基于改进YOLOv8的X线安检图像违禁品检测方法  

A Contraband Detection Method for X-ray Security ImagesBased on Improved YOLOv 8

作　　者：毛玮杨 杨军[1,2] 刘栩栋 梁道正 MAO Weiyang;YANG Jun;LIU Xudong;LIANG Daozheng(School of Computer Science,Sichuan Normal University,Chengdu 610101,Sichuan;Key Laboratory of Visual Computing and Virtual Reality,Sichuan Normal University,Chengdu 610101,Sichuan)

机构地区：[1]四川师范大学计算机科学学院,四川成都610101 [2]四川师范大学可视化计算与虚拟现实四川省重点实验室,四川成都610101

出　　处：《四川师范大学学报（自然科学版）》2025年第2期253-260,共8页Journal of Sichuan Normal University（Natural Science）

基　　金：国家自然科学基金(62006165);四川省自然科学基金(2022NSFSC0552)。

摘　　要：人工安检效率低,易出错,实现基于人工智能的自动安检是安检的发展趋势.针对YOLOv8目标检测模型在X线违禁品检测中检测精度低和对少量类别漏检率高的问题,对YOLOv8模型进行改进.在YOLOv8n的基础上修改网络结构,引入注意力机制,提出带有通道注意力(efficient channel attention,ECA)的YOLOv8-ECA目标检测模型,以便更好地提取X线图像中违禁品的特征,同时采用图像旋转等一系列数据增强方法,对少量类别样本进行样本扩充.在自建的X线安检图像数据集上进行实验.实验结果表明,改进后的算法较原始YOLOv8n模型在检测精度上提升6%,在检测速度上较原始YOLOv8n模型提升15.7%,同时降低少量类别的漏检率.The efficiency of manual security checks is low and prone to errors. Implementing automatic security checks basedon artificial intelligence is the development trend of security checks. The YOLOv8 object detection model has beenimproved to address the issues of low detection accuracy and high missed detection rate for a small number ofcategories in X-ray prohibited item detection. On the basis of YOLOv8n, the network structure was modified,attention mechanism was introduced, and a YOLOv8n-ECA object detection model with Efficient ChannelAttention (ECA) was proposed to better extract the features of prohibited items in X-ray images. At the same time, aseries of data augmentation methods such as image rotation were used to expand the sample size for a small numberof category samples. Experiments were conducted on a self-building X-ray security inspection image dataset, andthe results showed that the improved algorithm enhanced detection accuracy by 6% compared to the originalYOLOv8n model, increased detection speed by 15.7% compared to the original YOLOv8n model, and reduced themissed detection rate of a small number of categories.

关 键 词：YOLOv8n ECA注意力 深度学习 X线图像 违禁品检查 

分 类 号：O357.5[理学—流体力学]