Vehicle re-identification based on cross-layer attention and multi-memory

doi:10.3969/j.issn.1000-1565.2025.03.011

Abstract

Abstract: In order to improve the discriminant performance of the feature representation in the vehicle re-identification task,most of the current methods enrich the discriminant information by extracting the global information and local information of the vehicle respectively through the multi-branch structure,this will not only increase the network parameters, but also consume a lot of time.In order to solve the above problems,a vehicle re-identification method based on cross-layer attention and multiple memory units is proposed. Firstly, a cross-layer attention module is proposed to extract shallow global texture information and deep local semantic information in a single branch network,and then the cross-layer attention mechanism is used to assign weights to fuse the information of different layers.Then, the instance memory and the class memory are constructed to store the instance and class features respectively,and the influence of noise samples on feature representation is reduced by identity loss based on multiple memory.The experimental results show that the mAP,CMC@1 and CMC@5 of the proposed method on VeRi-776 dataset increase by- DOI:10.3969/j.issn.1000-1565.2025.03.011基于跨层注意力与多记忆单元的车辆重识别方法齐玉亮^{1, 2},王伟明²,王静³,熊彦臻⁴,李慧⁴(1.河北大学河北省机器视觉工程研究中心,河北保定 071002;2.河北高速公路集团有限公司京雄分公司,河北高碑店 074004;3.河北省交通运输运行监测与信息服务中心,河北石家庄 050031;4.河北大学网络空间安全与计算机学院,河北保定 071002)摘要:为了提高车辆重识别任务中特征表示的判别性能,目前大部分方法通过多分支结构来分别提取车辆的全局信息和局部细节信息,这样不仅会增加网络参数,同时还会消耗大量的时间.为了解决上述问题,提出了一种基于跨层注意力与多记忆单元的车辆重识别方法.首先,提出一种跨层注意力模块提取单分支网络中浅层的局部空间信息和深层的全局语义信息,再通过跨层注意力机制分配权重来融合不同层的信息,构建实例记忆单元和类别记忆单元来分别储存实例级特征和类别级特征,并通过基于多记忆单元的身份损失来削弱噪声样本对特征表示的影响.实验结果表明,所提出方法在VeRi-776数据集上的mAP、CMC@1、CMC@5相比于基线网络分别提高了1.5%、0.3%、0.1%.在VehicleID数据集中测试集大小为800、1 600、2 400时CMC@1相比于基线网络分别提高了1.9%、2.8%、3.8%.关键词:跨层注意力;实例记忆单元;类别记忆单元;车辆重识别中图分类号:TP183 文献标志码:A 文章编号:1000-1565(2025)03-0327-10Vehicle re-identification based on cross-layer attention and multi-memoryQI Yuliang^1,2, WANG Weiming², WANG Jing³, XIONG Yanzhen⁴, LI Hui⁴(1. Hebei Machine Vision Engineering Research Center,Hebei University,Baoding 071002,China;2. China Hebei Expressway Group Co.,Ltd. Jingxiong Branch,Gaobeidian 074004,China;3. Hebei Provincial Transportation Operation Monitoring and Information Service Center, Shijiazhuang 050031, China; 4. School of Cyber Security and Computer, Hebei University, Baoding 071002, China)Abstract: In order to improve the discriminant performance of the feature representation in the vehicle re-identification task,most of the current methods enrich the discriminant information by extracting the global information and local information of the vehicle respectively through the multi-branch structure,this will not only increase the network parameters, but also consume a lot of time.In order to solve the above problems,a vehicle re-identification method based on cross-layer attention and multiple memory units is proposed. Firstly, a cross-layer attention module is proposed to extract shallow global texture information and deep local semantic information in a single branch network,and then the cross-layer attention mechanism is used to assign weights to fuse the information of different layers.Then, the instance memory and the class memory are constructed to store the instance and class features respectively,and the influence of noise samples on feature representation is reduced by identity loss based on multiple memory.The experimental results show that the mAP,CMC@1 and CMC@5 of the proposed method on VeRi-776 dataset increase by- 收稿日期:2024-12-31;修回日期:2025-02-27 基金项目:河北高速公路集团“智能车载一体机”科技创新项目(冀高创〔2023〕294号) 第一作者:齐玉亮(1979—),男,河北省机器视觉工程研究中心工程师,主要从事高速车辆识别方向研究.E-mail:108715625@qq.com 通信作者:李慧(1993—),女,河北大学助理实验师,主要从事机器视觉方向研究.E-mail:lihui15794@hbu.edu.cn 第3期齐玉亮等:基于跨层注意力与多记忆单元的车辆重识别方法河北大学学报(自然科学版) 第45卷1.5%,0.3% and 0.1%,respectively,compared with the baseline network.With test set sizes of 800,1 600 and 2 400 in the VehicleID dataset CMC@1 increases by 1.9%,2.8% and 3.8% over the baseline network,respectively.

Key words: cross-layer attention, class memory, instance memory, vehicle re-identification

CLC Number:

TP183

QI Yuliang, WANG Weiming, WANG Jing, XIONG Yanzhen, LI Hui. Vehicle re-identification based on cross-layer attention and multi-memory[J]. Journal of Hebei University(Natural Science Edition), 2025, 45(3): 327-336.

References

[1] MENG D C, LI L, WANG S H, et al. Fine-grained feature alignment with part perspective transformation for vehicle ReID[C] //Proceedings of the 28th ACM International Conference on Multimedia. October 12 - 16, 2020, Seattle, WA, USA. ACM, 2020: 619-627. DOI:10.1145/3394171.3413573.
[2] MENG D C, LI L, LIU X J, et al. Parsing-based view-aware embedding network for vehicle re-identification[C] //2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 13-19, 2020, Seattle, WA, USA. IEEE, 2020:7101-7110.DOI:10.1109/CVPR42600.2020.00713.
[3] XU Z M, WEI L L, LANG C Y, et al. SSR-net: a spatial structural relation network for vehicle re-identification[J]. ACM Trans Multimed Comput Commun Appl, 2023, 19(6): 1-22. DOI:10.1145/3578578.
[4] QIAN J J, JIANG W, LUO H, et al. Stripe-based and attribute-aware network: a two-branch deep model for vehicle re-identification[J]. Meas Sci Technol, 2020, 31(9): 095401. DOI:10.1088/1361-6501/ab8b81.
[5] WANG H B, PENG J J, CHEN D Y, et al. Attribute-guided feature learning network for vehicle reidentification[J]. IEEE MultiMedia, 2020, 27(4): 112-121. DOI:10.1109/MMUL.2020.2999464.
[6] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C] //2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018, Salt Lake City, UT, USA. IEEE, 2018: 7132-7141. DOI:10.1109/CVPR.2018.00745.、
[7] ELSAYED G F, KORNBLITH S, LE Q V. Saccader: improving accuracy of hard attention models for vision[J]. 2019: 702-714. DOI:10.5555/3454287.3454351.
[8] CHENG D, WANG X L, WANG N N, et al. Cross-modality person re-identification with memory-based contrastive embedding[C] //Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence and Thirty-Fifth Conference on Innovative Applications of Artificial Intelligence and Thirteenth Symposium on Educational Advances in Artificial Intelligence. ACM, 2023: 425-432. DOI:10.1609/aaai.v37i1.25116.
[9] HE K M, FAN H Q, WU Y X, et al. Momentum contrast for unsupervised visual representation learning[C] //2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). June 13-19, 2020. Seattle, WA, USA. IEEE, 2020: 9729-9738. DOI:10.1109/cvpr42600.2020.00975.
[10] 彭锦佳,王辉兵.基于异构卷积神经网络集成的无监督行人重识别方法[J].电子学报, 2023, 51(10): 2902-2914. DOI:10.12263/DZXB.20220467.
[11] LIU X C, LIU W, MEI T, et al. A deep learning-based approach to progressive vehicle re-identification for urban surveillance[M] //Computer Vision – ECCV 2016. Cham: Springer International Publishing, 2016: 869-884. DOI:10.1007/978-3-319-46475-6_53.
[12] LIU H Y, TIAN Y H, WANG Y W, et al. Deep relative distance learning: tell the difference between similar vehicles[C] //2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). June 27-30, 2016, Las Vegas, NV, USA. IEEE, 2016: 2167-2175. DOI:10.1109/CVPR.2016.238.
[13] PAN X G, LUO P, SHI J P, et al. Two at once: enhancing learning and generalization capacities via IBN-net[C] //Computer Vision – ECCV 2018. ACM, 2018: 484-500. DOI:10.1007/978-3-030-01225-0_29.
[14] SUN Y F, ZHENG L, YANG Y, et al. Beyond part models: person retrieval with refined part pooling(and A strong convolutional baseline)[C] //Computer Vision – ECCV 2018. ACM, 2018: 501-518. DOI:10.1007/978-3-030-01225-0_30.
[15] JIANG G W, PANG X Y, TIAN X, et al. Global reference attention network for vehicle re-identification[J]. Appl Intell, 2023, 53(9): 11328-11343. DOI:10.1007/s10489-022-04000-6.
[16] SHEN F, ZHU J Q, ZHU X B, et al. Exploring spatial significance via hybrid pyramidal graph network for vehicle re-identification[J]. IEEE Trans Intell Transp Syst, 2022, 23(7): 8793-8804. DOI:10.1109/TITS.2021.3086142.
[17] QIAN J C, PAN M T, TONG W, et al. URRNet: a unified relational reasoning network for vehicle re-identification[J]. IEEE Trans Veh Technol, 2023, 72(9): 11156-11168. DOI:10.1109/TVT.2023.3262983.
[18] HUANG F X, LV X F, ZHANG L. Coarse-to-fine sparse self-attention for vehicle re-identification[J]. Knowl Based Syst, 2023, 270: 110526. DOI:10.1016/j.knosys.2023.110526
[19] XIE Y, WU H X, ZHU J Q, et al. Distillation embedded absorbable pruning for fast object re-identification[J]. Pattern Recognit, 2024, 152: 110437. DOI:10.1016/j.patcog.2024.110437.
[20] CUI B T, LI S, GUI J Y, et al. Viewpoint modeling with multi-task learning for vehicle re-identification[M] //PRICAI 2024: Trends in Artificial Intelligence. Singapore: Springer Nature Singapore, 2024: 37-49. DOI:10.1007/978-981-96-0125-7_4. [21] ZHENG Z D, RUAN T, WEI Y C, et al. VehicleNet: learning robust visual representation for vehicle re-identification[J]. IEEE Trans Multimed, 2020, 23: 2683-2693. DOI:10.1109/TMM.2020.3014488.
[22] PENG J J, JIANG G Q, WANG H B. Generalized multiple sparse information fusion for vehicle re-identification[J]. J Vis Commun Image Represent, 2021, 79: 103207. DOI:10.1016/j.jvcir.2021.103207.
[23] PANG X Y, YIN Y L, TIAN X. Global relational attention with a maximum suppression constraint for vehicle re-identification[J]. Int J Mach Learn Cybern, 2024, 15(5): 1729-1742. DOI:10.1007/s13042-023-01993-5.
[24] SUN K, PANG X Y, ZHENG M F, et al. Heterogeneous context interaction network for vehicle re-identification[J]. Neural Netw, 2024, 169: 293-306. DOI:10.1016/j.neunet.2023.10.032.
[25] HU W B, ZHAN H J, SHIVAKUMARA P, et al. TANet: Text region attention learning for vehicle re-identification[J]. Eng Appl Artif Intell, 2024, 133: 108448. DOI:10.1016/j.engappai.2024.108448.
[26] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C] //2017 IEEE International Conference on Computer Vision(ICCV). October 22-29, 2017, Venice, Italy. IEEE, 2017: 618-626. DOI:10.1109/ICCV.2017.74. (