摘要
为了提高X光下违禁品检测的准确性与效率,解决长时间、高强度的安检工作导致安检人员疲劳、检查准确性降低的问题。研究以行李物品中违禁品检测为对象,提出了一种将人工智能技术与安检仪相结合的高精度X光下违禁品检测方法。该方法利用基于深度学习的计算机视觉技术,自动检测并标注X光图像中的违禁品。高精度X光下违禁品检测方法在设计时采用了多种创新技术,以确保其在实际应用中的高效性和可靠性。具体而言,高精度X光下违禁品检测方法采用α-IoU损失函数,以缓解正负样本表达不均衡的问题。同时结合SimCSPSPPF空间金字塔池化方法和卷积块注意力机制,融合多层次语义信息,生成丰富特征表达,从而显著增强检测性能,使得高精度X光下违禁品检测在复杂的X光图像环境中也能保持较高的检测准确性和稳定性。此外,此方法还利用Merge极大值抑制方式减少误检情况,提升检测精度。从试验结果来看,在配置良好的环境下,此方法在EDS数据集和VOC数据集上均实现了较高的检测精度。同时在EDS数据集上的mAP达到了81.20%,表现出了良好的检测准确性和稳定性,满足实际安检工作的需求,对比近年来其他被广泛使用的目标检测模型,平均提升达到10.81%。研究结果表明,高精度X光下违禁品检测方法可以有效提高违禁品检测的准确率与效率,同时显著增强安检的智能化水平,为保障公共运输安全提供了重要支持。
To improve the accuracy and efficiency of contraband detection with X-ray,and to solve the problem of security personnel fatigue and reduced inspection accuracy caused by long-time and high-intensity security work,taking the contraband detection in baggage items as the object,the high precision X-ray contraband detection method is proposed.The method combines artificial intelligence technology with security checkers,and utilizes deep learning-based computer vision technology to automatically detect and label contraband in X-ray images.The high precision X-ray contraband detection method employs a variety of innovative techniques in its design to ensure its efficiency and reliability in practical applications.Specifically,the high precision X-ray contraband detection method employs theα-IoU loss function to alleviate the problem of uneven expression of positive and negative samples.It also combines the SimCSPSPPF spatial pyramid pooling method and the convolutional block attention mechanism to fuse multilevel semantic information and generate rich feature expressions,which significantly enhances the detection performance,enabling the high precision X-ray contraband detection to maintain high detection accuracy and stability even in complex X-ray image environments.In addition,the method also utilizes Merge great value suppression to reduce the false detection situation,and enhance the detection accuracy.From the experimental results,the high precision X-ray contraband detection method achieves high detection accuracy on both EDS dataset and VOC dataset in a well-configured environment.Meanwhile,the mAP on EDS dataset reaches 81.20%,which shows good detection accuracy and stability to meet the needs of actual security checking work.The average enhancement reaches 10.81%comparing with other widely used target detection models in recent years.The result indicates that the high precision X-ray contraband detection method can effectively improve the accuracy and efficiency of contraband detection,and simultaneously significantly enhance the intelligence level of security check,providing the important support for safeguarding public transportation safety.
作者
黄贤明
翁成康
黄海洋
HUANG Xian-ming;WENG Cheng-kang;HUANG Hai-yang(School of Computer Science,Hunan University of Technology,Zhuzhou,Hunan 412007,China)
出处
《公路交通科技》
CAS
CSCD
北大核心
2024年第8期55-65,共11页
Journal of Highway and Transportation Research and Development
基金
教育部中国高校产学研创新基金项目(2020ITA05043,2023DT002)
湖南省教育厅科学研究项目(21C0409)。
关键词
智能交通
违禁品检测
深度学习
计算机视觉
目标检测
注意力机制
intelligent transport
contraband detection
deep learning
computer vision
object detection
attention mechanism