89 International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/ 2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org Weapon Detection and Classification Based on Time-Frequency Analysis of Electromagnetic Transient Images Abdalrahman Al-Qubaa and Gui Yun Tian School of Electrical, Electronic and Computer Engineering Newcastle University, Newcastle upon Tyne, NE1 7RU, UK Abd.qubaa@ncl.ac.uk, g.y.tian@ncl.ac.uk Abstract—Terrorist groups, hijackers, and people hiding guns and knives are a constant and increasing threat. Concealed weapon detection has become one of the greatest challenges facing the law enforcement community today. The fact that most weapons are made from metallic materials makes electromagnetic detection methods the most prominent and preferred approach for concealed weapon detection. Each weapon has a unique electromagnetic fingerprint, determined by its size, shape and physical composition. A new detection system developed at Newcastle University that uses a walk-through metal detector with a Giant Magneto-Resistive sensor array has been utilized in this study. The system enables a two-dimensional image to be constructed from measured signals and used in later image processing. This paper addresses weapon detection using time and frequency feature extraction techniques based on this new system. The study also employs and compares two classification techniques for potential automated classification. Experimental results using guns and non-gun objects in controlled and non-controlled environments have demonstrated the potential and efficiency of the new system. The classification capabilities of the system could be developed to the point that individuals could pass through the system without the need to take off other metallic objects. The proposed techniques have the potential to produce major improvements in automatic weapon detection and classification. Keywords-sensor array; electromagnetic imaging; weapon detection; feature extraction; airport security. I. INTRODUCTION This paper, based on previous work from Al-Qubaa et al. [1], presents new results for the proposed weapon detection and classification system. There is a growing need for effective, quick and reliable security methods and techniques using new screening devices to identify weapon threats. Electromagnetic (EM) weapon detection has been used for many years, but object identification and discrimination capabilities are limited [2]. Many approaches and systems/devices have been proposed and realised for security in airports, railway stations, courts, etc. The fact that most weapons are made of metallic materials makes EM detection methods the most prominent and systems/devices built on the principle of EM induction have been prevalent for many years for the detection of suspicious metallic items carried covertly [3]. Walk-through metal detectors (WTMDs) and hand- held metal detectors (HHMDs) are commonly used as devices for detecting metallic weapons and contraband items using an EM field. Most WTMD and HHMD units use active EM techniques to detect metal objects [4][5]. Active EM means that the detector sets up a field with a source coil and this field is used to probe the environment. The applied/primary field induces eddy currents in the metal under inspection, which then generate a secondary magnetic field that can be sensed by a detector coil. The rate of decay and the spatial behaviour of the secondary field are determined by the conductivity, magnetic permeability, shape, and size of the target. Sets of measurements can then be taken and used to recover the position, the size and the shape of the objects. Many other EM imaging techniques have been used in WTMDs. These methods include microwave [6], millimetre waves [7], terahertz waves [8], infrared imaging [9], and X-ray imaging which has been used for luggage inspections in airports [10]. All these approaches have advantages and disadvantages linked to operating range, material composition of the weapon, penetrability and attenuation factors. Weapon detection systems currently available are primarily used to detect metal and have a high false alarm rate because they work by adjusting a threshold to discriminate between threat items and personal items, depending on the mass of the object. This leads to an increase in the false alarm level [11] [12]. Also, the human body can affect the sensitivity of the detector as when dealing with low conductivity or small materials, the human body can give a stronger signal than the material. This can cause the material to pass undetected, giving poor reliability [13]. Advanced signal processing algorithms have been used to analyse the magnetic field change generated when a person passes through a portal. Then pattern recognition and classification techniques can be used to calculate the probability that the acquired magnetic signature correlates to a weapon, or whether it is a non-weapon response [14]. Extracting distinct features from the EM signal is imperative for the proper classification of these signals. Feature extraction techniques are transforming the input image into a set of features. In other words, feature extraction is the use of a reduced representation, not a full representation, of an image to solve pattern recognition problems with sufficient accuracy. Extract or generate features from the EM signal is common method for metal detection and classification to represent the possible targets of interest. Feature extraction using Time-Frequency analysis has been used for stationary targets of backscattered signal [15]. Features are extracted from scattered field of a given candidate target from the joint time-frequency plane to obtain a single characteristic feature vector that can effectively represent the target of concern. Joint time frequency analysis was used to overcome the limitation of using the Fourier transform (FT) series to represent the EM signals which is require an infinite number of sinusoid functions [16]. The sinusoid function