基于卷积神经网络的雾天道路能见度测量方法

doi:10.12305/j.issn.1001-506X.2026.04.02

摘要/Abstract

摘要：

为改善雾天场景下的道路能见度，提高道路安全性和行车效率，通过构建基于道路监控图像的能见度检测及恢复方法，提出基于大气散射模型和卷积神经网络的能见度检测方法，实现对雾浓度的检测，进而对浓雾图像进行去雾处理，以达到对雾天道路监控图像清晰化增强的目的。通过自动特征提取和注意力机制模块，建立雾天光学图像特征与能见度之间的关系模型，改善能见度检测方法在复杂的高速道路场景下适用性差的问题。采用透射率特征像素与景深距离像素特征提取的策略，优化图像透射率与目标景深估计以提高道路能见度检测的精度。利用包含雾图的合成数据集进行训练，并采用高速监控拍摄的雾天实测图像进行测试，实现了对不同情况下雾浓度检测和评估，可用于智慧交通系统中的高速道路预警和提示。

关键词: 能见度检测, 卷积神经网络, 光学图像增强, 大气散射模型

Abstract:

In order to improve road visibility in foggy weather, enhance road safety and driving efficiency. This article proposes a visibility detection method based on atmospheric scattering model and convolutional neural network by constructing a visibility detection and restoration method based on road monitoring images, to achieve the detection of fog concentration. Furthermore, the foggy images are processed to remove fog and enhance the clarity of road monitoring images in foggy weather. By using automatic feature extraction and attention mechanism modules, a relationship model between foggy optical image features and visibility is established to improve the poor applicability of visibility detection methods in complex highway scenes; The strategy of using transmittance feature pixels and depth of field distance pixel feature extraction is adopted to optimize the estimation of image transmittance and target depth of field to improve the accuracy of road visibility detection. By using a synthetic dataset containing fog images for training and testing with actual foggy images captured by high-speed monitoring, fog concentration detection and evaluation under different conditions have been achieved, which can be used for highway warning and prompting in intelligent transportation systems.

Key words: visibility detection, convolutional neural network (CNN), optical image enhancement, atmospheric scattering model

中图分类号:

TP 391.41

孙景荣, 张华, 陈哲哲, 赵方正. 基于卷积神经网络的雾天道路能见度测量方法[J]. 系统工程与电子技术, 2026, 48(4): 1112-1124.

Jingrong SUN, Hua ZHANG, Zhezhe CHEN, Fangzheng ZHAO. Method for measuring road visibility in foggy condition based on convolutional neural network[J]. Systems Engineering and Electronics, 2026, 48(4): 1112-1124.

图/表 19

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图13

表1

图14

表2

图15

表3

图16

参考文献 32

1	李欣雨, 李晓雷, 詹银霞, 等. 基于视觉信息熵的高速公路雾区诱导研究综述[J]. 公路, 2024, 69 (8): 295- 306.
	LI X Y, LI X L, ZHAN Y X, et al. A review of the research on fog area induction of highway based on visual information entropy[J]. Highway, 2024, 69 (8): 295- 306.
2	周俊, 买买提江·吐尔逊. 基于支持向量机回归的道路能见度检测[J]. 现代电子技术, 2024, 47 (23): 154- 158. doi: 10.16652/j.issn.1004-373x.2024.23.023
	ZHOU J, TURSUN M. Road visibility detection based on support vector machine regression[J]. Modern Electronic Technology, 2024, 47 (23): 154- 158. doi: 10.16652/j.issn.1004-373x.2024.23.023
3	ZHANG K K, ZHENG X Y, YAN X K, et al. Research on visibility detection based on forward scattering technology[C]//Proc. of the Advanced Sensor Systems and Applications IX, 2019.
4	韦金阳, 王科平, 杨艺, 等. 基于多重迁移注意力的增量式图像去雾算法[J]. 系统仿真学报, 2024, 36 (4): 969- 980.
	WEI J Y, WANG K P, YANG Y, et al. Incremental image dehazing algorithm based on multiple transfer attention[J]. Journal of System Simulation, 2024, 36 (4): 969- 980.
5	刘万军, 程裕茜, 曲海成. 基于生成对抗网络的图像自增强去雾算法[J]. 系统仿真学报, 2024, 36 (5): 1093- 1106.
	LIU W J, CHENG Y Q, QU H C. Image self-enhanced dehazing algorithm based on generative adversarial network[J]. Journal of System Simulation, 2024, 36 (5): 1093- 1106.
6	PALVANOV A, CHO Y I. Visnet: deep convolutional neural networks for forecasting atmospheric visi bility[J]. Sensors, 2019, 19 (6): 1343- 1354. doi: 10.3390/s19061343
7	MCCARTNEY E J. Optics of the atmosphere: scattering by molecules and particles[J]. Optica Acta International Journal of Optics, 1977, 24 (7): 779- 779. doi: 10.1126/science.196.4294.1084.b
8	THOMPSON J L, PHUNG S L, BOUZERDOUM A. D-Net: a generalised and optimised deep network for monocular depth estimation[J]. IEEE Access, 2021, 9, 134543- 134555.
9	KOSCHMIEDER H, Theorie der horizontalen sichtweite[J]. Beitrage Zur Physik Der Freien Atmosphare, 1924: 33−53.
10	DUNTLEY S Q. The visibility of distant objects[J]. Journal of the Optical Society of America, 1948, 38 (3): 237- 249. doi: 10.1364/JOSA.38.000237
11	GIYENKO A, PALVANOV A, CHO Y. Application of convolutional neural networks for visibility estimation of CCTV images[C]//Proc. of the International Conference on Information Networking, 2018: 875−879.
12	WU J , WU X , GUO R X, et al. Enhancing water mist environment image visibility using an improved dark channel prior and polarimetric imaging method[J]. Optics Communications, 2025, 590: 132016.
13	HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2009, 33 (12): 2341- 2353.
14	FENG P W, WEI X W. Road extraction using modified dark channel prior and neighborhood FCM in foggy aerial images[J]. Multimedia Tools and Applications, 2019, 78 (1): 947- 964. doi: 10.1007/s11042-018-5962-0
15	SONG H , LIU N . Research and analysis of dark channel priori dehazing algorithm based on guided filtering[J]. Frontiers in Computing and Intelligent Systems, 2024, 7(2): 6−10.
16	CAI B L , XU X M, JIA K , et al. DehazeNet: an end-to-end system for single imagehaze removal[J]. IEEE Trans. on Image Processing, 2016, 25(11): 5187−5198.
17	YANG Z L, LIU B , BI X L, et al. D-Net: a dual-encoder network for image splicing forgery detection and localization[J]. Pattern Recognition, 2024, 155: 110727.
18	YANG S, SUN Z Y, JIANG Q, et al. A mixed transmission estimation iterative method for single image dehazing[J]. IEEE Access, 2021, 9 (1): 63685- 63699.
19	ZHANG L B, WANG S, WANG X H. Single image dehazing based on bright channel prior model and saliency analysis strategy[J]. IET Image Processing, 2021, 15 (5): 1023- 1031. doi: 10.1049/ipr2.12082
20	SALAZAR C S, MOYA S E U, ARREGUIN J M R, et al. Fast single image defogging with robust sky detection[J]. IEEE Access, 2020, 99 (8): 149176- 149189.
21	DONG H, PAN J S, XIANG L, et al. Multi-scale boosted dehazing network with dense feature fusion[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 2154−2164.
22	XIE B, YANG J X, SHEN J H, et al. Image defogging method combining light field depth estimation and dark channel[C]//Proc. of the IEEE International Conference on Consumer Electronics and Computer Engineering, 2021: 745−749.
23	ZHU Q S, MAI J M, SHAO L. A fast single image haze removal algorithm using color attenuation prior[J]. Digital Signal Processing, 2022, 24 (11): 3522- 3533. doi: 10.1109/TIP.2015.2446191
24	XIAO L G, HANG Y H, ZHENG W W, et al. Dark-channel based attention and classifier retraining for smoke detection in foggy environments[J]. Digital Signal Processing, 2022, 9 (1): 63685- 63699. doi: 10.1016/j.sigpro.2022.108872
25	孙景荣, 陈哲哲, 谢林昌, 等. 基于图像天空区域分割的去雾算法[J]. 系统工程与电子技术, 2023, 45(6): 1606−1615.
	SUN J R, CHEN Z Z, XIE L C, et al. Haze removal algorithm based on image sky region segmentation[J]. Systems Engineering and Electronics, 2023, 45(6): 1606−1615.
26	HORE A, ZIOU D. Image quality metrics: PSNR vs. SSIM[C]//Proc. of the International Conference on Pattern Recognition, 2010: 2366−2369.
27	HERMAN R G, SCHUMACHER S R. Using reforecasts to improve forecasting of fog and visibility for aviation[J]. Weather and Forecasting, 2016, 31 (2): 467- 482.
28	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770−778.
29	GEIGER A, LENZ P, STILLER C, et al. Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32 (11): 1231- 1237.
30	LI B Y, REN W Q, FU, D P, et al. RESIDE: a benchmark for single image dehazing[EB/OL].[2025-01-18] https://arxiv.org/pdf/1712.04143.
31	KHUDHAIR T A, MAOLOOD T A, GBASHI K E. Symmetric keys for lightweight encryption algorithms using a pre–trained VGG16 Model[J]. Telecom, 2024, 5 (3): 892- 906.
32	SARSHAR T N, SADEGHI S, KAMSARI M, et al. Advancing brain MRI image classification: integrating VGG16 and ResNet50 with a multi-verse optimizationmethod[J]. BioMed, 2024, 4 (4): 499- 523.

数据集	数量
KITTI	14999
SOTS	1000
REDE	810
FRIDA	330

真实等级	识别等级
真实等级	1	2	3	4	5	6
1	0.882	0.117	0.023	0.0032	0.002	0.001
2	0.089	0.861	0.055	0.0014	0.003	0.001
3	0.007	0.087	0.881	0.0091	0.004	0.002
4	0.005	0.023	0.054	0.8174	0.043	0.021
5	0.003	0.011	0.021	0.0093	0.823	0.063
6	0.001	0.003	0.032	0.0042	0.085	0.843

算法	FRIDA 数据集	KITTI 数据集	SOTS 数据集	高速数据集
VGG16	93.12	92.11	94.98	78.43
depth+trans	94.21	91.42	94.46	79.33
ResNet50	98.32	93.52	98.78	87.33
本文算法	99.12	96.12	99.46	89.43

[1]	蒋明煜, 张顺生, 肖思瑶. 面向轻量级交叉注意力卷积网络的SAR目标识别[J]. 系统工程与电子技术, 2025, 47(9): 2853-2861.
[2]	陈文洁, 张浦, 史高翔, 刘林, 刘烜. 基于余弦校验关系的卷积神经网络LDPC码盲识别[J]. 系统工程与电子技术, 2025, 47(9): 3117-3125.
[3]	段阿敏, 张朝辉. 基于二次分解的混合神经网络蜂窝流量预测[J]. 系统工程与电子技术, 2025, 47(5): 1687-1697.
[4]	付卫红, 张鑫钰, 刘乃安. 基于多尺度融合神经网络的同频同调制单通道盲源分离算法[J]. 系统工程与电子技术, 2025, 47(2): 641-649.
[5]	孔令宸, 刘桐辛, 周晨, 赵正予. 基于区域电离层高分辨率重构的到达时差定位分析[J]. 系统工程与电子技术, 2025, 47(11): 3521-3530.
[6]	师英杰, 周东东, 雷泰然, 黄三傲, 徐科. 纵波EMAT优化与非接触管道液位监测[J]. 系统工程与电子技术, 2025, 47(11): 3531-3542.
[7]	韦娟, 何德华, 宁方立. 基于自适应多分支卷积的声学场景分类[J]. 系统工程与电子技术, 2025, 47(10): 3148-3154.
[8]	邵永琪, 杨丽花, 常澳, 任露露. RIS辅助的OFDM系统中时变信道估计方法[J]. 系统工程与电子技术, 2025, 47(1): 324-331.
[9]	于蕾, 邓秋月, 郑丽颖, 吴昊宇. 二阶逐层特征融合网络的图像超分辨重建[J]. 系统工程与电子技术, 2024, 46(2): 391-400.
[10]	肖博文, 马泽远, 鲁天宇, 夏群利. 平台式导引头隔离度模型在线辨识方法[J]. 系统工程与电子技术, 2024, 46(11): 3595-3604.
[11]	季然, 肖茂森, 李硕, 刘宇, 罗湛仪, 程嘉维. 基于机器学习的MRTD客观测试方法研究[J]. 系统工程与电子技术, 2024, 46(10): 3265-3270.
[12]	杨帆, 马萍, 李伟, 杨明. 基于孪生网络的仿真模型智能排序评估方法[J]. 系统工程与电子技术, 2023, 45(7): 2060-2068.
[13]	姜雨, 袁琪, 胡志韬, 吴薇薇, 顾欣. 基于气象因素的机场进离港延误预测[J]. 系统工程与电子技术, 2023, 45(6): 1722-1731.
[14]	汪锐, 张天骐, 安泽亮, 王雪怡, 方竹. 基于联合特征参数和一维CNN的MIMO-OFDM系统调制识别算法[J]. 系统工程与电子技术, 2023, 45(3): 902-912.
[15]	闫啸家, 梁伟阁, 张钢, 佘博, 田福庆. 基于RCNN-ABiLSTM的机械设备剩余寿命预测方法[J]. 系统工程与电子技术, 2023, 45(3): 931-940.