基于多目标的能见度检测方法

doi:10.19678/j.issn.1000-3428.0063745

摘要/Abstract

摘要： 能见度对人类生产生活、交通运输安全等具有重要影响，是地面自动气象观测的重要内容之一，但由于受影响因素较多，目前能见度检测仍缺乏统一的标准和检定规程。现有基于图像的能见度检测方法大多从整幅图像或局部区域中提取视觉特征估计能见度，未考虑不同景深目标物对应子图像的质量衰减程度不同，导致检测结果精确度和稳定性不高。提出一种新的能见度检测方法，使用预训练的神经图像评估，从不同景深目标物对应的子图像中提取视觉特征，并将提取的特征和能见度真值输入到全连接网络，以训练子图像的能见度映射模型。根据子图像与全局图像间的关系，动态建立各目标在能见度整体估计过程中的权重回归模型，按照权重融合各目标物能见度估计值，得到整幅图像的能见度检测值。实验结果表明，该方法能有效提升回归模型的预测精度，其在不同能见度区间的检测正确率均超过85%。

关键词: 能见度检测, 多目标检测, 视觉特征, 神经图像评估, 全连接网络

Abstract: Visibility is extremely important in all aspects of human life and plays a critical role in automated meteorological observations from ground stations.However, no unified standards for specifications or verification procedures have been established to characterize visibility.Existing image-based visibility detection methods extract visual features from an entire image or local area to estimate visibility without considering the differences in the attenuation of the visual quality of sub-images corresponding to objects with different depths of field in a given scene. This typically results in poor precision and stability.Hence, this study presents a new visibility detection method designed to reliably detect visibility automatically.First, a pre-trained Neural Image Assessment(NIMA) network is utilized to extract visual features from sub-images of objects with different depths of field.Then, the extracted features and ground-truth labels are input into a fully connected network to train a model for mapping sub-images to visibility. Based on the relationships between sub-images and the entire image, a weight regression model is dynamically established for each object.Finally, the visibility estimates of each object are fused by weight to obtain visibility detection values for the entire image.Experimental results show that the proposed method can effectively improve the prediction accuracy of the regression model and achieve a detection accuracy of more than 85% in different visibility intervals.

Key words: visibility detection, multi-object detection, visual features, Neural Image Assessment(NIMA), fully connected network

中图分类号:

TP393

唐榕, 李骞, 唐绍恩. 基于多目标的能见度检测方法[J]. 计算机工程, 2023, 49(2): 314-320.

TANG Rong, LI Qian, TANG Shaoen. Visibility Detection Method Based on Multi-Object[J]. Computer Engineering, 2023, 49(2): 314-320.

http://www.ecice06.com/CN/Y2023/V49/I2/314

图/表 14

20230216183222

20230216183226

20230216183229

20230216183233

20230216183237

20230216183240

20230216183243

20230216183246

20230216183249

20230216183253

20230216183257

20230216183300

20230216183303

20230216183307

参考文献

[1] International Commission on Illumination.Contrast and visibility (1st edition) (E):CIE 95-1992[S].1992-01-01.
[2] 左芬.武汉天河国际机场低能见度的特征分析[J].空运商务, 2019, 37(1):58-61. ZUO F.Characteristic analysis of low visibility in Wuhan Tianhe international airport[J].Air Transport & Business, 2019, 37(1):58-61.(in Chinese)
[3] 于堃.江苏省能见度的分布特征分析及低能见度的影响研究[D].南京:南京信息工程大学, 2019. YU K.Distribution characteristic analysis of visibility & research of low visibility impact in Jiangsu Province[D].Nanjing:Nanjing University of Information Science & Technology, 2019.(in Chinese)
[4] 袁成松.低能见度交通气象信息服务系统的研究与应用[D].南京:南京农业大学, 2007. YUAN C S.Study and application on the transportation meteorological information and service system for low-visibility processes[D].Nanjing:Nanjing Agricultural University, 2007.(in Chinese)
[5] 赵晓莉, 闫军, 苑跃, 等.能见度自动观测仪与人工观测资料对比分析[J].计算技术与自动化, 2013, 32(4):138-141. ZHAO X L, YAN J, YUAN Y, et al.Comparative analysis visibility automatic and artificial observation data[J].Computing Technology and Automation, 2013, 32(4):138-141.(in Chinese)
[6] 花毓幸, 曾燕, 邱新法.基于图像兴趣窗格测算大气能见度的方法研究[J].科技通报, 2017, 33(8):39-42. HUA Y X, ZENG Y, QIU X F.Research of measurement method of visibility based on optimal ROI[J].Bulletin of Science and Technology, 2017, 33(8):39-42.(in Chinese)
[7] 项文书.基于交通视频的能见度估计研究[D].上海:上海交通大学, 2014. XIANG W S.Visibility estimation using traffic surveillance video[D].Shanghai:Shanghai Jiao Tong University, 2014.(in Chinese)
[8] 邢向楠, 崔岩梅, 李涛, 等.基于透射法的能见度测量装置设计与实验研究[J].计测技术, 2011, 31(3):10-13. XING X N, CUI Y M, LI T, et al.Design and experiment research of visibility measuring device based on transmission method[J].Metrology & Measurement Technology, 2011, 31(3):10-13.(in Chinese)
[9] 程寅, 陆亦怀, 连翠华, 等.前向散射型能见度仪的研制[J].大气与环境光学学报, 2006, 1(4):59-63. CHENG Y, LU Y H, LIAN C H, et al.Investigation on the forward-scattering visibility sensor[J].Journal of Atmospheric and Environmental Optics, 2006, 1(4):59-63.(in Chinese)
[10] 孙学金, 王晓蕾, 李浩等.大气探测学[M].北京:气象出版社, 2009. SUN X, WANG X, LI H, et al.Atmosphere sounding and observing[M].Beijing:China Meteorological Press, 2009.(in Chinese)
[11] 许茜, 殷绪成, 李岩, 等.基于图像理解的能见度测量方法[J].模式识别与人工智能, 2013, 26(6):543-551. XU X, YIN X C, LI Y, et al.Visibility measurement with image understanding[J].Pattern Recognition and Artificial Intelligence, 2013, 26(6):543-551.(in Chinese)
[12] 陆天舒, 吕伟涛, 杨俊等.一种能见度观测方法:CN201610720847.6[P].2017-02-15. LU T, LÜ W, YANG J, et al.A visibility observation method:CN201610720847.6[P].2017-02-15.
[13] 苗苗.视频能见度检测算法综述[J].现代电子技术, 2012, 35(15):72-75. MIAO M.Survey on video-based visibility detection method[J].Modern Electronics Technique, 2012, 35(15):72-75.(in Chinese)
[14] 唐绍恩, 李骞, 顾大权, 等.一种基于多元回归的能见度检测方法[J].信息技术与网络安全, 2018, 37(3):70-73, 79. TANG S E, LI Q, GU D Q, et al.A method of visibility detection based on multiple regression[J].Information Technology and Network Security, 2018, 37(3):70-73, 79.(in Chinese)
[15] 唐绍恩, 李骞, 胡磊, 等.一种基于迁移学习的能见度检测方法[J].计算机工程, 2019, 45(9):242-247. TANG S E, LI Q, HU L, et al.A visibility detection method based on transfer learning[J].Computer Engineering, 2019, 45(9):242-247.(in Chinese)
[16] ISHIMOTO K, TAKEUCHI M, NAITOU S, et al.Development and certification of a visibility-range monitor by image processing[J].Annals of Glaciology, 1989, 13:117-119.
[17] GAZZI M.A field experiment on contrast reduction law[J].Atmospheric Environment, 1994, 28(5):901-907.
[18] HE K M, SUN J, TANG X O.Single image haze removal using dark channel prior[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(12):2341-2353.
[19] 石玉立, 王彬, 卜帆.基于图像特征的大气能见度估算方法[J].南京理工大学学报, 2018, 42(5):552-559. SHI Y L, WANG B, BU F.Atmospheric visibility measurement based on image feature[J].Journal of Nanjing University of Science and Technology, 2018, 42(5):552-559.(in Chinese)
[20] YIN X C, HE T T, HAO H W, et al.Learning based visibility measuring with images[C]//Proceedings of the 18th International Conference on Neural Information.Berlin, Germany:Springer, 2011:711-718.
[21] OUTAY F, TAHA B, CHAABANI H, et al.Estimating ambient visibility in the presence of fog:a deep convolutional neural network approach[J].Personal and Ubiquitous Computing, 2021, 25(1):51-62.
[22] LI J P, LO W L, FU H, et al.A transfer learning method for meteorological visibility estimation based on feature fusion method[J].Applied Sciences, 2021, 11(3):997.
[23] TALEBI H, MILANFAR P.NIMA:neural image assessment[J].IEEE Transactions on Image Processing, 2018, 27(8):3998-4011.
[24] HOWARD A G, ZHU M L, CHEN B, et al.MobileNets:efficient convolutional neural networks for mobile vision applications[EB/OL].[2021-12-01].https://arxiv.org/abs/1704.04861.
[25] CHAABANI H, et al.A neural network approach to visibility range estimation under foggy weather conditions[J].Procedia Computer Science, 2017, 113:466-471.

选择文件类型/文献管理软件名称

选择包含的内容