融合自适应感受野与多支路特征的鞋型识别算法

doi:10.19678/j.issn.1000-3428.0061942

摘要/Abstract

摘要： 随着监控摄像头的普及和图侦技术的快速发展，“鞋印+监控”技战法成为公安机关侦破案件的重要手段。该技战法根据现场嫌疑鞋印推断出嫌疑鞋型，进而在犯罪现场周围监控视频中查找对应鞋型，锁定犯罪嫌疑人。然而现有鞋型识别算法无法充分提取嫌疑鞋印的重要特征，导致识别准确率降低。针对该问题，提出一种融合自适应感受野模块与多支路特征的鞋型识别算法。通过设计一种自适应感受野模块，使网络自适应选择合适大小的感受野特征，增强网络的特征提取能力，同时构建多支路特征融合模型，融合网络的深层和浅层特征，以充分利用有效特征进行鞋型识别，从而提高识别精度。在此基础上，采用中心损失函数和标签平滑损失函数联合训练的方法，在增大类间差距的同时缩小类内差距，增强模型的泛化能力。在多背景鞋型数据集上进行实验，结果表明，该算法Rank-1和mAP精度分别为79.77%和62.18%，具有较优的识别效果，为公安刑侦实战提供了一种可行方案。

关键词: 鞋型识别, 自适应感受野, 特征融合, 中心损失函数, 标签平滑

Abstract: With the popularization of surveillance cameras and the rapid development of graphic investigation technology, the "shoe print and surveillance" technique has become an essential means for public security organs to detect cases.This technique infers the shoe type of the suspect from the shoe print on the scene, then search the corresponding shoe type from the surveillance video around the crime scene, and then locates the suspect.A shoe type recognition algorithm based on adaptive receptive field module and multi-branch feature fusion is proposed to address the low automation problems and insufficient extraction of important shoe features by existing algorithms.An adaptive receptive field module is designed to help the network adaptively select suitable receptive field features, and then enhance the network feature extraction ability.A multi-branch feature fusion model is constructed to fuse the deep and shallow features in the network, using effective features to improve the recognition accuracy.Based on this, the Center Loss function and Label Smoothing(LS) function are used to jointly train the square method, which can effectively reduce the intra-class spacing while increasing the inter-class spacing and enhancing the generalization performance of the model.Experiments are carried out on the established multi-background shoe data set, and the results show that the algorithm is improved by rank-1 and mAP reaches 79.77% and 62.18%, respectively, which is more accurate recognition effect and provides a feasible scheme for combating public security criminal investigation.

Key words: shoe type recognition, adaptive receptive field, feature fusion, Center Loss function, Label Smoothing(LS)

中图分类号:

TP391.4

张家钧, 唐云祁, 杨智雄. 融合自适应感受野与多支路特征的鞋型识别算法[J]. 计算机工程, 2022, 48(6): 295-303.

ZHANG Jiajun, TANG Yunqi, YANG Zhixiong. Shoe Type Recognition Algorithm with Adaptive Receptive Field and Multi-Branch Feature[J]. Computer Engineering, 2022, 48(6): 295-303.

http://www.ecice06.com/CN/Y2022/V48/I6/295

图/表 15

20220625181434

20220625181437

20220625181442

20220625181446

20220625181450

20220625181454

20220625181457

20220625181501

20220625181506

20220625181510

20220625181514

20220625181519

20220625181523

20220625181527

20220625181531

参考文献

[1] 许磊, 黎智辉, 李志刚, 等.视频侦查模拟实验在案件侦破中的应用[J].刑事技术, 2018, 43(4):330-333. XU L, LI Z H, LI Z G, et al.A murder case investigated and solved by applying the simulation experiment into the collected video[J].Forensic Science and Technology, 2018, 43(4):330-333.(in Chinese)
[2] WEN Y D, ZHANG K P, LI Z F, et al.A discriminative feature learning approach for deep face recognition[C]//Proceedings of European Conference on Computer Vision.Berlin, Germany:Springer, 2016:499-515.
[3] SZEGEDY C, VANHOUCKE V, IOFFE S, et al.Rethinking the inception architecture for computer vision[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:2818-2826.
[4] LOWE D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision, 2004, 60(2):91-110.
[5] KRIZHEVSKY A, SUTSKEVER I, HINTON G E.ImageNet classification with deep convolutional neural networks[J].Communications of the ACM, 2017, 60(6):84-90.
[6] SIMONYAN K, ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[EB/OL].[2021-05-10].https://arxiv.org/abs/1409.1556.
[7] SZEGEDY C, LIU W, JIA Y Q, et al.Going deeper with convolutions[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2015:1-9.
[8] HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:770-778.
[9] HUANG G, LIU Z, VAN DER MAATEN L, et al.Densely connected convolutional networks[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:2261-2269.
[10] REN S Q, HE K M, GIRSHICK R, et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6):1137-1149.
[11] GHODRATI A, DIBA A L, PEDERSOLI M, et al.DeepProposal:hunting objects by cascading deep convolutional layers[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2015:2578-2586.
[12] WAN J, WANG D Y, HOI S C H, et al.Deep learning for content-based image retrieval:a comprehensive study[C]//Proceedings of the 22nd ACM International Conference on Multimedia.New York, USA:ACM Press, 2014:157-166.
[13] LI Y, KONG X W, ZHENG L, et al.Exploiting hierarchical activations of neural network for image retrieval[C]//Proceedings of the 24th ACM International Conference on Multimedia.New York, USA:ACM Press, 2016:132-136.
[14] 侯媛媛, 何儒汉, 李敏, 等.结合卷积神经网络多层特征融合和K-Means聚类的服装图像检索方法[J].计算机科学, 2019, 46(S1):215-221. HOU Y Y, HE R H, LI M, et al.Clothing image retrieval method combining convolutional neural network multi-layer feature fusion and K-Means clustering[J].Computer Science, 2019, 46(S1):215-221.(in Chinese)
[15] GONG Y C, WANG L W, GUO R Q, et al.Multi-scale orderless pooling of deep convolutional activation features[C]//Proceedings of European Conference on Computer Vision.Berlin, Germany:Springer, 2014:392-407.
[16] YANDEX A B, LEMPITSKY V.Aggregating local deep features for image retrieval[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2015:1269-1277.
[17] KALANTIDIS Y, MELLINA C, OSINDERO S.Cross-dimensional weighting for aggregated deep convolutional features[C]//Proceedings of European Conference on Computer Vision.Berlin, Germany:Springer, 2016:685-701.
[18] TOLIAS G, SICRE R, JÉGOU H.Particular object retrieval with integral max-pooling of CNN activations[EB/OL].[2020-05-15].https://arxiv.org/pdf/1511.05879.pdf.
[19] WEI X S, LUO J H, WU J X, et al.Selective convolutional descriptor aggregation for fine-grained image retrieval[J].IEEE Transactions on Image Processing, 2017, 26(6):2868-2881.
[20] XU J, SHI C Z, QI C Z, et al.Unsupervised part-based weighting aggregation of deep convolutional features for image retrieval[EB/OL].[2020-05-15].https://arxiv.org/pdf/1705.01247.pdf.
[21] RADENOVIC F, TOLIAS G, CHUM O.Fine-tuning CNN image retrieval with no human annotation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(7):1655-1668.
[22] 朱杰, 赵相坤, 谢博鋆, 等.基于深度特征加权的图像表示方法[J].郑州大学学报(理学版), 2020, 52(1):47-53. ZHU J, ZHAO X K, XIE B J, et al.Deep feature weighting based image representation[J].Journal of Zhengzhou University (Natural Science Edition), 2020, 52(1):47-53.(in Chinese)
[23] 李恒, 赵广社, 王鼎衡, 等.加权聚合深度卷积特征的图像检索方法[J].信息与控制, 2020, 49(1):55-61. LI H, ZHAO G S, WANG D H, et al.An image retrieval method based on weighting aggregation of deep convolutional features[J].Information and Control, 2020, 49(1):55-61.(in Chinese)
[24] SHELHAMER E, LONG J, DARRELL T.Fully convolutional networks for semantic segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4):640-651.
[25] CHEN W, LIU Y, WANG W P, et al.Deep learning for instance retrieval:a survey[EB/OL].[2020-05-15].https://arxiv.org/abs/2101.11282.
[26] DING Z Y, SONG L, ZHANG X T, et al.Selective deep ensemble for instance retrieval[J].Multimedia Tools and Applications, 2019, 78(5):5751-5767.
[27] 熊炜, 杨荻椿, 艾美慧, 等.基于多支路特征融合的行人重识别研究[J].计算机工程与科学, 2021, 43(2):329-339. XIONG W, YANG D C, AI M H, et al.Person re-identification based on multi-branch feature fusion[J].Computer Engineering & Science, 2021, 43(2):329-339.(in Chinese)
[28] AI J Q, TIAN R T, LUO Q W, et al.Multi-scale rotation-invariant haar-like feature integrated CNN-based ship detection algorithm of multiple-target environment in SAR imagery[J].IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(12):10070-10087.
[29] SHI Q Q, LI W, ZHANG F, et al.Deep CNN with multi-scale rotation invariance features for ship classification[J].IEEE Access, 2018, 6:38656-38668.
[30] YU F, KOLTUN V, FUNKHOUSER T.Dilated residual networks[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:636-644.
[31] XIE S N, GIRSHICK R, DOLLÁR P, et al.Aggregated residual transformations for deep neural networks[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:5987-5995.
[32] 张家钧, 唐云祁, 杨智雄, 等.基于注意力机制的鞋型识别算法[J].激光与光电子学进展, 2022, 59(2):365-373. ZHANG J J, TANG Y Q, YANG Z X, et al.Shoe type recognition algorithm based on attention mechanism[J].Laser & Optoelectronics Progress, 2022, 59(2):365-373.(in Chinese)
[33] 刘航, 汪西莉.自适应感受野机制遥感图像分割模型[J].中国图象图形学报, 2021, 26(2):464-474. LIU H, WANG X L.Remote sensing image segmentation model based on an adaptive receptive field mechanism[J].Journal of Image and Graphics, 2021, 26(2):464-474.(in Chinese)
[34] IOFFE S, SZEGEDY C.Batch normalization:accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on International Conference on Machine Learning.New York, USA:ACM Press, 2015:448-456.
[35] PAN X G, LUO P, SHI J P, et al.Two at once:enhancing learning and generalization capacities via IBN-Net[C]//Proceedings of European Conference on Computer Vision.Berlin, Germany:Springer, 2018:484-500.

选择文件类型/文献管理软件名称

选择包含的内容