基于ResNet和双注意力机制的赤足图像年龄预测
Age prediction of barefoot images based on ResNet and dual-attention mechanism
张涛 1韩晓雪 1成文超 1王慧 1王宇轩1
作者信息
- 1. 辽宁师范大学物理与电子技术学院,辽宁大连 116029
- 折叠
摘要
足迹是人在行走时足部与地面等客体接触时所留下的痕迹,也是犯罪现场嫌疑人最容易遗留的生物特征之一,它隐含着人体的身高、体重、性别和年龄等身份属性信息,利用足迹信息进行人的年龄预测,对指明侦察方向和缩小侦察范围有着极其重要的意义.传统侦查工作中,刑侦专家会依据积累的案件经验,依据现场遗留的足迹进行嫌疑人身份和属性的预测,但这个过程需要大量的领域知识,据此,提出了基于赤足图像的年龄自动预测方法,其由伪彩色变换模块、在线随机几何变换模块、特征提取模块、空间注意力模块和年龄预测模块组成.算法在由1818幅赤足灰度图像组成的数据集上进行了测试,预测准确率指标Acc_5和Acc_10分别达到了 55.5％和83.4％,优于现有的年龄预测方法.
Abstract
Footprints are the traces left when a person's foot comes into contact with an object such as the ground while walking,and they are also one of the most common biological features be left be-hind by a suspect at a crime scene,Footprints can provide information about a person's,weight,gender,age and other identity attributes,Using footprint information for human age prediction,to indicate the direction of reconnaissance and narrow the scope of reconnaissance,is of great signifi-cance.In traditional investigations,forensic experts make predictions about the identity and attributes of suspects based on accumulated case experience and footprints left at the scene,but this process requires a great deal of domain knowledge.Accordingly,an automatic age prediction method based on barefoot images is proposed in this paper,which consists of a pseudo-color transformation module,an online random geometric transformation module,a feature extraction module,a spatial attention module and an age prediction module.The algorithm was tested on a dataset consisting of 1,818 barefoot greyscale images,and the prediction accuracy metrics of Acc_5 and Acc_10 reached 55.5％and 83.4％,respectively,which are better than existing age prediction methods.
关键词
年龄预测/赤足图像/ResNet18网络/瓶颈注意力机制/空间注意力机制Key words
age prediction/barefoot images/ResNet18 network/bottleneck attention mechanisms/spa-tial attention mechanisms引用本文复制引用
基金项目
国家自然科学基金资助项目(62076114)
辽宁省应用基础研究计划项目(2023JH/101300189)
出版年
2024
NETL
NSTL
万方数据