基于卷积神经网络的安检X光图像违禁品多标签识别
Multi-label Recognition of Prohibited Items in Security X-ray Images Based on Convolutional Neural Network
杨登杰 1江式坤 1周亮2
作者信息
- 1. 中国联合网络通信有限公司政企客户事业群解决方案部,北京 100033
- 2. 中国联合网络通信有限公司甘肃省分公司政企客户事业群,甘肃 兰州 730030
- 折叠
摘要
在安全检查工作中,违禁品的检测与识别仍然过度依赖于安检员的视觉经验.如何自动判别出X光图像中存在的常见违禁品,进而辅助安检员进行决策,成为安检领域的一个亟待解决的问题.基于深度学习技术研究安检X光图像中的违禁品多标签识别方法.通过引入multi-hot向量标注法对安检X光数据集进行标注,迁移训练Darknet-53 卷积神经网络实现对X光图像中违禁品的类别判定.实验结果表明,安检X光图像违禁品多标签识别平均精度达到了 98%以上,满足现实安检场景下的应用需求.
Abstract
In safety inspections,the detection and identification of prohibited items still excessively rely on the visual ex-perience of security inspectors.How to automatically identify common prohibited items in X-ray images and assist security inspectors in making decisions has become an urgent issue in the field of security inspection.This paper is based on deep learning technology to study the multi label recognition method for prohibited items in security X-ray images.By introducing the multi hot vector annotation method to annotate the security X-ray dataset,the Darknet-53 convolutional neural network is transferred and trained to determine the category of prohibited items in X-ray images.The experimental results show that the average accuracy of multi label recognition for prohibited items in X-ray images of security checks has reached over 98%,meeting the application requirements in real security check scenarios.
关键词
卷积神经网络/安检X光图像/违禁品/多标签识别Key words
convolutional neural network/security X-ray images/prohibited items/multi-label recognition引用本文复制引用
出版年
2024
NETL
NSTL
万方数据