Security threats on Data-Driven Approaches for Luggage Screening
Security threats on Data-Driven Approaches for Luggage Screening
复制标题
DOI：
10.1109/ccwc60891.2024.10427869
复制
发表时间：
2024-01
期刊：
影响因子：
--
通讯作者：
Debashis Gupta;Aditi Golder;Kishor Datta Gupta;Dipankar Dasgupta;Roy George;K. Shujaee
中科院分区：
文献类型：
--
作者：
Debashis Gupta;Aditi Golder;Kishor Datta Gupta;Dipankar Dasgupta;Roy George;K. Shujaee
自过去几十年以来，航空和其他运输部门越来越受欢迎，使得更难以在安全检查站周围进行严格的安全检查，以减少恐怖活动的危险。自20世纪70年代初以来，X光成像设备使安全人员能够确定货物和行李中可能存在的危险。然而，人工筛查潜在危险物体需要时间，并且容易出现人为错误。为了使用2D X射线和3D CT数据识别行李安全威胁，研究人员最近开发了计算机视觉方法，这些方法已被机器学习（ML）模型彻底改变。然而，严重的遮挡、类别不平衡和缺乏标记数据严重损害了这些技术的有效性，而狡猾的隐藏威胁（如对抗性攻击）则进一步加剧了这种情况。因此，研究界必须利用现有文献中的信息制定适当的策略，以新的方式进行研究。为了实现这一目标，我们建议实施一项系统调查，全面分析行李筛查领域的当代进步。在本文中，我们对基准数据集上分析方法的有效性进行了比较分析。此外，我们还讨论了目前行李筛查和对抗性攻击中可用的AI开发及其对行李筛查的影响。我们还讨论了现有的开放问题和未来的研究方向。
Since the previous several decades, the aviation and other transportation sectors have grown in popularity, making it more difficult to keep tight security checks surrounding security checkpoints to reduce the danger of terrorist activity. Since the early 1970s, X-ray imaging equipment has allowed security officers to locate possible hazards in the cargo and luggage. However, the manual screening of potentially dangerous objects takes time and is prone to human mistakes. To identify luggage security threats using 2D X-ray and 3D CT data, researchers have recently developed computer vision approaches that have been revolutionized by Machine Learning (ML) models. However, significant occlusion, class imbalance, and a lack of labeled data seriously impair these techniques’ effectiveness, which is further exacerbated by cunningly hidden developing threats like adversarial attacks. As a result, the research community has to develop appropriate strategies by using the information from current literature to go in new ways. In order to accomplish this objective, we suggest implementing a methodical survey that provides comprehensive analysis of contemporary advancements in the field of luggage screening. In this paper, we offer a comparative analysis of the efficacy of the analysed approaches on benchmark datasets. Furthermore, we discussed currently available AI development in luggage screening and adversarial attack and it’s impact on luggage screening. We also talk about existing open problems and prospective directions for future study.