To solve the practical problems in vehicle type classification on toll roads, a deep learning network based on dual-view feature fusion, named ETCLNet, was proposed in this paper. A dual-branch architecture was adopted by the network to extract the features of vehicle head and side body images, respectively, and combined with an innovatively designed multi-scale feature extraction module, namely IPFE and an adaptive weighting-based bilinear feature fusion (AWBF) mechanism, efficient and fine-grained vehicle feature representation was achieved. Through the design of parallel multi-scale convolutions and residual connections, the problem of gradient vanishing was effectively alleviated, and the adaptability to complex scenarios was enhanced by the IPFE module. The high-order interaction of features between angles was further optimized by the AWBF mechanism, and efficient fusion of multi-angle features was achieved, and classification performance was significantly improved through adaptive weighting and element-wise multiplication. Experimental results indicate that ETCLNet outperforms existing mainstream models in multiple evaluation metrics (e.g., accuracy, precision, recall, F1 value, and SAUC), which solves the problem of the low vehicle type recognition rate of cameras in existing toll road audit systems. In addition, existing hardware devices are fully utilized by this design, which reduces deployment costs. A scientific and efficient solution for multi-view vehicle recognition and smart transportation is provided by this paper, and new ideas for deep learning network design are offered.
PENGYubin. Vehicle trajectory tracking and vehicle identification system based on FMCW millimeter wave radar[D]. Harbin: Harbin Institute of Technology, 2020.
[7]
王君逸. 基于多特征融合的交通车辆声信号识别方法研究[D]. 南昌: 华东交通大学, 2022.
[8]
WANGJunyi. Research on traffic vehicle acoustic signal recognition method based on multi-feature fusion[D]. Nanchang: East China Jiaotong University, 2022.
ZENGXiaming, HEJiaxiong, CAOLingen, et al. Research on vehicle recognition based on laser radar vehicle inspector[J]. Chinese Journal of Construction Machinery, 2021, 19(4): 324-330.
ZHENGJianlong, CHENMengjie, LIUChaochao. Review of intelligent construction development of highway infrastructures and its prospects[J]. Journal of China & Foreign Highway, 2025, 45(2): 1-20.
[15]
邓航. 基于Gabor特征简化的车型识别[D]. 西安: 西安电子科技大学, 2017.
[16]
DENGHang. Vehicle recognition based on Gabor feature simplification[D]. Xi’an: Xidian University, 2017.
LUOLiang, JunjieLYU, LITao, et al. Automatic recognition of engineering vehicle type based on improved NC-HOG feature[J]. Computer Engineering and Design, 2021, 42(11): 3164-3173.
[19]
BISHARAA, MAZEE H, MAZEM. Considerations for the implementation of machine learning into acute care settings[J]. British Medical Bulletin, 2022, 141(1): 15-32.
[20]
ZHAOP N, LAIL F. Minimax rate optimal adaptive nearest neighbor classification and regression[J]. IEEE Transactions on Information Theory, 2021, 67(5): 3155-3182.
[21]
宋嘉冀. 基于视频的车辆信息提取与检索技术研究[D]. 南京: 东南大学, 2017.
[22]
SONGJiaji. Research on vehicle information extraction and retrieval technology based on video[D]. Nanjing: Southeast University, 2017.
CHENGShuhong, GAOXu, ZHOUBin. Vehicle recognition based on multi-feature extraction and SVM parameter optimization[J]. Acta Metrologica Sinica, 2018, 39(3): 348-352.
[25]
LECUNY, BENGIOY, HINTONG. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[26]
YAOD H, ZHUW X, CHENY J, et al. Chinese license plate character recognition based on convolution neural network[C]//Jinan: 2017 Chinese Automation Congress (CAC). 2017: 1547-1552.
MULiang, ZHAOHong, LIYan, et al. Vehicle recognition based on gradient compression and YOLO v4 algorithm[J]. Chinese Journal of Engineering, 2022, 44(5): 940-950.
ZHANGNian, ZHANGLiang. Type recognition of highway trucks based on deep learning[J]. Journal of Traffic and Transportation Engineering, 2023, 23(1): 267-279.
FENGHuijie, ZHAOHongdong, YUKuaikuai, et al. Vehicle acoustic recognition based on lightweight CNN[J]. Transducer and Microsystem Technologies, 2024, 43(7): 136-140.
WANGDai. Detection and identification of vehicle lanes and vehicles in bad weather based on improved YOLOX[D]. Beijing: China University of Geosciences, 2023.
WANGKaiming, LUOZhehao, SUNWenqi, et al. Vehicle recognition algorithm based on FIPA model and adaptive collaborative optimization[J]. Journal of Computer Applications, 2024, 44(sup 2): 267-273.
DONGYibing, ZENGHui, LIJianke, et al. DPRT-YOLO: Real-time object detector for intelligent and connected vehicles in complex driving environments[J]. Computer Engineering and Applications, 2025, 61(14): 148-162.
ZENGYongqiang, ZHANGYonghua, ZHAOHui, et al. Object detection and tracking based on fusion of radar and camera on highway side[J]. Journal of China & Foreign Highway, 2025, 45(6): 261-267.
[47]
VASWANIA, SHAZEERN, PARMARN, et al. Attention is all you need[PP/OL]. arXiv.(2011-06-03)[2025-02-12].
[48]
SZEGEDYC, LIUW, JIAY Q, et al. Going deeper with convolutions[C]//Boston:2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015: 1-9.
[49]
SZEGEDYC, VANHOUCKEV, IOFFES, et al. Rethinking the inception architecture for computer vision[C]//Las Vegas: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 2818-2826.
[50]
SANDLERM, HOWARDA, ZHUM L, et al. MobileNet v2: Inverted residuals and linear bottlenecks[C]//Salt Lake City: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 4510-4520.
[51]
KOONCEB. Convolutional Neural networks with swift for tensorflow: Image recognition and dataset categorization[M]. Berkeley, CA: Apress, 2021.
[52]
IANDOLAF N, HANS, MOSKEWICZM W, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5 MB model size[PP/OL]. arXiv.(2016-02-07)[2025-02-12].
[53]
DAIY Q, ZHENGT, XUEC B, et al. MViT-PCD: A lightweight ViT-based network for Martian surface topographic change detection[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 3000405.
[54]
WANGC Y, YEHI H, LIAOH M. YOLO v9: Learning what you want to learn using programmable gradient information[PP/OL]. arXiv.(2024-02-13)[2025-02-12].