOLO: a cross-modal enhanced YOLO algorithm for adverse weather object detection in autonomous driving[J]. Big Data and Cognitive Computing, 2025, 9(4): No.92. |
| [20] | Cai W, Chen Y, Qiu X, et al. LLD-YOLO: a low-light object detection algorithm based on dynamic weighted fusion of shallow and deep features[J]. IEEE Access, 2025, 13: 69967-69979. |
| [21] | Gao S, Zhang P, Yan T, et al. Multi-scale and detail-enhanced segment anything model for salient object detection[C]// MM 2024. New York: ACM, 2024: 9894-9903. |
| [22] | Chen L, Fu Y, Gu L, et al. Frequency-aware feature fusion for dense image prediction[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12): 10763-10780. |
| [23] | Li H, Li J, Wei H, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21(3): No.62. |
| [24] | Feng C, Zhong Y, Gao Y, et al. TOOD: task-aligned one-stage object detection[C]// ICCV 2021. Piscataway: IEEE, 2021: 3490-3499. |
| [25] | Li B, Ren W, Fu D, et al. Benchmarking single-image dehazing and beyond[J]. IEEE Transactions on Image Processing, 2018, 28(1): 492-505. |
| [26] | Sakaridis C, Dai D, van Gool L. Semantic foggy scene understanding with synthetic data[J]. International Journal of Computer Vision, 2018, 126(9): 973-992. |
| [27] | Tong J. AOD-Net: a lightweight real-time fruit detection algorithm for agricultural automation[J]. Journal of Food Measurement and Characterization, 2025, 19(4): 2818-2830. |
| [28] | Liu X, Ma Y, Shi Z, et al. GridDehazeNet: attention-based multi-scale network for image dehazing[C]// ICCV 2019. Piscataway: IEEE, 2019: 7313-7322. |
| [29] | Meng X, Liu Y, Fan L, et al. YOLOv5s-Fog: an improved model based on YOLOv5s for object detection in foggy weather scenarios[J]. Sensors, 2023, 23(11): No.5321. |
| [30] | Wang C, He W, Nie Y, et al. Gold-YOLO: efficient object detector via gather-and-distribute mechanism[C]// NeurIPS 2023. Red Hook: Curran Associates Inc., 2023: 51094-51112. |
| [31] | Kang M, Ting C M, Ting F F, et al. ASF-YOLO: a novel YOLO model with attentional scale sequence fusion for cell instance segmentation[J]. Image and Vision Computing, 2024, 147: No.105057. |
| [1] | 张纾豪, 何坤金, 徐佳晨, 沙河山, 陈正鸣. 融合透视校正与轻量注意力机制的轮毂缺陷检测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(6): 2007-2015. |
| [2] | 尹秋燕, 丁婧, 聂志刚. 无人机航拍视角下的人体姿态估计算法YOLO-AirPose[J]. 《计算机应用》唯一官方网站, 2026, 46(6): 1989-1997. |
| [3] | 杜艺, 续明进, 孔佳仪, 王力瑶, 赵晨. 基于YOLOv11的低秩自适应参数高效微调算法[J]. 《计算机应用》唯一官方网站, 2026, 46(6): 1738-1745. |
| [4] | 吴闵奇, 杨元华, 李航, 胡雅琴, 汤智豪, 梅腾. 基于图Transformer和RT-DETR的轻量化水下小目标检测[J]. 《计算机应用》唯一官方网站, 2026, 46(5): 1586-1595. |
| [5] | 张红瑞, 冯威铭, 杨潞霞, 马永杰. 基于YOLO11改进的水下小目标检测算法CSAF-YOLO[J]. 《计算机应用》唯一官方网站, 2026, 46(5): 1578-1585. |
| [6] | 张永兵, 闫丽蓉, 唐晓芬. 渐进式双阶段模态交互的单域泛化目标检测[J]. 《计算机应用》唯一官方网站, 2026, 46(4): 1264-1274. |
| [7] | 何帅, 邓春华. 基于YOLO-World的少样本学习目标检测算法[J]. 《计算机应用》唯一官方网站, 2026, 46(4): 1275-1282. |
| [8] | 陈鹏, 李旭, 余肖生. RGB-D双流镜像伪装目标检测网络[J]. 《计算机应用》唯一官方网站, 2026, 46(4): 1253-1263. |
| [9] | 郭阳, 王海亮, 高需, 王海涛, 王翌博. BEV三维目标检测算法体系综述[J]. 《计算机应用》唯一官方网站, 2026, 46(4): 1238-1252. |
| [10] | 于银山, 唐旭, 丁明鉴, 黄文凯, 毕嘉文, 谭国辰. 基于YOLOv10的实时车辆检测算法[J]. 《计算机应用》唯一官方网站, 2026, 46(3): 950-958. |
| [11] | 刘权捷, 顾兆一, 王春源. 复杂光照条件下的不安全驾驶行为检测[J]. 《计算机应用》唯一官方网站, 2026, 46(2): 613-619. |
| [12] | 樊跃波, 陈明轩, 汤显, 高永彬, 李文超. 基于多维频域特征融合的人物交互检测[J]. 《计算机应用》唯一官方网站, 2026, 46(2): 580-586. |
| [13] | 李明光, 陶重犇. 基于Mamba模型的分级跨模态融合三维目标检测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(2): 572-579. |
| [14] | 谢斌红, 王瑞, 张睿, 张英俊. 代理原型蒸馏的小样本目标检测算法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 233-241. |
| [15] | 李世伟, 周昱峰, 孙鹏飞, 刘伟松, 孟竹喧, 廉浩杰. 基于煤尘对激光雷达电磁波散射和吸收效应的点云数据增强方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 331-340. |
| 阅读次数 | ||||||
|
全文 |
|
|||||
|
摘要 |
|
|||||