Loading...
Recognizing Plastic Threats in Baggage X-Rays Using Deep Networks
Alawar, Mohammad Suhail Abdulwahid Abdulrahman
Alawar, Mohammad Suhail Abdulwahid Abdulrahman
Date
2019-12
Advisor
Type
Thesis
Degree
Description
A Master of Science thesis in Electrical Engineering by Mohammad Suhail Abdulwahid Abdulrahman Alawar entitled, “Recognizing Plastic Threats in Baggage X-Rays Using Deep Networks”, submitted in December 2019. Thesis advisor is Dr. Usman Tariq and thesis co-advisors are Dr. Nasser Qaddoumi and Dr. Hasan Mir. Soft copy is available (Thesis, Approval Signatures, Completion Certificate, and AUS Archives Consent Form).
Abstract
Transportation authorities are faced with the challenge of detecting contraband items, particularly firearms. In most cases, security personnel can reliably perform this task. However, these tasks require strong focus, which causes the operators to become stressed out in a short amount of time. During baggage inspections, some bags can be very difficult to classify due to the superimposing effect that can occur in cluttered bags. Additionally, officials desire enhanced passenger throughput in order to boost the profitability of the airport. This implies a need to increase the rate of passengers going through the X-ray detection machine. These challenges motivate the urgent need for an automated X-ray recognition system that can classify various types of objects, thus enhancing the accuracy of the X-ray machine operators’ final decision through the use of computer vision aided software. In this research, we propose new methods as well as modifications of existing methods that can be used to automatically classify hard-torecognize threats. In particular, we focus on detecting plastic threats instead of the easyto-recognize metal threats. The main contribution of this work is the development of a novel paradigm to classify 3D-printed plastic threats (such as a gun) in baggage, including a method for systematic data collection and an approach for Threat Image Projection. We also prove the effectiveness of our assumptions and approaches by generalizing our system to recognize new threats that the system has not been trained for.