Detection of Contraband Items from X-ray Scans using Semantic Deep Learning

  • Muhammad Shafay

Student thesis: Master's Thesis


Automatic detection and classification of prohibited items in passenger baggage is a challenging task, especially in cluttered and occluded concealment scenarios. To overcome the constraints of manual screening, advances in computer vision, inspired by deep learning models, have led to the creation of numerous computer-aided screening systems for threat detection in 2D and 3D CT Xray baggage security imaging. These methods seek to detect forbidden objects with enhanced effectiveness by resolving a variety of domain-specific problems such as significant occlusion, class imbalance, and insufficient labeled data, among others. However, such automated security threat identification methods are not capable of dealing with a real-life antagonist, that is looking for weaknesses in the system with malevolent intentions, such as by concealing disassembled threat components to appear neutral. In this research, we aim to test different frameworks for segmentation, detection, and classification on four publicly available datasets for X-ray security, named as GDXray, SIXRay, OPIXray, and Compass-XP. All these datasets consist of different types of contraband items that are prohibited on air travel. Each dataset has its forte such as GDXray consists of grayscale images, SIXray consists of the largest number of images of all the datasets and OPIXray consists of five different types of knives and images of SIXray and OPIXray are occluded by the baggage items as well while Compass-XP is specialized for the classification tasks. The goal is to find an architecture that is well suited to address such diverse problems and provide a comparison of different machine learning and deep learning algorithms for segmentation, detection, and classification.
Date of AwardDec 2021
Original languageAmerican English


  • Deep Learning
  • X-ray classification
  • Broad Learning System
  • Semantic segmentation
  • Object Detection.

Cite this