Approaching real‐time terahertz imaging with photo‐induced coded apertures and compressed sensing
2014; Institution of Engineering and Technology; Volume: 50; Issue: 11 Linguagem: Inglês
10.1049/el.2014.0993
ISSN1350-911X
AutoresMd. Itrat Bin Shams, Zhenguo Jiang, Syed M. Rahman, Jubaid Abdul Qayyum, Li‐Jing Cheng, Huili Grace Xing, Patrick Fay, L. Liu,
Tópico(s)Advanced Optical Sensing Technologies
ResumoElectronics LettersVolume 50, Issue 11 p. 801-803 Image and vision processing and display technologyFree Access Approaching real-time terahertz imaging with photo-induced coded apertures and compressed sensing M.I.B. Shams, M.I.B. Shams Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorZ. Jiang, Z. Jiang Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorS. Rahman, S. Rahman Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorJ. Qayyum, J. Qayyum Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorL.-J. Cheng, L.-J. Cheng School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR, 97331 USASearch for more papers by this authorH. G. Xing, H. G. Xing Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorP. Fay, P. Fay Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorL. Liu, Corresponding Author L. Liu [email protected] Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this author M.I.B. Shams, M.I.B. Shams Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorZ. Jiang, Z. Jiang Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorS. Rahman, S. Rahman Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorJ. Qayyum, J. Qayyum Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorL.-J. Cheng, L.-J. Cheng School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR, 97331 USASearch for more papers by this authorH. G. Xing, H. G. Xing Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorP. Fay, P. Fay Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this authorL. Liu, Corresponding Author L. Liu [email protected] Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN, 46556 USASearch for more papers by this author First published: 01 May 2014 https://doi.org/10.1049/el.2014.0993Citations: 32AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract Photo-induced coded-aperture imaging (PI-CAI) based on compressed sensing (CS) at 590 GHz using a WR-1.5 (500–750 GHz) vector network analyser is demonstrated. For a 256-pixel (16 × 16) frame, the acquisition time can be reduced by 40%, as compared with PI-CAI without CS, while maintaining high imaging quality. On the basis of this approach, it is envisioned that real-time (26 fps) THz imaging with 1000 pixels (32 × 32) can be realised using high-speed digital micromirror device chipsets and optimised data-acquisition software. Introduction The imaging in the terahertz (THz) spectral region is promising for applications in radio astronomy, biological sensing, medical diagnostics, security screening and defence [1, 2]. In addition, the emerging applications in plasma diagnostics and avionics (e.g. brownout landing guidance) have driven the demand for high-performance real-time THz imaging systems [3, 4]. Imagers based on single detector elements and mechanical scanning are impractical for these real-time applications due to the inherently low frame rates [5]. THz focal-plane arrays could achieve the required frame rates, but are typically complicated and expensive, especially for large-scale arrays with high imaging resolutions [6]. As an alternative technique, coded-aperture imaging (CAI) based on wavefront spatial encoding and modulation has shown great potential for realising both high-performance and simple, low-cost THz imaging systems [7]. CAI requires only a single detector, which greatly reduces the system cost. To obtain an N × N pixel image, N × N measurements with a series of N × N aperture masks are needed. Coded-aperture masks for THz CAI have been demonstrated using patterns fabricated on printed circuit boards [7], and electronically actuated reconfigurable arrays have been reported based on Schottky varactors [8] and graphene modulators [9]. To date, however, all these approaches have resulted in either low imaging speed or complicated system implementations. THz photo-induced CAI (PI-CAI) has been demonstrated as an alternative [10]. In this approach, reconfigurable coded masks are directly projected on an unpatterned Si wafer by a commercial digital light processing (DLP) projector. It has been shown that each pixel 'aperture' can be optically turned on and off with a modulation depth of ∼20 dB and a modulation speed of 1.3 kHz [10]. Although imaging at 590 GHz using Hadamard codes [11] has been performed, the resulting acquisition time was nearly 30 s for a single 64-pixel frame which is far too slow for real-time applications. To achieve real-time THz PI-CAI, frame acquisition time needs to be significantly decreased. To address this issue, we report here the demonstration of PI-CAI based on compressed sensing (CS) at 590 GHz using a WR-1.5 (500–750 GHz) vector network analyser (VNA) as the source and receiver. For a 256-pixel (16 × 16) frame, the acquisition time can be reduced by 40%, as compared with PI-CAI without CS, while maintaining high imaging quality. On the basis of this approach, we project that real-time (>24 fps) THz imaging with 1000 pixels (32 × 32) can be realised by using commercially available high-speed digital micromirror device (DMD) chipsets. CS principle and experiment CS is a signal processing technique in which an image can be reconstructed accurately with a smaller number (M) of measurements than the total pixel number (N × N), i.e. M < N2. In contrast to general CAI using N2 coded masks (e.g. Hadamard coding) [11], CAI based on CS uses only M masks chosen from the N × N set of masks, leading to a much higher image acquisition speed [7]. In this Letter, the M masks were chosen from the original N2 Hadamard masks described in [11]. Image reconstruction based on M measurements can then be achieved by several different algorithms [12, 13]. We have chosen the basis pursuit method [13] in which the image, s (N2 × 1pixels) is reconstructed using L1 norm minimisation by solving (1)Here, is the L1 norm, A is the reduced measurement matrix (M × N2 pixels, N2 for each mask) and y is the measurement (M × 1). The method works by having an initial guess for the image s and updating s iteratively until a certain accuracy level is reached. To quantify the degree of compression, the compression rate (<1) is defined as R = M/N2. For a prototype demonstration, PI-CAI was performed based on CS for imaging at 590 GHz with 256 pixels (16 × 16). An Agilent N5245A PNA-X network analyser with full two-port WR-1.5 extenders (Virginia Diodes, Inc.) was used as the THz source and receiver. The THz signal transmitted from port 1 was coupled to free space using a WR-1.5 diagonal horn antenna. Two off-axis parabolic mirrors were used to collimate and focus the THz beam. As shown in Fig. 1, the focused beam (after the second parabolic mirror) was first spatially modulated by a photo-induced reconfigurable mask (realised with an unpatterned semi-insulating silicon wafer) and then reflected by an indium tin oxide (ITO) glass plate to a second WR-1.5 horn antenna (connected to port 2 of the VNA). For each mask, the measured transmission (S21) was recorded. The aperture mask was generated by illuminating Hadamard-coded-aperture arrays directly on the Si wafer by a commercial DLP projector. The ITO plate (Fig. 1) was mounted at 45° with respect to the THz beam to reflect the THz signal while allowing optical patterns from the DLP to be projected onto the Si wafer. A lens was used in the optical path to set the size of the projected aperture arrays on the Si wafer to ∼2 × 2 cm. An 'H'-shaped stencil (∼11 mm height and ∼4 mm line width) cut from an aluminium foil was placed in front of the Si wafer (Fig. 1) to be used as the target object. Images of the letter 'H' were then taken using the process described in [10] with compression rates varying from 100% (no compression) to 5%. Fig 1Open in figure viewerPowerPoint Experimental setup for THz PI-CAI based on CS. WR-1.5 VNA used as source and receiver Fig 2Open in figure viewerPowerPoint PI-CAI (16 × 16) of 'H'-shape aperture at 590 GHz with different compression rates and images after median filtering a PI-CAI of 'H'-shape aperture with different compression rates b Images after median filtering Results Fig. 2a shows the imaging results for different compression rates. The letter 'H' is clearly mapped using 256 masks (512 measurements, since two physical masks are required to realise the ' −1' pixels in standard Hadamard masks [10]), or 100% compression rate. The dimension of the 'H' in the obtained THz image (100%) agrees quite well with the original object. With decreasing R, the image quality degrades and more artefacts are seen in the images. To assist with interpreting these images, processed images using median filtering are shown in Fig. 2b. The letter 'H' can be barely discerned for R < 40%. However, imaging using R = 60–70% provides good image quality; therefore R = 0.6 was treated as a threshold level for accurate imaging for this particular target. It should be noted that the threshold level is determined by the choice of M masks. Prior knowledge of the expected target objects may help in further reducing R, resulting in further gains in imaging speed. In the case of R = 0.6, only 307 measurements (instead of 512) are required to reconstruct the original image, which reduces the image acquisition time from 24 to 14 s without significantly degrading the image quality. To understand the above effect, mean square errors (MSEs) for imaging with different compression rates are plotted in Fig. 3. MSE is calculated using , where Xr,i represents the ith reconstructed image pixel and Xi is the ith original image pixel. In this calculation, we choose the image using 100% compression rate as the 'original' image. The red dots (measured) in Fig. 3 are MSEs for real measured images with different R, as shown in Fig. 2a. The solid black (theory) line is the 'theoretical' MSE for virtually reconstructed images based on sets of randomly selected masks with a given R. As shown in Fig. 3, the measurement results show quite good agreement with theory. In addition, MSE increases slowly with decreasing R for R > 50%, whereas much faster increasing is observed for smaller R, especially when R < 30%. This indicates that a smaller number of measurements (masks) can be used for imaging while maintaining high imaging quality (i.e. small MSE) as long as an appropriate R is chosen. In the prototype demonstration in this Letter, the image acquisition time of 14 s is limited by the relatively slow (∼1.5 kHz) DMD control electronics in the DLP. By employing a 32 kHz DMD chipset (e.g. Texas Instruments DLP7000) and the optimised data-acquisition software, real-time THz imaging (∼26 fps, R = 0.6) with 1000 pixels (32 × 32) appears possible based on this approach. Fig 3Open in figure viewerPowerPoint Calculated MSE of images acquired from CS Conclusion THz imaging at 590 GHz has been performed using photo-induced coded apertures and CS. For the experimental demonstration reported here, 256-pixel (16 × 16) images have been taken with compression rates varying from 100 to 5%. Imaging performance has been fully evaluated and analysed. The calculated MSE against compression rate indicates that the image acquisition time can be reduced by at least 40%, as compared with PI-CAI without CS, while maintaining high imaging quality. Video-rate (>24 fps) THz imaging with 1000 pixels is achievable by employing faster DMD chipsets and improved data acquisition. The reported approach represents a significant step towards the realisation of cost-effective real-time THz imaging for a wide range of applications. Acknowledgments This work is partially supported by NSF grants ECCS-1002088, ECCS-1102214 and ECCS-1202452. The authors also acknowledge partial support from the Advanced Diagnostics and Therapeutics (AD&T) Initiative and the Center for Nano Science and Technology (NDNano) at the University of Notre Dame. References 1Gregory, I.S., Tribe, W.R., Cole, B.E., Baker, C., Evans, M.J., Bradley, I.V., Linfield, E.H., Davies, A.G., Missous, M.: 'Phase sensitive continuous-wave THz imaging using diode lasers', Electron. Lett., 2004, 40, (2), pp. 143– 145 (https://doi/org/10.1049/el:20040102) 2Siebert, K.J., Loffler, T., Quast, H., Thomson, M., Bauer, T., Leonhardt, R., Czasch, S., Roskos, H.G.: 'All-optoelectronic continuous wave THz imaging for biomedical applications', Phys. Med. Biol., 2002, 47, (21), pp. 3743– 3748 (https://doi/org/10.1088/0031-9155/47/21/310) 3Domier, C.W., Luhmann, N.C.Jr, Park, H.K., Xia, Z., Zhang, P.: ' Advances in millimeter wave/THz plasma diagnostics instrumentation'. 32nd Int. Conf. Infrared, Millimeter and Terahertz Waves (IRMMW-THz), Cardiff, UK, September 2007, pp. 8– 11 4Smith, R., Sundstrom, B., Belcher, B., Ewen, D.: ' Radiometric one second camera (ROSCAM) airborne evaluation'. SPIE Conf. Passive Millimeter-Wave Imaging Technology, 1999, pp. 2– 12 5Zimdars, D.: 'High speed terahertz reflection imaging', Proc. SPIE, 2005, 5692, pp. 255– 259 (https://doi/org/10.1117/12.590897) 6Liu, L., Xu, H., Lichtenberger, A.W., Weikle II, R.M.: 'Integrated 585 GHz hot-electron mixer focal-plane arrays based on annular-slot antennas for imaging applications', IEEE Trans. Microw. Theory Tech., 2010, 58, (7), pp. 1943– 1951 (https://doi/org/10.1109/TMTT.2010.2050101) 7Chan, W.L., Charan, K., Takhar, D., Kelly, K.F., Baraniuk, R.G., Mittleman, D.M.: 'A single-pixel terahertz imaging system based on compressed sensing', Appl. Phys. Lett., 2008, 93, p. 121105 (https://doi/org/10.1063/1.2989126) 8Hawasli, S., Alijabarri, N., Weikle, R.M.II: ' Schottky diode arrays for submillimeter-wave sideband generation'. 37th Int. Conf. Infrared, Millimeter, and Terahertz Waves, Wollongong, NSW, Australia, September 2012, pp. 1– 2 9Sensale-Rodriguez, B., Rafique, S., Yan, R., Zhu, M., Protasenko, V., Jena, D., Liu, L., Xing, H.G.: 'Terahertz imaging employing grapheme modulator array', Opt. Express, 2013, 21, (2), pp. 2324– 2330 (https://doi/org/10.1364/OE.21.002324) 10Cheng, L.J., Liu, L.: 'Optical modulation of continuous terahertz waves towards cost-effective reconfigurable quasi-optical terahertz devices', Opt. Express, 2013, 21, (23), pp. 28657– 28667 (https://doi/org/10.1364/OE.21.028657) 11Kannegulla, A., Jiang, Z., Rahman, S., Shams, M.I.B., Fay, P., Xing, H.G., Cheng, L.J., Liu, L.: 'Coded-aperture imaging using photo-induced reconfigurable aperture arrays for mapping terahertz beams'. IEEE Trans. THz Sci. Technol., to appear, 2014 12Tropp, J., Gilbert, A.: 'Signal recovery from random measurements via orthogonal matching pursuit', IEEE Trans. Inf. Theory, 2007, 53, (12) pp. 4655– 4666 (https://doi/org/10.1109/TIT.2007.909108) 13Chen, S.S., Donoho, D.L., Saunders, M.A.: 'Atomic decomposition by basis pursuit', SIAM J. Sci. Comput., 1998, 20, (1), pp. 33– 61 (https://doi/org/10.1137/S1064827596304010) Citing Literature Volume50, Issue11May 2014Pages 801-803 FiguresReferencesRelatedInformation
Referência(s)