On the Evaluation of Prohibited Item Classification and Detection in Volumetric 3D Computed Tomography Baggage Security Screening Imagery

Abstract

Automatic prohibited object detection within 2D/3D X-ray Computed Tomography (CT) has been studied in literature to enhance the aviation security screening at checkpoints. Deep Convolutional Neural Networks (CNN) have demonstrated superior performance in 2D X-ray imagery. However, there exists very limited proof how deep neural networks perform in materials detection within volumetric 3D CT baggage screening imagery. We attempt to close this gap by applying Deep Neural Networks in 3D contraband substance detection based on their material signatures. Specifically, we formulate it as a 3D semantic segmentation problem to identify material types for all voxels based on which contraband materials can be detected. To this end, we firstly investigate 3D CNN based semantic segmentation algorithms such as 3D U-Net and its variants. In contrast to the original dense representation form of volumetric 3D CT data, we propose to convert the CT volumes into sparse point clouds which allows the use of point cloud processing approaches such as PointNet++ towards more efficient processing. Experimental results on a publicly available dataset (NEU ATR) demonstrate the effectiveness of both 3D U-Net and PointNet++ in materials detection in 3D CT imagery for baggage security screening.

On the Evaluation of Semi-Supervised 2D Segmentation for Volumetric 3D Computed Tomography Baggage Security Screening

“…Automatic threat detection has been studied within 2D Xray imagery [1]- [3] and 3D CT imagery [4], [5], [10]- [13]. State-of-the-art object detection frameworks such as Faster R-CNN [16] has been used to detect firearms, knives and other prohibited items in 2D baggage screening imagery.…”

“…Pretrained on large-scale ImageNet and MS-COCO, these object detection models can be easily fine-tuned for X-ray imagery. In [4], [5], the object detectors were adapted from 2D to 3D and used for firearm, bottle detection within volumetric 3D CT imagery. Promising results have been reported whilst the scale of dataset in terms of the numbers of category and training CT samples remains moderate compared with 2D X-ray datasets due to the difficulties of collating, annotating and storing large 3D CT datasets.…”

“…3D volumetric CT imagery provides more information for baggage security as well as the possibility of distinguishing different types of materials based on their density and effective-Z characteristics [9]. 3D Convolutional Neural Network (CNN) based methods have been used for prohibited item detection within 3D CT imagery and achieved promising results [4], [5], [10]- [13]. Similar to the situation in 2D Xray imagery [1], these approaches to prohibited item detection rely on the specific shapes and appearances hence unsuitable for contraband material detection.…”

Contraband Materials Detection Within Volumetric 3D Computed Tomography Baggage Security Screening Imagery

“…In recent years, impressive progress in deep learning techniques has enabled the possibility of fully automatic prohibited object detection in 2D X-ray imagery with high precision and very low false alarm rates [1], [2]. With the success of automatic threat object detection in 2D X-ray imagery, attempts have been made to extend this idea to 3D CT imagery with promising results achieved in prior work [3], [4]. However, the techniques used in [3] and [4] rely on the detection of specific object appearance and shape (e.g., handguns, bottles, knives, etc.)…”

“…With the success of automatic threat object detection in 2D X-ray imagery, attempts have been made to extend this idea to 3D CT imagery with promising results achieved in prior work [3], [4]. However, the techniques used in [3] and [4] rely on the detection of specific object appearance and shape (e.g., handguns, bottles, knives, etc.) hence is likely to fail in detecting contraband materials (e.g., explosive material, drugs, etc.)…”