| 大数据环境下深度学习行人再识别技术研究与应用 |
| |
| 引用本文： | 马吉忠,谢一,马全海,武文魁,李文琪,李玥. 大数据环境下深度学习行人再识别技术研究与应用[J]. 南方农机, 2021, 0(8): 29-30,33 |
| |
| 作者姓名： | 马吉忠 谢一 马全海 武文魁 李文琪 李玥 |
| |
| 作者单位： | 甘肃农业大学信息科学技术学院 |
| |
| 基金项目： | 甘肃省自然科学基金(18JR3RA165);甘肃农业大学信息科学技术学院发展基金项目(GAU-XKFZJJ-2020-02);甘肃农业大学SRTP(202016024)。 |
| |
| 摘 要： | 行人再识别(Re-ID)是网络匹配行人图像任务.它与属性识别在学习行人描述上有共同目标,而区别是粒度.通过基于属性标签和ID标签的互补性,多数Re-ID方法仅考虑行人身份标签,而包含详细本地描述的属性有利于Re-ID模型学习更多判别式特征表示.因此提出属性的人识别(APR)网络.APR网络是通过学习Re-ID嵌入并同时...
|
| 关 键 词： | 行人再识别 深度学习 图像 属性 |
Research and Application of Pedestrian Re-identification Technology Based on Deep Learning in Big Data Environment |
| |
| Ma Jizhong,Xie Yi,Ma Quanhai,Wu Wenkui,Li. Research and Application of Pedestrian Re-identification Technology Based on Deep Learning in Big Data Environment[J]. , 2021, 0(8): 29-30,33 |
| |
| Authors: | Ma Jizhong Xie Yi Ma Quanhai Wu Wenkui Li |
| |
| Affiliation: | (College of Information Science Technology of Gansu Agriculture University,Gansu Lanzhou 730070) |
| |
| Abstract: | Pedestrian Re-ID(Re-ID)is the task of matching pedestrian images with the network.Based on the complementarity of attribute tags and ID tags,most Re-ID methods only consider pedestrian identity tags,and the attributes that include detailed local descriptions help the Re-ID model to learn more discriminative feature representations.Therefore,an attributed person recognition(APR)network is proposed.The APR network is a multi-task network that learns Re-ID embedding and predicts pedestrian attributes at the same time.In the experiment,after considering the dependence and correlation between attributes and weighting the attribute prediction,the APR retrieval process is ten times faster,and the accuracy on Market-1501 drops by 2.92%. |
| |
| Keywords: | Pedestrian re-identification Deep learning Image Attributes |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
