基于知识蒸馏的道路交通标志识别神经网络

doi:10.3778/j.issn.1002-8331.2308-0446

摘要/Abstract

摘要： 自然场景下的交通标志识别易受到光照、遮挡和模糊等因素的干扰，从而影响检测精度；同时现有的深度学习模型参数量多、计算复杂度较高导致模型推理时间较长。提出了一种基于知识蒸馏的神经网络架构AFE-ViT（adaptive feature extraction-vision Transformer）用于道路交通标志识别，该架构由自适应特征提取模块和轻量级ViT（vision Transformer）分类器组成，其融合了图像中局部和全局特征信息，对自然场景下的道路交通标志识别具有更好的适应性。其中，自适应特征提取模块结合了InceptionNetV1、SKNet思想和残差结构，实现了多感受野的自适应选择，并作为ViT的前置模块，有效提高了特征提取效率。选择ResNet18作为教师网络，AFE-ViT作为学生网络，采用特征级和输出级知识蒸馏方法对AFE-ViT进行蒸馏，压缩模型参数。实验结果表明，该方法的识别准确率可达98.98%，模型参数量仅为9.9×105，表现优于同类深度学习模型。

关键词: 交通标识, 知识蒸馏, 自适应特征提取

Abstract: Recognition of traffic signs in natural scenes is susceptible to interference from factors such as lighting, occlusions, and blurriness, which can affect detection accuracy. Additionally, existing deep learning models have a large number of parameters and high computational complexity, resulting in longer model inference times. The article proposes a neural network architecture adaptive feature extraction-vision Transformer (AFE-ViT) based on knowledge distillation for road traffic sign recognition. The architecture consists of an adaptive feature extraction module and a lightweight vision Transformer (ViT) classifier. It combines local and global feature information in the image, and has better adaptability to road traffic sign recognition in natural scenes. Among them, the adaptive feature extraction module combines InceptionNetV1, SKNet ideas and residual structure to realize the adaptive selection of multiple receptive fields, and as the front module of ViT, it effectively improves the efficiency of feature extraction. It chooses ResNet18 as the teacher network and AFE-ViT as the student network, and uses feature-level and output-level knowledge distillation methods to distill AFE-ViT and compress model parameters. The experimental results show that the recognition accuracy of this method can reach 98.98%, and the number of model parameters is only 9.9×105, which is better than similar deep learning models.

Key words: traffic sign identification, knowledge distillation, adaptive feature extraction

葛怡源, 于明鑫. 基于知识蒸馏的道路交通标志识别神经网络[J]. 计算机工程与应用, 2024, 60(19): 110-119.

GE Yiyuan, YU Mingxin. Lightweight Road Traffic Sign Identification Neural Network Based on Knowledge Distillation[J]. Computer Engineering and Applications, 2024, 60(19): 110-119.

参考文献

[1] 潘峰, 潘振华, 熊亮, 等. 人工智能技术在智能驾驶控制中的应用[J]. 北京联合大学学报, 2022, 36(3): 30-37.
PAN F, PAN Z H, XIONG L, et al. Application of artificial intelligence in intelligent driving control[J]. Journal of Beijing Union University, 2022, 36(3): 30-37.
[2] 杨康. 人工智能在汽车驾驶技术领域的应用与发展[J]. 汽车实用技术, 2022, 47(11): 16-19.
YANG K. Application and development of artificial intelligence in the field of automobile driving Technology[J]. Automobile Applied Technology, 2022, 47(11): 16-19.
[3] 黄东风. 人工智能在汽车驾驶技术领域的应用与发展[J]. 时代汽车, 2022(1): 42-43.
HUANG D F. Application and development of artificial intelligence in the field of automobile driving technology[J]. Auto Time, 2022 (1): 42-43.
[4] 牟凯, 张舒, 曹洪斌. 人工智能技术在智慧交通领域的应用研究[J]. 物流科技, 2022, 45(20): 98-100.
MOU K, ZHANG S, CAO H B. Application research on artificial intelligence technology in intelligent transportation field[J]. Logistics Sci-Tech, 2022, 45(20): 98-100.
[5] 刘少博, 张晖, 吴超仲.TRB2017年会—002行人交通及智能交通研究综述[J]. 交通信息与安全, 2017, 35(3): 1-10.
LIU S B, ZHANG H, WU C Z. A review of 2017 transportation research board 96th annual meeting with the focuses and areas of pedestrian traffic and intelligent transportation[J].?Journal of Transport Information and Safety, 2017, 35(3): 1-10.
[6] 朱金好, 罗晓萍. 基于决策树型SVM的交通标志图像识别[J]. 长沙理工大学学报(自然科学版), 2004(2): 13-17．
ZHU J H, LUO X P. Recognition of traffic sign image based on decision-tree-based support vector machine[J]. Journal of Changsha University of Science and Technology(Natural Science), 2004(2): 13-17.
[7] 罗晓萍, 蒋加伏, 唐贤瑛. 基于SVM和模糊免疫网络的交通标志图像识别[J]. 计算机工程与设计, 2006, 27(9): 1542-1544．
LUO X P, JIANG J F, TANG X Y. Recognition of traffic sign image based on support vector machine and fuzzy immune networks[J]. Computer Engineering and Design, 2006, 27(9): 1542-1544.
[8] 刘红, 高向东, 杨大鹏. 一种基于仿射变换的三角形交通标志矫正方法[J]. 汽车零部件, 2010(6): 56-57.
LIU H, GAO X D, YANG D P. A correction method for triangle traffic signs based on affine transformation[J].Automobile Parts, 2010(6): 56-57.
[9] LIU M, MAO J X. Traffic signs recognition based on affine invariant Hu’s moment features[J]. Applied Mechanics and Materials, 2013, 321/324: 945-949.
[10] 孙露霞, 张尤赛, 李永顺. 基于感兴趣区域和HOG-CTH特征的交通标志检测[J]. 计算机与数字工程, 2018, 46(6): 1222-1226．
SUN L X, ZHANG Y S, LI Y S. Traffic sign detection based on regions of interest and HOG-CTH features[J].Computer and Digital Engineering, 2018, 46(6): 1222-1226.
[11] 辛靖宇, 徐伟昊, 赵子亮, 等. 基于深度学习的交通标志识别[J]. 北京汽车, 2023(2): 35-38.
XIN J Y, XU W H, ZHAO Z L, et al. Traffic sign recognition based on deep learning[J]. Beijing Automotive Engineering, 2023(2): 35-38.
[12] 徐仙伟, 曹霁. 基于深度学习的交通标志识别算法[J]. 计算机时代, 2019(6): 67-70.
XU X W, CAO J. Traffic sign recognition algorithm with deep learning[J]. Computer Era, 2019(6): 67-70.
[13] ZHU Y Z, YAN W Q. Traffic sign recognition based on deep learning[J]. Multimedia Tools and Applications, 2022, 81(13):17779-17791.
[14] FARAG W. Traffic signs classification by deep learning for advanced driving assistance systems[J]. Intelligent Decision Technologies, 2019, 13(3): 305-314.
[15] 邓乐平, 李伟. 深度学习多场景下交通标志快速检测方法研究[J]. 测绘技术装备, 2022, 24(3): 86-92.
DENG L P, LI W. Fast detection method of traffic signs in multi-scenarios of deep learning[J]. Geomatics Technology and Equipment, 2022, 24(3): 86-92.
[16] 何锐波, 狄岚, 梁久祯. 一种改进的深度学习的道路交通标识识别算法[J].智能系统学报, 2020, 15(6): 1121-1130.
HE R B, DI L, LIANG J Z. An improved deep learning algorithm for road traffic identification[J]. CAAI Transactions on Intelligent Systems, 2020, 15(6): 1121-1130.
[17] 耿经邦, 梁正友. 基于改进ResNet的交通标识识别[J]. 电子技术与软件工程, 2020(6): 138-140.
GENG J B, LIANG Z Y. Traffic sign recognition based on Improved ResNet[J]. Electronic Technology and Software Engineering, 2020(6): 138-140.
[18] 廖聪, 郭凰, 赵茂军, 等. 基于图像增强和SKNet的交通标志识别[J]. 计算机与现代化, 2023(3): 23-28.
LIAO C, GUO H, ZHAO M J, et al. Traffic sign recognition based on image enhancement and SKNet[J]. Computer and Modernization, 2023(3): 23-28.
[19] LI X, WANG W, HU X, et al. Selective kernel networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 510-519.
[20] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words: Transformers for image recognition at scale[J]. arXiv:2010.11929, 2020.
[21] HINTON E G, VINYALS O, DEAN J. Distilling the knowledge in a neural network[J]. arXiv:1503.02531, 2015.
[22] 赵胜伟, 葛仕明, 叶奇挺, 等. 基于增强监督知识蒸馏的交通标识分类[J]. 中国科技论文, 2017, 12(20): 2355-2360.
ZHAO S W, GE S M, YE Q T, et al. Traffic sign classification based knowledge distillation with augmentation supervision[J]. China Sciencepaper, 2017, 12(20): 2355-2360.
[23] ALEXANDER W, JAVAD M S, MICHAEL J S. MicronNet: a highly compact deep convolutional neural network architecture for real-time embedded traffic sign classification[J]. IEEE Access, 2018, 6: 59803-59810.
[24] REZA F R, GOU K, KOHICHI O. Lightweight spatial pyramid convolutional neural network for traffic sign classification[C]//Proceedings of the 2018 Indonesian Association for Pattern Recognition International Conference (INAPR), Jakarta, Indonesia, 2018:23-28.
[25] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018.
[26] WANG Q L, WU B G, ZHU P F, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.
[27] SANGHYUN W, JONGEHAN P, JOON-YOUNG L, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision, 2018.
[28] ZHANG Q L, YANG Y B. SA-Net: shuffle attention for deep convolutional neural networks[C]//Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2021.
[29] HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, TN, USA, 2021: 13708-13717.
[30] 袁穆佳惠, 陈晓.基于卷积神经网络的交通标志识别方法研究[J].计算机与数字工程, 2023, 51(6):1323-1327.
YUAN M J H, CHEN X. Research on traffic sign recognition method based on CNN[J].Computer & Digital Engineering, 2023, 51(6):1323-1327.
[31] HE S H, CHEN L, ZHANG S Y, et al. Automatic recognition of traffic signs based on visual inspection[J]. IEEE Access, 2021, 9: 43253-43261.
[32] 赵泽毅, 周福强, 王少红, 等. 应用双通道卷积神经网络的交通标识识别方法[J]. 中国测试, 2024, 50(6): 35-41.
ZHAO Z Y, ZHOU F Q, WANG S H, et al. Traffic sign recognition method based on dual channel CNN[J]. China Measurement & Testing Technology, 2024, 50(6): 35-41.
[33] VASU P K A, GABRIEL J, ZHU J, et al. FastViT: a fast hybrid vision transformer using structural reparameterization[J]. arXiv:2303.14189, 2023.
[34] LIU X Y, PENG H, ZHENG N X, et al. EfficientViT: memory efficient vision transformer with cascaded group attention[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 14420-14430.