Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2023, Vol. 57 Issue (4): 712-718    DOI: 10.3785/j.issn.1008-973X.2023.04.008
自动化技术、计算机技术     
基于注意力和自适应权重的车辆重识别算法
苏育挺(),陆荣烜,张为*()
天津大学 电气自动化与信息工程学院，天津 300072
Vehicle re-identification algorithm based on attention mechanism and adaptive weight
Yu-ting SU(),Rong-xuan LU,Wei ZHANG*()
School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
 全文: PDF(1937 KB)   HTML
摘要：

为了解决车辆重识别算法受类内差异性与类间相似性的干扰无法充分表示车辆特征的问题，提出基于注意力机制和自适应损失权重的车辆重识别算法. 该算法采用改进的主干网络ResNet50_ibn，避免了颜色、光照、视角等客观因素的干扰，提取关于目标的不变性特征. 搭建基于注意力机制的组表示网络，融合特征间的相互依赖关系，从不同分组的特征表示中提取更加丰富的特征信息. 设计自适应损失权重计算损失函数，使用多损失函数策略对网络模型进行训练. 该算法在公开数据集VeRi776与VehicleID上的首位击中率和平均精度均值分别达到了96.0%、79.8%和81.5%、80.9%. 实验结果表明，利用该算法提取的特征更具判别性，综合性能优于现有的其他车辆重识别算法.
关键词： 车辆重识别;  注意力机制;  自适应损失权重;  机器视觉;  深度学习    
Abstract:

A vehicle re-identification algorithm based on attention mechanism and adaptive loss weights was proposed in order to solve the problem that the vehicle re-identification algorithms cannot adequately represent vehicle features due to intra-class differences and inter-class similarities. The improved backbone network ResNet50_ibn was used to avoid the influence of objective factors such as color, illumination and perspective, and extract invariant features about the target. An attention mechanism was introduced to build a group representation network, which fused the interdependence between features by extracting from different groups and extracted abundant feature information from different feature representations. The loss function was improved by adaptive loss weight, and the network was trained by using a multi-loss function strategy. The algorithm achieved 96.0% Rank-1, 79.8% mAP and 81.5% Rank-1, 80.9% mAP, respectively on the public datasets VeRi776 and VehicleID. The experimental results show that the features extracted by the algorithm are more discriminative. The comprehensive performance is better than other existing vehicle re-identification algorithms.
Key words: vehicle re-identification    attention mechanism    adaptive loss weight    machine vision    deep learning
收稿日期: 2022-04-23 出版日期: 2023-04-21
CLC:  TP 391  
基金资助: 国家自然科学基金资助项目(61802277)
通讯作者: 张为     E-mail: ytsu@tju.edu.cn;tjuzhangwei@tju.edu.cn
作者简介: 苏育挺（1972—），男，教授，博导，从事多媒体信息处理研究. orcid.org/0000-0001-5165-204X. E-mail： ytsu@tju.edu.cn
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
苏育挺
陆荣烜
张为

引用本文:

苏育挺,陆荣烜,张为. 基于注意力和自适应权重的车辆重识别算法[J]. 浙江大学学报(工学版), 2023, 57(4): 712-718.

Yu-ting SU,Rong-xuan LU,Wei ZHANG. Vehicle re-identification algorithm based on attention mechanism and adaptive weight. Journal of ZheJiang University (Engineering Science), 2023, 57(4): 712-718.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2023.04.008    或    https://www.zjujournals.com/eng/CN/Y2023/V57/I4/712

图 1  基于注意力和自适应权重的车辆重识别网络结构
图 2  改进的残差块结构
图 3  相似性计算模块的结构图
%
方法 Rank-1 Rank-5 mAP
ResNet50 84.2 88.1 65.6
ResNet50+GroupNet 85.9 89.6 67.8
ResNet50_ibn 88.7 92.1 71.2
ResNet50_ibn+GroupNet 90.4 94.6 72.9
表 1  不同模块对算法性能的影响
%
损失函数 Rank-1 Rank-5 mAP
CE+Triplet 90.4 94.6 72.9
CE+SupCon 91.8 96.0 74.4
Label smoothing CE+Triplet 91.2 95.3 73.8
Label smoothing CE+SupCon 92.5 96.6 75.2
表 2  不同损失函数的实验结果
%
损失权重 Rank-1 Rank-5 mAP
1∶1 92.5 96.6 75.2
1∶2 93.8 97.2 76.7
0.5∶0.5 94.5 97.7 77.8
自适应损失权重 96.0 98.6 79.8
表 3  不同损失权重的实验结果
%
方法 Rank-1 Rank-5 mAP
OIFE+ST[19] 92.4 — 51.4
VAMI+STR[20] 85.9 91.8 61.3
PNVR[21] 94.3 98.3 74.3
MRL[22] 94.3 97.3 78.5
UMTS[6] 95.8 — 75.9
SPAN[9] 93.9 97.6 68.6
PVEN[23] 95.6 98.4 79.5
TBE[24] 96.0 98.5 79.5
本文方法 96.0 98.6 79.8
表 4  VeRi776数据集下与主流算法的结果对比
%
方法 小尺度 中尺度 大尺度
Rank-1 mAP Rank-1 mAP Rank-1 mAP
OIFE[19] — — — — 67.0 —
VAMI[20] 63.1 — 52.9 — 47.3 —
PNVR[21] 78.4 — 75.0 — 74.2 —
MRL[22] 75.7 78.3 71.6 75.4 66.5 68.2
UMTS[6] 74.4 80.4 72.4 77.1 69.8 75.2
PVEN[23] 78.4 78.3 75.0 78.3 74.2 78.3
TBE[24] 86.0 — 82.3 — 80.7 —
本文方法 84.5 85.3 82.5 82.6 81.5 80.9
表 5  VehicleID数据集下与主流算法的结果对比
图 4  重识别检测结果的对比
1 LIU X, WU L, MA H, et al. Large-scale vehicle re-identification in urban surveillance videos [C]// Proceedings of the IEEE International Conference on Multimedia and Expo. Seattle: IEEE, 2016: 1-6.
2 ZAPLETAL D, HEROUT A. Vehicle re-identification for automatic video traffic surveillance [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. California: IEEE, 2016: 1568-1574.
3 CORMIER M, SOMMER L W, TEUTSCH M. Low resolution vehicle re-identification based on appearance features for wide area motion imagery [C]// Proceedings of the IEEE Winter Applications of Computer Vision Workshops. New York: IEEE, 2016: 1-7.
4 LUO H, GU Y, LIAO X, et al. Bag of tricks and a strong baseline for deep person re-identification [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. Long Beach: IEEE, 2019: 1487-1495.
5 ZHENG Z, RUAN T, WEI Y, et al. VehicleNet: learning robust visual representation for vehicle re-identification [C]// Proceedings of the IEEE Transactions on Multimedia. New York: IEEE, 2021: 2683-2693.
6 JIN X, LAN C, ZENG W, et al. Uncertainty-aware multi-shot knowledge distillation for image-based object re-identification [C]// Proceedings of the AAAI Conference on Artificial Intelligence. New York: AAAI, 2020: 11165-11172.
7 KIM Y, PARK W, ROH M C, et al. GroupFace: learning latent groups and constructing group-based representations for face recognition [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 5620-5629.
8 谢秀珍, 罗志明, 连盛, 等 一种融合表观与属性信息的车辆重识别方法[J]. 厦门大学学报: 自然科学版, 2021, 60 (1): 72- 79
XIE Xiu-zhen, LUO Zhi-ming, LIAN Sheng, et al A vehicle re-identification method by fusing the vehicle appearance and attribute information[J]. Journal of Xiamen University: Natural Science Edition, 2021, 60 (1): 72- 79
9 CHEN T S, LIU C T, WU C W. et al. Orientation-aware vehicle re-identification with semantics-guided part attention network [C]// European Conference on Computer Vision. Amsterdam: Elsevier, 2020: 330–346.
10 刘晗煜, 黄宏恩, 郑世宝 基于视角一致性三元组损失的车辆重识别技术[J]. 测控技术, 2021, 40 (8): 47- 53
LIU Han-yu, HUANG Hong-en, ZHENG Shi-bao View consistency triplet loss for vehicle re-identification[J]. Measurement and Control Technology, 2021, 40 (8): 47- 53
doi: 10.19708/j.ckjs.2021.08.009
11 NGUYEN B X, NGUYEN B D, DO T, et al. Graph-based person signature for person re-identifications [C]// Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE, 2021: 3487-3496.
12 HUYNH S V, NGUYEN N H, NGUYEN N T, et al. A strong baseline for vehicle re-identification [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE, 2021: 4142-4149.
13 HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
14 WU Y, HE K Group normalization[J]. International Journal of Computer Vision, 2018, 128: 742- 755
15 PRANNAY K, PIOTR T, CHEN W, et al. Supervised contrastive learning [EB/OL]. [2022-04-20]. https://arxiv.org/abs/2004.11362.
16 LIU X, LIU W, MEI T, et al PROVID: progressive and multimodal vehicle reidentification for large-scale urban surveillance[J]. IEEE Transactions on Multimedia, 2019, 20 (3): 645- 658
17 LIU H, TIAN Y, WANG Y, et al. Deep relative distance learning: tell the difference between similar vehicles [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 2167-2175.
18 KINGMA D, BA J. Adam: a method for stochastic optimization [C]// International Conference on Learning Representations. San Diego: [s. n.], 2015: 1412-1426.
19 WANG Z, TANG L, LIU X, et al. Orientation invariant feature embedding and spatial temporal regularization for vehicle reidentification [C]// Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 832-837.
20 YI Z, LING S. Viewpoint-aware attentive multi-view inference for vehicle re-identification [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2017: 6489-6498.
21 HE B, LI J, ZHAO Y, et al. Part-regularized near-duplicate vehicle re-identification [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 3992-4000.
22 LIN W, Y LI, YANG X, et al. Multi-view learning for vehicle re-identification [C]// IEEE International Conference on Multimedia and Expo. Shanghai: IEEE, 2019: 832-837.
23 MENG D, LI L, LIU X, et al. Parsing-based view-aware embedding network for vehicle re-identification [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 7101-7110.
[1] 程艳芬,吴家俊,何凡. 基于关系门控图卷积网络的方面级情感分析[J]. 浙江大学学报(工学版), 2023, 57(3): 437-445.
[2] 曾耀,高法钦. 基于改进YOLOv5的电子元件表面缺陷检测算法[J]. 浙江大学学报(工学版), 2023, 57(3): 455-465.
[3] 兰欢,余建波. 基于深度学习三维成型的钢板表面缺陷检测[J]. 浙江大学学报(工学版), 2023, 57(3): 466-476.
[4] 曾菊香,王平辉,丁益东,兰林,蔡林熹,管晓宏. 面向节点分类的图神经网络节点嵌入增强模型[J]. 浙江大学学报(工学版), 2023, 57(2): 219-225.
[5] 杨帆,宁博,李怀清,周新,李冠宇. 基于语义增强特征融合的多模态图像检索模型[J]. 浙江大学学报(工学版), 2023, 57(2): 252-258.
[6] 刘超,孔兵,杜国王,周丽华,陈红梅,包崇明. 高阶互信息最大化与伪标签指导的深度聚类[J]. 浙江大学学报(工学版), 2023, 57(2): 299-309.
[7] 鲁建厦,包秦,汤洪涛,邵益平,赵文彬. 无设备人体追踪系统的择优标签方法[J]. 浙江大学学报(工学版), 2023, 57(2): 415-425.
[8] 马骏驰,迪骁鑫,段宗涛,唐蕾. 程序表示学习综述[J]. 浙江大学学报(工学版), 2023, 57(1): 155-169.
[9] 叶晨,战洪飞,林颖俊,余军合,王瑞,钟武昌. 基于推理-情境感知激活模型的设计知识推荐[J]. 浙江大学学报(工学版), 2023, 57(1): 32-46.
[10] 王林涛,毛齐. 基于RGB与深度信息融合的管片抓取位置测量方法[J]. 浙江大学学报(工学版), 2023, 57(1): 47-54.
[11] 凤丽洲,杨阳,王友卫,杨贵军. 基于Transformer和知识图谱的新闻推荐新方法[J]. 浙江大学学报(工学版), 2023, 57(1): 133-143.
[12] 刘近贞,陈飞,熊慧. 多尺度残差网络模型的开放式电阻抗成像算法[J]. 浙江大学学报(工学版), 2022, 56(9): 1789-1795.
[13] 王万良,王铁军,陈嘉诚,尤文波. 融合多尺度和多头注意力的医疗图像分割方法[J]. 浙江大学学报(工学版), 2022, 56(9): 1796-1805.
[14] 郝琨,王阔,王贝贝. 基于改进Mobilenet-YOLOv3的轻量级水下生物检测算法[J]. 浙江大学学报(工学版), 2022, 56(8): 1622-1632.
[15] 夏杰锋,唐武勤,杨强. 光伏航拍红外图像的热斑自动检测方法[J]. 浙江大学学报(工学版), 2022, 56(8): 1640-1647.