Automated detection of smuggled high-risk security threats using Deep Learning

Abstract

The security infrastructure is ill-equipped to detect and deter the smuggling of non-explosive devices that enable terror attacks such as those recently perpetrated in western Europe. The detection of so-called "small metallic threats" (SMTs) in cargo containers currently relies on statistical risk analysis, intelligence reports, and visual inspection of X-ray images by security officers. The latter is very slow and unreliable due to the difficulty of the task: objects potentially spanning less than 50 pixels have to be detected in images containing more than 2 million pixels against very complex and cluttered backgrounds. In this contribution, we demonstrate for the first time the use of Convolutional Neural Networks (CNNs), a type of Deep Learning, to automate the detection of SMTs in fullsize X-ray images of cargo containers. Novel approaches for dataset augmentation allowed to train CNNs from-scratch despite the scarcity of data available. We report fewer than 6% false alarms when detecting 90% SMTs synthetically concealed in streamof-commerce images, which corresponds to an improvement of over an order of magnitude over conventional approaches such as Bag-of-Words (BoWs). The proposed scheme offers potentially super-human performance for a fraction of the time it would take for a security officers to carry out visual inspection (processing time is approximately 3.5s per container image).

A deep learning framework for the automated inspection of complex dual-energy x-ray cargo imagery

“…In our previous work on cargo, we showed that using a trained-from-scratch Convolutional Neural Network (CNN), we could detect 90% of SMTs synthetically concealed in stream-of-commerce images of ISO containers, whilst raising 6% false alarms. 7 In addition, we found that by feeding the CNN the log-image as an additional input channel, the detection performance of the network was improved considerably.…”

“…[6][7][8][9][10][11] Of these, cargo offers the most challenge for Automated Threat Detection (ATD). Images are typically much larger, and threats easier to conceal amongst dense or complex cargo.…”

“…As in our prior work, 7 we use CNN architectures based on the 19-layer very deep networks first described by Simonyan and Zisserman. 22 Our main modification to their original architecture, is the use of batch normalisation layers.…”

Multi-view X-Ray R-CNN

“…However, the use of CNNs on X-ray images for baggage inspection has been limited to the direct application of basic 2D detection algorithms, either with pretraining on photographic images or training from scratch. Jaccard et al [16] propose a black-box approach to multi-view detection by extracting CNN features from all views, concatenating them, and feeding them to fully-connected layers. Yet, the accuracy fell short of that of the original single-view detection.…”

Transferring x-ray based automated threat detection between scanners with different energies and resolution

“…[1][2][3] The Convolutional Neural Networks (CNNs) which were used obtained accuracy by training on 10 5 labelled images generated through Threat Image Projection (TIP) onto a stream-of-commerce (SoC) dataset. It is easy to conceive of scenarios where, even with TIP, training data is limited in quantity and scope-for example when a new generation of scanner is introduced, or a new threat emerges-while producing additional training data would be costly and time consuming.…”