Elsevier

Neurocomputing

Volume 134, 25 June 2014, Pages 247-253
Neurocomputing

Millimeter-wave security imaging using complex-valued self-organizing map for visualization of moving targets

https://doi.org/10.1016/j.neucom.2013.06.048Get rights and content

Abstract

We propose a millimeter-wave imaging system for moving targets, and show the dependence of its dynamics on the self-organization parameters in the complex-valued self-organizing map (CSOM) by presenting systematic experimental results. The system consists of one-dimensional array antenna, parallel front-end and a CSOM to deal with complex texture. Experiments demonstrate that the CSOM visualizes successfully a liquid-filled small plastic bottle. The system construction is presented with the adaptive processing dynamics as well as the complex-valued inner product used for determining the winner in the CSOM. It is found that the dynamics shows a higher tolerance on the self-organization coefficient for neighbors β, showing almost stable self-organization for changes over one order of magnitude, than on that for winners α.

Introduction

Recent progress in electromagnetic-wave imaging by use of microwave, millimeter-wave, and X-ray has visualized many objects in various situations. Among them, millimeter-wave imaging is expected to be utilized more and more in human life including security scenes with advantages of high directivity and penetration ability in propagation as well as its non-invasive property.

There are two bases in the physical realization. One is passive imaging. A passive system observes the black-body radiation of human body and other objects to show the temperature by taking into account the reflectance and absorptance of the object surface [1]. Passive images are obtained by using millimeter-wave antennas, wideband amplifiers, and high-frequency detectors such as schottky diodes. The passive imaging process is similar to photography since it is incoherent, and suitable for capturing target shape directly [2]. However, the black-body power is so small and the contrast is so low in most cases that the target should stay still for a certain period. Hence, the system is unsuitable for moving targets.

The other is active imaging where we illuminate the target area to receive scattered and/or reflected wave with much higher contrast [3]. Potentially it realizes quick image acquisition, practical use in combination with various types of modulation of illuminating wave, and utilization of coherence (phase information) [4]. In addition, the receiving power is so high that paralleled receiving circuits may be free from millimeter-wave amplifiers, resulting in large cost reduction. However, it often suffers from the so-called speckle noise originating from the high coherence [5] and, hence, not so good at capturing target shape in a straightforward manner. Then we employ synthetic aperture technique at a high calculation cost or neural networks for quick adaptive processing.

Passive imaging has been investigated by many research groups such as Mizuno′s group [6]. Their system uses 35 GHz wave, 100 mm lens, and corrugated Fermi antennas. Active systems have also been developed by, for example, Sheen′s group [7] to visualize weapons and liquid bottles concealed in clothes. Speckle reduction has also been investigated based on, e.g., physical Hadamard transform [8]. Presently there are three serious problems in active imaging, namely (i) long observation time for spatial scanning and aperture synthesis, (ii) high cost of millimeter-wave amplifiers and other electronics, and (iii) the privacy invasion in the synthetic aperture observation in which the body-line is visible.

This paper proposes a millimeter-wave active imaging system using one-dimensional array antenna and a complex-valued self-organizing map (CSOM) to visualize moving targets such as passengers walking through ticket gates of Shinkansen as shown in Fig. 1. Possible targets include polyethylene terephthalate (PET)-bottle liquid bombs concealed in clothes. We assume 1000 walking people per hour per gate. The use of an array antenna is one of the powerful solutions for moving-target visualization [9]. In this case, we have to develop a low-cost parallel front-end to mitigate the cost problem, which is one of the most serious problems in millimeter systems in general. Our new antenna, i.e., the bulk linearly tapered slot antenna (bulk LTSA) [10] gives the solution in combination with envelope phase detection (EPD) technique [11].

To solve the privacy problem and to shorten the measurement time by reducing the spatial resolution, we employ the CSOM so that we can visualize targets even with a low spatial resolution [12]. The CSOM has been effective in ground penetrating radar (GPR) systems to visualize plastic landmines buried underground. In the GPR case, we pay attention to the variation in the complex-amplitude textures from stones, clods, metal fragments and plastic landmines, though the reflectance values of them are very near to one another except for metal. The situation is similar to the present passenger case. A PET bottles filled with liquid explosives has a reflectance almost same as our body. Then we again pay attention to the phase information, or to the complex-amplitude in total, in its texture. Previously we tried to observe targets with a scanning pair of transmitter and receiver antennas [13] where we were successful in visualization. However, the target had to stay still.

Here, previously we proposed briefly a millimeter-wave imaging system for moving targets consisting of one-dimensional array antenna, parallel front-end and CSOM processing unit. In a previous conference paper, we presented a preliminary experiment [14]. In this paper contrarily, we conduct systematic experiments to present the stability of the CSOM dynamics. It is found that the dynamics shows a higher tolerance on the self-organization coefficient for neighbors β, showing almost stable self-organization for changes over one order of magnitude, than on that for winners α. The results promise us to develop future actually practical real-time visualization systems.

Section snippets

Measurement and processing in total

Fig. 2 shows the processing flowchart showing the total processing conducted in our previous or novel imaging system. First we acquire the scattering/reflection image in three dimensions, i.e., (two-dimensional in space)×(one-dimensional in frequency), over a wide frequency band using coherent active imaging technique. Then we extract local textural features in a window by calculating local correlations between pixel values in respective local areas in space and frequency domains [13]. The

Experiments

In this paper, the target is an 8 cm tall, 4 cm in diameter, small plastic bottle filled with water. It is attached to a bar moving at 3.0 cm/s as shown in Fig. 8. The observation area is about 20 cm×50 cm with an observation time of 16 s. There is nothing in the background in the present experiment except for a wall 1 m away from the target. We conduct calibration first without the target.

Fig. 9 shows an example of captured raw data represented in normalized amplitude in dB and phase in rad. The

Conclusion

We proposed a millimeter-wave imaging system for moving targets consisting of one-dimensional array antenna, parallel front-end and CSOM processing unit. Experiments demonstrated that the CSOM visualized successfully the target even for low-resolution measurement data in which we cannot see almost random data. We will conduct further experiment with various backgrounds to elucidate strength and weakness of our system.

Shogo Onojima received the B.S. degree in electronic and information engineering in 2010, and the M.S. in bioengineering in 2012, both from the University of Tokyo, Tokyo, Japan. His research interests include neural networks and bioelectronics.

References (15)

  • T. Hara et al.

    Plastic mine detecting radar system using complex-valued self-organizing map that deals with multiple-frequency interferometric images

    Neural Netw.

    (2004)
  • A. Sasaki et al.

    Millimeter-wave imaging using an electrooptic detector as a harmonic mixer

    IEEE J. Sel. Top. Quantum Electron.

    (2000)
  • A. Mizutani et al.

    Grating lobe suppression of narrow-wall slotted hollow waveguide millimeter-wave planar antenna for arbitrarily linear polarization

    IEEE Trans. Antennas Propag.

    (2007)
  • V. Tavakol, Q. Feng, I. Ocket, D. Schreurs, B. Nauwelaers, System modeling for active MM-wave imaging systems using an...
  • F. Qi, I. Ocket, V. Tavakol, D. Schreurs, B. Nauwelaers, Millimeter wave imaging: system modeling and phenomena...
  • G. Koers et al.

    Study of active millimeter-wave image speckle reduction by Hadamard phase pattern illumination

    J. Opt. Soc. Am., Part A

    (2008)
  • K. Mizuno, H. Matono, Y. Wagatsuma, H. Warashina, H. Sato, S. Miyanaga, Y. Yamanaka, New applications of...
There are more references available in the full text version of this article.

Cited by (19)

View all citing articles on Scopus

Shogo Onojima received the B.S. degree in electronic and information engineering in 2010, and the M.S. in bioengineering in 2012, both from the University of Tokyo, Tokyo, Japan. His research interests include neural networks and bioelectronics.

Yuya Arima received the B.S. degree in electronic and information engineering from the University of Tokyo, Tokyo, Japan, in 2013, and is now studying toward the M.S. degree. His research interests include neural networks and radar electronics.

Akira Hirose received the Ph.D. degree from the University of Tokyo, Tokyo, Japan, in 1991 in electronic engineering. In 1987, he joined the Research Center for Advanced Science and Technology (RCAST), The University of Tokyo, as a Research Associate. In 1991, he was appointed as an Instructor at the RCAST. From 1993 to 1995, on leave of absence from the University of Tokyo, he joined the Institute for Neuroinformatics, University of Bonn, Germany. Presently he is a Professor at the Department of Electrical Engineering and Information Systems, The University of Tokyo, Japan. The main fields of his interest are neural networks and wireless electronics. He served as the Chair of the Neurocomputing Technical Group in the Institute of Electronics, Information and Communication Engineers (IEICE), and presently is the President of Japanese Neural Networks Society (JNNS), the Vice-President of Electronics Society, IEICE, a member of IEEE CIS Neural Networks Technical Committee, and a Governing Board Member of Asia-Pacific Neural Network Assembly (APNNA). He also served as the Editor-in-Chief of the IEICE Transactions on Electronics and an Associate Editor of journals such as the IEEE Transactions on Neural Networks, and now serves as an Associate Editor of IEEE Geoscience and Remote Sensing Newsletter.

Dr. Hirose is a Fellow of the IEEE, a senior member of the IEICE, and a member of the JNNS.

A part of the contents was presented in the International Conference on Neural Information Processing (ICONIP) 2012 Doha. This paper has been prepared for the Special Issue on the ICONIP 2012 Doha.

View full text