The Federal Aviation Administration (FAA) is the leading federal agency responsible for encouraging and fostering the development of a safe, secure, and efficient national airspace system (NAS). Our goal is to establish an operating environment that ensures a threat-free system to preclude acts of terrorism and fatalities. As part of the process to meet this goal, our research and development activities continually search for technologies to ensure aviation security. Recent acts of terrorism against the aviation community have demonstrated an increasing level of sophistication in the design and deployment of explosive devices. In order to prevent the introduction of explosives onto an aircraft they must be detected prior to passenger and baggage loading. The Bulk Detection program is one method of developing a number of technologies that 'see' into and 'alarm' on suspect baggage. These detection devices must be capable of providing this serve with a confidence commensurate with the state-of-the- art available today. This program utilizes the expertise of government agencies, universities and industries working toward constructing their plans and executing their designs to produce the best available equipment.
This paper presents a current overview of the Vapor/Particle Detection Program that is currently in place at the FAA Technical Center. The Purpose of the program is to develop the systems and technologies that can eventually be deployed in airports that can screen luggage and passengers for explosives. The paper will discuss how current program is structured in terms of long term research, new term development, support activities and cooperative efforts with other agencies. In addition some of the current vapor detection technologies will be discussed as an introduction to the areas that requires signal processing.
The Federal Aviation Administration (FAA) has been granted the authority to issue research grants and to establish aviation research centers of excellence. This flexible instrument is of benefit to the FAA and to colleges, universities and other appropriate research institutions. Its purpose is to encourage and support innovative, advanced research in areas of potential benefit to the long-term growth of civil aviation. The following discussion outlines some of the principal features and indicates areas of interest to the FAA.
The potential of x-ray computed tomography (CT) for detection and identifying explosives concealed in suitcases and packages was recognized soon after the development of the first medical scanner, and several studies in this area have been published over the past 15 years. CT images are created by quantitatively determining the x-ray mass attenuation of materials within a cross section and mapping these values within a reconstruction matrix on a pixel-by- pixel basis. It is the quantitative nature of CT images that makes them useful for explosives detection. However, even with the short image cycle times currently achievable, the time required to obtain contiguous CT images throughout every suitcase is prohibitive, and means must be developed to maximize the information which can be derived from a limited number of CT images. One means of doing this is to fuse the data obtained from CT images with the data obtained from conventional or dual-energy x-ray projection images.
X-ray computerized tomography (CT) is nowadays a standard radiological imaging technique in medical applications. CT would be ideal also for many industrial non-destructive testing in such cases we can use time enough for each separate test operation. As well known in medical CT, hundreds of projections are needed to reconstruct an accurate cross-sectional image of an object. But in industrial cases a process speed will put a time limit. We seldom have time to measure hundreds of projections of a cross section, neither can we rotate the object rapidly enough with control. So, from industrial point of view, Quality and Security Control can be realized economically only, if it is technically based on only few projections (1...3) using fixed geometry, i.e. one x-ray source-detector pair for each projection including special detectors and special image processing technique.
The Radon transform and its inverse, commonly used for computed tomography (CT), are computationally burdensome for single processor computers. Since projection-based computations are easily executed in parallel, multiprocessor architectures have been proposed for high-speed operation. In this paper, we describe an architecture for a high-speed (30 MHz raster-scan image data rate), high accuracy (12-bits per pixel) computed-tomography system for use in non-destructive inspection system. This architecture reconstructs images from fan- or parallel-beam data using either single-pass or iterative reconstruction techniques. Our architecture uses a number of identical processor modules in a pipeline. Each processor module consists of memory for data storage, a commercially available digital signal processing (DSP) chip for filtering, and our custom IC which performs 450 million mathematical operations per second (MOPS). This architecture can reconstruct CT images as large as 1024 X 1024 pixels from a variety of image reconstruction algorithms. The details of the implementation and performance of our expandable architecture are discussed.
Tomographic data and tomographic reconstructions are naturally periodic in the angle of rotation of the turntable and the polar angel of the coordinates in the object, respectively. Similarly, acoustic waves are periodic and have amplitude and wavelength as free parameters that can be fit to another representation. Work has been in progress for some time in bringing the acoustic senses to bear on large data sets rather than just the visual sense. We will provide several different acoustic representations of both raw data and density maps. Rather than graphical portrayal of the data and reconstructions, you will be presented various 'tone poems.'
We have taken Cormack's original algorithm for parallel beam tomography and generalized it for fan beam tomography. The algorithm involves three steps: (1) do a FFT in angle space (angle of rotation) for each detector; (2) solve a system of integral equations for each Fourier components; and (3) back FFT to obtain the density function. Step (2) above we accomplish by a singular value decomposition (for each Fourier component). The advantages are (1) the SVD does not change from object to object under interrogation, (2) the method is fast--capable of real time analysis, (3) the null space is explicitly displayed, and (4) the SVD is usable in second-generation hunting strategies. A video of a simulated model reconstruction will be shown.
Detection of a signal from explosives at the level of a few parts in a trillion presents a formidable challenge for analysts who use detectors which rely on the chemical signatures of the material. Whether in vapor form in air, or in solid form in a matrix of ambient debris, the target chemical compounds may be present in only minute quantities compared to a large mass or volume of innocuous material. To be practicable in routine use by security personnel the false alarm rate from any detector must be very low, and the speed of analysis must be rapid to maintain a high throughput of samples. Thus, the main signal detection criteria can be defined, based on the following requirements: . very high sensitivity to specific compounds of interest, . high specificity to these compounds in order to maintain a very low false alarm rate, and . rapid data acquisition and data processing times to provide instantaneous (or real-time) detection. This paper outlines a detection process based on a tandem mass spectrometer coupled with an ionization source that operates at atmospheric pressure (APITMSTMS). This instrument has the intrinsic high sensitivity of mass spectrometry, particularly with respect to compounds such as explosives, and achieves the required high specificity and throughput by a series of rapid, but effective, filtering steps. While sensitivity is important, selectivity or the ability to discriminate between analytes of interest and the background signal is the most important factor in obtaining very low detection limits. Background signals are comprised both of electronic and chemical noise, with the sample matrix chemical interference being the primary potential source of background. Chemical and ion optical filtering of ionized vapor samples is used to select only those ions which have the correct massto-charge ratio corresponding to the target compounds. These selected ionized molecules are then introduced into a region of the MS/MS where the ions are collisionally fragmented to produce fragment ion spectra (referred to as daughter ion spectra) which are related to the structures of the initial target molecules. By identifying and matching the daughter spectral peaks with known target compound spectra, explosive compounds can be specifically detected with low detection limits.
We discuss the potential use of nonlinear optical phase conjugation to enhance the performance of various classes of optical interferometric sensors and optical fiber biosensor devices, with potential application to trace-compounds detection of explosives and other species. Examples include Michelson interferometers, laser homodyne sensors, ellipsometers, and modulation spectrometers. We speculate that compensated interferometric devices using nonlinear optical phase conjugation may lead to a new class of fieldable remote sensor which is robust, compact, inexpensive, and portable, with the capability of functioning in real-world environments. In addition to explosive sensors, these compensated remote sensors have potential application for in situ monitoring of (epitaxial) growth processes, etch and wear operations, drug interdiction, impurity sensing, and environmental monitoring.
This paper addresses the detection of Cyclotrimethylenetrinitramine (RDX) using Ion Mobility Spectrometry (IMS). It evaluates a number of post ionization Digital Signal Processing (DSP) techniques for improving the selectivity and detection limit is IMS. The results indicate that derivative methods provide the best selectivity, with minimum detectable peak separations of 0.20 msec. However, their susceptibility to noise results in poor sensitivity. Cross-correlation methods provide the best sensitivity, or detection limit, with minimum detectable RDX quantities of 0.01 nanogram. Despite their good detection limit, correlation methods suffer from a poor selectivity due to their smoothing effect. A multiresolution algorithm combining derivative methods with correlation methods is presented in this paper, and is shown to provide the advantages of both. In the proposed algorithm cross correlation methods are applied to the original signal in order to determine potential peak locations. Derivative methods are then applied within specific time windows with proper smoothing at the edges. The paper also discusses the applicability of neural networks to the peak detection problem in IMS. A brief overview of neural networks is presented with an emphasis of the Hopfield network which was selected as the most appropriate structure for the IMS peak detection problem. Peak separations down to 0.18 msec were resolved.
Thermedics Detection's explosive detectors use fast gas chromatograph analyzers to identify key components in explosives. The analyzer produces a time series of measurements which identify mobility through the columns. This time series of measurements appears as a spectrum of values with peaks corresponding to certain substances, explosive and otherwise. The analytical task of the system is to isolate the signal peaks from the detection noise, background, pedestals and peaks from extraneous substances. A uniquely modified back- propagation neural network (non-linear adaptive filter) was developed to perform the signal analysis. The unique feature of this signal analysis system was the analysis to train the network to provide only signal amplitudes but, additionally, a measure of confidence in the derived amplitudes with respect to the simulated interferents, random noise and peak time jitter included in the training. Alarms can then be set according to confidence of detection.
The Vivid Rapid Explosives Detection Systems is a true dual energy x-ray machine employing precision x-ray data acquisition in combination with unique algorithms and massive computation capability. Data from the system's 960 detectors is digitally stored and processed by powerful supermicro-computers organized as an expandable array of parallel processors. The algorithms operate on the dual energy attenuation image data to recognize and define objects in the milieu of the baggage contents. Each object is then systematically examined for a match to a specific effective atomic number, density, and mass threshold. Material properties are determined by comparing the relative attenuations of the 75 kVp and 150 kVp beams and electronically separating the object from its local background. Other heuristic algorithms search for specific configurations and provide additional information. The machine automatically detects explosive materials and identifies bomb components in luggage with high specificity and throughput, X-ray dose is comparable to that of current airport x-ray machines. The machine is also configured to find heroin, cocaine, and US currency by selecting appropriate settings on-site. Since January 1992, production units have been operationally deployed at U.S. and European airports for improved screening of checked baggage.
Minimizing signal errors and losses in high-rate nuclear-imaging systems places demands on the signal-processing and data acquisition electronics. We will describe both the data acquisition system being developed for the resonant absorption project and techniques used to minimize signal errors and losses. The data acquisition system acquires pulse-height spectra from an array of gamma-ray detectors. The data is made available to multiple processors by using the VMEbus standard to provide concurrent data analysis. In addition, we use the VxWorks real-time operating system in conjunction with a SUN workstation to develop the application software. We have designed a pulse-height-analysis board that is optimized for low dead time. This board has eight independent signal channels, each consisting of a charge integrator, a fast analog-to-digital converter, and a first-in/first-out memory. This board also contains a 68020 CPU that performs the initial data compression and stores digitized data into dual-ported memory. By using an independent high-speed signal channel for each detector, we are able to improve performance over the standard multiplexed techniques commonly in use.
Research and development are in progress in signal processing for explosive detection for airport security. The recent advancements in real-time computer vision, however, will provide possibilities of far more new and innovative ways to augment airport's human security force. For example, a computer vision system with real-time object tracking capability can monitor the behavior of large groups of people, computers can keep detailed records of the flow of people through certain areas, allowing security personnel to predict and quickly respond to crowd or isolation situations, as well as identifying potential suspects through suspicious movement patterns. Further, such a system would enable airport security officers to accurately monitor a particular suspect's movements within the airport, relayed from camera to camera, without alerting him to their observation or diverting manpower from other areas. Computerized face recognition and facial expression extraction are other actively researched areas in computer vision that have potential for airport security applications. I present an overview of the state-of-the-art computer vision research than can have relevance to airport security, and discuss how they can be applied and what aspects of that technology need to be researched further before they can be put into practical application.
Array Systems Computing Inc. (ASC) is developing a prototype Computer Assisted X-ray Screening System (CAXSS) which uses state-of-the-art image processing and computer vision technology to detect threats seen in x-ray images of passenger carry-on luggage at national and international airports. This system is successful in detecting weapons including guns, knives, grenades, aerosol cans, etc. Currently, bomb detection is also being implemented; preliminary results using this bomb detector are promising.
Recognition of partially occluded objects has been an important issue to airport security because occlusion causes significant problems in identifying and locating objects during baggage inspection. In this paper, we present the annealed Hopfield network (AHN) for occluded object matching problems. In AHN, the mean field theory is applied to the hybrid Hopfield network in order to improve computational complexity of the annealed Hopfield network and provide reliable matching under heavily occluded conditions. AHN is slower than HHN. However, AHN provides near global solutions without initial restrictions and provides less false matching than HHN. In conclusion, a new algorithm based upon a Neural Network approach was developed to demonstrate the feasibility of the automated inspection of threat objects from X-ray images. The robustness of the algorithm is proved by identifying occluded target objects with large tolerance of their features.
X-ray images by their very nature are difficult to interpret. For most security applications these images have been generated by using Linear Array type imaging sensors, and presented on standard video monitors. A major problem for the observer is the lack of three dimensional information present. It must be remembered that the images are essentially shadows which have been projected onto a plane by transmitted radiation. Therefore, the cues to depth which we say associate with a normal photograph (i.e. produced by reflected light), such as, occlusion and to some extent linear perspective, are missing. The loss of depth cues can, and does, cause serious ambiguities to arise in the interpretation of complex x-ray images.
In this paper we provide the results of an empirical investigation of iterative maximum entropy spectrum estimation in two dimensions. The Lim-Malik algorithm is compared to the classical periodogram algorithm on a number of analytic as well as practical two-dimensional signals. We study the convergence of the Lim-Malik algorithm and suggest some criterion to assure convergence in the two-dimensional case.
A new technology has been developed for detecting explosives and other dangerous objects concealed under persons' clothing. The 'Subambient Exposure, Computer Utilized Reflected Energy' (SECURE) method uses a very low level of back-scattered x-rays in conjunction with digital image processing to produce an image of the person and any concealed objects. Image processing algorithms, used in the system are directed at presenting information to a human operator in the best possible manner for foreign object detection. These algorithms are viewed as being near optimum, and additional development is probably not justified. Algorithm development is needed in the area of automatic threat detection. This has the potential of reducing the invasion of privacy associated with having a security operator view each image. It also has the potential of reducing the serious problem of operator complacency. In one approach, the new algorithm must (1) recognize and isolate objects in the image, (2) discriminate between concealed objects and human anatomy, and (3) discriminate between dangerous and benign concealed objects. The images produced with the SECURE technology are extremely noisy due to the low levels of radiation used. Any algorithm developed must perform well in this noisy environment. Execution of the algorithm must be accomplished in less than a few seconds. Hardware to implement the algorithm must be of a complexity and cost compatible with the commercial SECURE system.
An automated detection and recognition system is described for use with x-ray images of luggage inspection systems. The development goal was to achieve an automated analysis which may support operator-based control of luggage. The focus of attention lies on the recognition and checking of specified objects to which the system has been adapted during a training phase. The system trained so far concentrates on the detection of detonators and fire- arms. The segmentation gives various objects from which several features are extracted. These features are presented to a classifier which assigns the objects to predetermined categories. For classification a specially trained neural network (multilayer perceptron) is used. For detection of weapons first performance data are available. Detection of detonators is in the laboratory stage and shown first results.
High-resolution nuclear magnetic resonance (NMR) provides detailed spectral information on the molecular level. As such, it has been a principal tool of chemists since the early 1950s and used to probe the molecular structure, configuration and composition of liquids. However, high-resolution NMR techniques require very homogeneous DC and RF magnetic fields and are thus inappropriate as the basis for a practical liquid explosives screening system. These field requirements are relaxed for low-resolution NMR, but the unique spectral features of the high-resolution NMR signal are obscured. In spite of this, low-resolution NMR does provide substantial chemical information regarding liquids. Specific parameters available from low- resolution NMR include the signal amplitude (A0), the spin-lattice relaxation time (T1), the spin-spin relaxation time (T2), the diffusion constant (D), and the spin-spin coupling constant (J). Sequences of RF pulses can be designed to respond to one or more of these parameters and, therefore, unique NMR signatures for various liquids can be defined. General considerations of the relative importance of each of these parameters, along with practical considerations regarding allowable scanning and data processing times, will in large part determine the nature of signal processing methods to be used in the NMR Liquid Explosives Screening System.
Various millimeter-wave imaging systems capable of imaging through clothing for the detection of contraband metal, plastic, or ceramic weapons, have been developed at PNL. Two dimensional scanned holographic systems, developed at 35, 90, and 350 GHz, are used to obtain high resolution images of metal and plastic targets concealed by clothing. Coherent single-frequency amplitude and phase data, which is gathered over a two-dimensional scanned aperture, is reconstructed to the target plane using a holographic wavefront reconstruction technique. Practical weapon detection systems require high-speed scanning. To achieve this goal, a 35 GHz linear sequentially switched array has been built and integrated into a high speed linear scanner. This system poses special challenges on calibration/signal processing of the holographic system. Further, significant improvements in speed are required to achieve real time operation. Toward this goal, a wideband scanned system which allows for a two- dimensional image formation from a one-dimensional scanned (or array) system has been developed. Signal/image processing techniques developed and implemented for this technique are a variation on conventional synthetic aperture radar (SAR) techniques which eliminate far- field and narrow-bandwidth requirements. Performance of this technique is demonstrated with imaging results obtained from a Ka-band system.