OSA's Digital Library

Optics Express

Optics Express

  • Editor: Michael Duncan
  • Vol. 13, Iss. 17 — Aug. 22, 2005
  • pp: 6572–6586
« Show journal navigation

Multi-spectral mid-infrared laser stand-off imaging

Yi Wang, Yang Wang, and Han Q. Le  »View Author Affiliations


Optics Express, Vol. 13, Issue 17, pp. 6572-6586 (2005)
http://dx.doi.org/10.1364/OPEX.13.006572


View Full Text Article

Acrobat PDF (1944 KB)





Browse Journals / Lookup Meetings

Browse by Journal and Year


   


Lookup Conference Papers

Close Browse Journals / Lookup Meetings

Article Tools

Share
Citations

Abstract

A multi-spectral mid-IR laser imaging study including system engineering, experiments, and image processing and analysis is described. A 4-λ scalable system was built with semiconductor lasers, covering from 3.3-9.6 μm. The X-Y scanning system was capable of 2-dimensional (2D) multi-spectral imaging at a stand-off distance from 13-40 m. The system was applied to diverse targets that consist of man-made and natural materials and objects, and shown capable to resolve and distinguish small spectral differences among the various targets. Colorless objects in the visible were shown with “colorful” signatures in the mid-IR. Image processing algorithm based on spectral contrast was shown most effective to exploit the laser sensitivity and accuracy, as opposed to algorithms that operate mainly on the image spatial intensity. The results also showed the complexity of laser imaging phenomenology, involving both spectroscopic and geometrical scattering effects. A demonstration of 3D multi-spectral imaging was also given. The system design is suitable for compact packages with semiconductor lasers, and the results suggest that laser-based multi-spectral imaging can be a unique and powerful technology for target discrimination.

© 2005 Optical Society of America

1. Introduction

Multi-spectral imaging is a widely used technology in remote sensing, military surveillance, chemical identification, and machine vision. The technology achieves target enhancement/detection or material identification by relying on the distinctive spectral signatures of objects. An important spectral region is the mid-infrared (MIR) 3–14 μm, which contains unique vibration-band spectral signatures of virtually all compounds. There are two types of systems, passive and active. Passive systems [1

1 . See e. g. J. B. Campbell , Introduction to Remote Sensing , 2nd ed. ( Guilford Press. 1996 ), Chap. 14.

] rely on ambient light or target thermal radiation. A very large number of spectral images (hyperspectral) can be obtained using various optical band-pass filtering techniques. But the lack of control over the ambient light condition limits its application.

This paper is a study of multi-spectral laser imaging to investigate its potential and issues for applications. The study involved building a system using mid-IR semiconductor lasers, which was applied to a range of diverse targets that consist of common man-made and natural targets, including household materials and objects, plants, chemicals, soils, foods, … and others. Unlike other works [2–7

2 . T. P. Jannson , P. I. Shnitser , A. A. Kostrzewski , I. P. Agurok , W. Wang , A. Goldsmith , R. M. Kurtz , S. A. Kupiec , G. D. Savant , and J. L. Jannson , “ HWIL LIDAR imaging sensor, 3D synthetic and natural environment, and temporal ATR ,” in Technologies for Synthetic Environments: Hardware-in-the-Loop Testing VII, R. L. Murrer , ed., Proc. SPIE 4717 , 68 – 76 ( 2002 ). [CrossRef]

], which were concerned with either spatial features, or designed for specific spectroscopy of gas/aerosol targets with selected wavelengths, this work uses nonspecific wavelengths on generic solid, liquid targets to evaluate the technology general features and capability phenomenologically. Image processing and statistical analysis were applied to assess the effectiveness of multi-spectral data for target discrimination.

There are two key features of this work: the system engineering aspect, and the broadband multi-spectral imaging experiments. For system engineering, the paper describes a design using mid-IR semiconductor lasers that can be applied to compact portable systems. A key feature is that the system is truly broad-band with a large spectral spread from 3.3–9.6 μm, as opposed to systems with narrow tunable wavelength ranges. Although the number of wavelengths is still modest, 4 λ’s (which was cost- and not engineering-limited), the large spectral spread helps reveal the spectral diversity of various objects, and also makes the wavelength-dependence optical effects more relevant than those with narrow spectral ranges. Semiconductor lasers are attractive as they offer a wide spectral range, and allow ease of scaling to a large number of wavelengths. They do not have high power for long range, but in recent years, mid-IR semiconductor lasers can offer up to ~1 W [11

11 . J.S. Yu , S. Slivken , A. Evans , J. David , and M. Razeghi , “ Very High Average Power at Room Temperature from λ ~ 5.9 μm Quantum Cascade Lasers ,” Appl. Phys. Lett. , 82 , 3397 – 3399 ( 2003 ). [CrossRef]

], which is sufficient for many short-range (~100’s m) applications. With regard to the economic issue, semiconductor lasers are potentially inexpensive if there is a sizable economy of scale in the long term.

Another feature is a system architecture that is highly suitable and practical for semiconductor lasers. This includes the use of: i) coarse wavelength-division-multiplexing (CWDM) to combine many wavelengths into a single aperture, which avoids parallax artifacts that cause spectral distortion otherwise; ii) the intensity-modulation code-division-multiplexing (CDM) technique to distinguish signals of different wavelengths [12

12 . Z. Morbi , D. B. Ho , H.-W. Ren , H. Q. Le , and S. S. Pei , “ Short-range remote spectral sensor using mid-infrared semiconductor lasers with orthogonal code-division multiplexing approach ,” Opt. Eng. 41 , 2321 – 2337 ( 2002 ). [CrossRef]

,13

13 . Y. Wang , C. Peng , H. Zhang , and H. Q. Le , “ Remote spectral imaging with multi-wavelength and tunable, wavelength-modulation lasers ,” in Proceedings of the Conference on Laser and Electro-Optics 1 , 3 ( 2004 ).

], which helps simplify both hardware and signal processing designs as well as enhance the system robustness against interference and systematic artifacts; and iii) all-digital signal processing.

Most significant are the experimental results, which show substantial spectral discrimination capability for diverse targets even with just 4 generic wavelengths. A number of algorithms were used to produce false-color images for illustration. A specific focus is spectral contrast, which is based on the variation among different wavelengths, as opposed to spatial intensity contrast within an image. This approach is shown most relevant and effective for laser-based imaging. The rest of the paper is organized as follow. Section 2 describes some key issues of multi-spectral laser imaging that are of concern in this work. Section 3 describes the system and experimental approach. Section 4 describes the experimental results and discussion; and Section 5 provides a summary and conclusion.

2. Some key issues with multi-spectral laser stand-off imaging

2.1 The physical basis of laser-based multi-spectral images

For the sake discussion, Fig. 1 illustrates a nominal multi-spectral laser imaging system. A multi-wavelength laser beam illuminates a target scene. A receiver measures the scattered power. The laser beam can illuminate a small spot, and an image is formed by scanning the beam. The beam can also illuminate the entire scene, and a focal plane array receiver captures the entire image. A hybrid approach is also possible. A comparative study of different designs is not the topic of this paper. The interest here is the physical basis of the acquired image.

As illustrated in Fig. 1, multi-spectral laser imaging is a map of the backscattering coefficient vs. beam position on a target, S(λ;r⃗) = PR (λ;r⃗)/PI (λ;r⃗), where PR (λ;r⃗) is the received scattered power from the illuminated location r⃗ on the target, and PI (λ;r⃗) is the incident power; r⃗ can be 2- or 3-dimensional; and λ is the wavelength. In a scanning system, the entire laser power can be assumed to be on one spot, one can write PI (λ;r⃗) = e -α(λ)R PT;λ where PT;λ is the transmitter power, R is the object range, and α(λ) is the transmitting loss coefficient. The atmospheric effects on the laser phase and wavefront are complex [14

14 . See e. g. A. Ishimaru “ Wave Propagation and Scattering in Random Media ,” Academic Press, San Diego, CA ( 1978 ).

], but for simplicity, only the loss effect is included here. Correction for distance fading effect can also be applied: Ŝ (λ;r⃗) = (R/R 0 )2 e 2α(λ)R PR (λ;r⃗)/P T;λ, where R 0 is just a nominal distance. An exception is when α(λ) itself is the signature of interest, e. g. the detection of absorption of a gas cloud.

Fig. 1. Left: block diagram of the multi-spectral laser imaging system. Right: scattering process.

The behavior of the scattering function S(λ;r⃗) is typically complex [15

15 . See e. g., A. A. Kokhanovsky , Optics of Light Scattering Media, 2nd ed. ( Praxis Publishing , 2001 ); T. A. Ger-mer, “Model Integrated Scattering Tool,” http://physics.nist.gov/Divisions/Div844/facilities/MIST/mist.htm

]. The key interest is its λ-dependence, which is a function of both the target dielectric property and the geometrical property. The former is the spectroscopy of the target material, which is the basis for chemical imaging. The latter includes the surface microscopic morphology on the dimension of ~λ, (which often cause speckles), and the macroscopic surface profile on a dimension ≫λ. A more general description is S(λ;r⃗;Ω^ I;Ω^ R;ε̂I,ε̂R) that includes Ω^ I, Ω^ R, which are the laser incident angle and the receiver line-of-sight (LOS) angle relative to the target tangential surface, respectively; and ε̂I , ε̂R are the polarizations of the incident and received scattered waves, respectively. Ω^ I = Ω^ R , then S(λ;r⃗; Ω^ ) is proportional to the bidirectional scattering distribution function. If only Stokes vector is measured, S is proportional to the Mueller tensor. For transmissive targets, the volumetric scattering property, e. g. diffuse reflectance is also important. The only way to determine S(λ;r⃗) correctly is to solve the exact Maxwell equations, an impractical proposition for unknown and complex targets. Most models [15

15 . See e. g., A. A. Kokhanovsky , Optics of Light Scattering Media, 2nd ed. ( Praxis Publishing , 2001 ); T. A. Ger-mer, “Model Integrated Scattering Tool,” http://physics.nist.gov/Divisions/Div844/facilities/MIST/mist.htm

] are statistical approximations of Mie, Rayleigh, and other diffraction theories.

2.2 Issue of geometrical scattering in laser-based multi-spectral imaging

If S(λ;r⃗) can be approximately decomposed into two components, one to represent the mostly spectroscopic signature, denoted as S Spec(λ;r⃗), and the other for the mostly geometrical effects, denoted as Q(λ;r⃗), then both offer an extra dimension for target discrimination. An analogy is the ability to recognize two objects to have the same spectroscopic “color” but different surface morphology, for example. However, because of its complexity, the issue of spectroscopic or geometrical λ-dependence is usually ignored. An approach is to adopt a purely phenomenological and data-driven viewpoint, and apply unsupervised learning algorithms to classify pixels with similar signatures into the same group.

This approach may also be reasonable in systems with very narrow spectral range, as the λ-dependence geometrical effects may be neglected compared with the spectroscopic effects. In general however, large spectral spread is desirable for multi-spectral imaging. In this work, there is a nearly 3X wavelength difference between the two extremes. In such cases, the λ-scaling of geometrical effects can be very pronounced. For a number of common and typical target geometries, various theories produce λ-scaling behavior for Q(λ;r⃗) that can be parametrized as ~∑,cqλ-q where q (not necessary an integer) can range from <-2 to 4 (Rayleigh scattering). Q(λ;r⃗) can have behavior on both extremes: sharp long-λ drop-off and rapid long-λ increase. For example, a common assumption for many objects in remote sensing is q=2, which is the diffraction λ -2 scaling behavior. The λ -2 scaling can be correct at near normal incident for certain statistically incoherent diffractive targets, but incorrect for large angle. Consider for example Fraunhoffer diffraction for a microscopic circular scatterer of radius a, which scales as ~λ -2 (cos2 θR + sin2 θR cos2 ϕR )cosθI [J 1(u)/u]2 where θI,R , ϕI,R are the angles of incident (I) or received LOS (R), and u is a geometrical parameter that is a function of λ, a, and the angles. The λ -2 scaling is correct for u=0, which is normal incident, but shows [J 1(u)/u]2 oscillatory behavior for off-normal u≠0. A rough surface consisting of many phase-incoherent scatterers produces non-trivial λ-q , with q<0 or >0 depending on the incident/LOS angle and the specific statistical distributions of these microscopic structures. An example of opposite λ-scaling is the case that a surface is diffusive at short λ, but specular at long λ, with a highly varying λ-scaling behavior as a function of the viewing angles.

A comprehensive study of the scattering behaviors is not a topic of this paper, but some effects were observed experimentally and an understanding is necessary for the discussion. From a broader perspective, the laser-based technology with total control on laser power and modulation techniques, allows measurement accuracy. Thus, although there is no a priori mean to determine the scattering property of unknown targets, an algorithm can be probabilistic taking into the consideration of these effects. A possible approach is to decompose S(λ;r⃗) into a product of a low- and high-frequency spectral component. The low-frequency component is likely to be Q(λ;r⃗) -dominated, and the high-frequency is likely to be S Spec(λ;r⃗)dominated. This approach can not be meaningfully evaluated here since there are only 4 λ’s. But it is conceivable that with many more wavelengths, e. g. 10–20 λ’s, it is possible to discern the smooth, monotonic Q(λ;r⃗) from both random speckles and the more varying S Spec(λ r⃗) to achieve a more accurate target discrimination and classification.

More generally, the laser-based technology can adopt capabilities to help develop physics-based, rather than purely phenomenological image processing algorithms. For example, measurements of polarimetric signature can provide some information on the surface property that can be used for scattering correction. It will be evidenced in some experiments that object “shininess” can be detected, from which some surface roughness model can be assumed. This in turn can be used for “best guess”, e. g. Bayesian inference of some forms of Q(λ;r⃗) and S Spec(λ r⃗) for object classifications and perhaps with feedback learning.

2.3 Distinction of multi-spectral image from single-λimage: spectral vs. intensity contrast

Multi-spectral laser imaging is not the collection of many single-λ imaging with different wavelengths. A set of 2 single-λ images {S(λ 1;r⃗), S(λ 2;r⃗)} do not constitute a multi-spectral image unless a radiometric factor between them is known. Thus, a multi-spectral image is a set of spectra {S({λm }m=1N;rp)} mapped to pixel location rp , but not a set of intensity images at different λ’s. Mathematically, the data is organized as a set {S({λm }m=1N;r1), S({λm }m=1N;r2), ⋯S({λm }m=1N;rp)} as opposed to the arrangement {S(λ 1;{rk}k=1P), S(λ 2;{rk}k=1P), ⋯S(λ N;{rk}k=1P}. If each spatial image is individually normalized, the relative radiometric information and spectral contrast between the pixels would be distorted. Two pixels can have similar color but different intensities, i. e. S({λm }m=1N;r1)~η S({λm }m=1N;r2) where η≠1 indicates different intensities. Vice versa, they can have comparable total intensity, i.e., m=1NS(λm;r1)~m=1NS(λm;r2) but different colors, i. e. ∥S r1({λm }m=1N)-η S r2({λm }m=1N)∥≠0 for all η; the spectral distance represented by symbol ∥⋯∥ is an algorithm-dependent concept that will be discussed later in Section 4. The preferred multi-spectral imaging approach in this work emphasizes on spectral contrast, and preserves the radiometric information among the different λ’s, as opposed to algorithms that emphasizes only on the intensity contrast of each spectral image, and may lose the spectral information.

2.4 Analysis of spectral resolution and diversity

With only four wavelengths, an issue is how much spectral information this system can provide. The interest here is not a theoretical assessment based on information theory, but to adopt a statistical measure to evaluate the experimental results. Clearly, one can choose targets with the widest range of spectral diversity at the available wavelengths. But this is not the objective; the work aimed to be an experimental phenomenological demonstration. From the information perspective, a measure for the spectral diversity of an image is:

HS=rPrlog2(Pr),
(1)

which is the entropy of a spectral distribution and serves as an estimate for the spectral information content in bit; pr is the probability density of a given spectral measure. This will be used in Section 4.

3. System and experimental approach

3.1 Multispectral remote imaging scanning system and data acquisition

The multi-spectral imaging system is diagrammed in Fig. 1. Some aspects of the system have been described elsewhere [8

8 . See e. g., IEEE Signal Processing Magazine 19 , No 1 Jan ( 2002 ).

,10

10 . Y. Wang , Y. W., C. Peng , H. Zhang , A. Seetheraman , and H. Q. Le , “ Concepts for Scalable, CDMA-Networked, M/LWIR Semiconductor Laser Standoff Chemical Detection System ,” in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas , Eds., Proc. SPIE 5617 , 179 – 189 ( 2004 ). [CrossRef]

]. The X-Y scanning system is primarily for 2-D imaging, although it also has the 3-D capability using a near-IR diode laser as described in Section 4. The transmitter consists of a coarse wavelength-division-multiplexed unit of up to 4 commercial lasers, as described in Table 1. The CWDM used band-pass filters with typical transmission losses ≤2 dB on pass bands and reflection losses ≤1.5 dB on reflection bands. Each laser beam was adjusted for comparable beam size before WDM combination. All are diffraction-limited from single-mode waveguides except for the multi-mode 3.3-μm laser with a beam quality M 2 ~20. The single-mode waveguide lasers also have small profile differences.

Table 1. MIR semiconductor laser characteristics

table-icon
View This Table

Alignment for optimal beam overlap was critical to the spectral fidelity of the multi-spectral images. Non-overlapping beams have parallax errors that cause spectral distortion. The alignment was achieved by optimizing the far-field power of all beams in a common adjustable bucket. As the beams have profile variation, the overlapping criterion was bases on the beam centroid. The misalignment is <50 μrad, which is ≪ beam divergence of ~0.5 mrad.

The multi-spectral MIR beam was coupled into an X-Y galvanometer scanner, capable of up to 150-Hz scanning rate. The effective scanner aperture is 2 cm. The receiver consists of a number of plug-and-play configurations. Receiver optics options include a 6-cm-diameter f/1 lens for targets with strong return, or a modified 25-cm-Dobsonian reflector for weaker signal. The IR detector options included an InSb detector for 3–5 μm, and a MCT detector for 3–12 μm. In some experiments, the InSb was used for short wavelengths and the MCT was used for long wavelengths. The system field-of-view was limited by the detector size, and was ~38×38 mrad for 3-mm-diameter detectors used in most experiments. The set up also has two IR focal plane array (FPA) cameras, a PtSi for 3–5 μm, and a pyroelectric FPA for 2–14 μm.

The main element of data acquisition consists of a 4-channel analog-to-digital converter (ADC) capable of 100 MS/s. A first-in-first-out memory buffer was used between a 160-MHz DSP board (Motorola DSP56L307) and the ADC as cache to ensure fast data acquisition. The digital data is pre-processed in DSP, including filtering, noise reduction, and demodulation. The multi-spectral images are generated and processed by a computer.

3.2 Operation

The transmitters and receivers are modulated/demodulated using intensity-modulation direct-sequence spread spectrum (IM-DSSS) CDM technique. Some description of this CDM technique for stand-off sensing can be found elsewhere [8

8 . See e. g., IEEE Signal Processing Magazine 19 , No 1 Jan ( 2002 ).

,10

10 . Y. Wang , Y. W., C. Peng , H. Zhang , A. Seetheraman , and H. Q. Le , “ Concepts for Scalable, CDMA-Networked, M/LWIR Semiconductor Laser Standoff Chemical Detection System ,” in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas , Eds., Proc. SPIE 5617 , 179 – 189 ( 2004 ). [CrossRef]

,12

12 . Z. Morbi , D. B. Ho , H.-W. Ren , H. Q. Le , and S. S. Pei , “ Short-range remote spectral sensor using mid-infrared semiconductor lasers with orthogonal code-division multiplexing approach ,” Opt. Eng. 41 , 2321 – 2337 ( 2002 ). [CrossRef]

,13

13 . Y. Wang , C. Peng , H. Zhang , and H. Q. Le , “ Remote spectral imaging with multi-wavelength and tunable, wavelength-modulation lasers ,” in Proceedings of the Conference on Laser and Electro-Optics 1 , 3 ( 2004 ).

]. Each laser has a unique code, and the receiver can decode the signals of all wavelengths simultaneously. Accurate real-time ratio-ing of received and transmitted power for each λ was critical. Alternative approaches that use tunable lasers or a non-laser broad-band source with a tunable band-pass filter must be concerned with sufficiently rapid switching between different wavelengths to avoid time-dependent spectrum distortion. Although this system has only one transmitter (Tx) and one receiver (Rx), a general system may have several spatially distributed Tx’s and Rx’s to form a multi-point imaging network [10

10 . Y. Wang , Y. W., C. Peng , H. Zhang , A. Seetheraman , and H. Q. Le , “ Concepts for Scalable, CDMA-Networked, M/LWIR Semiconductor Laser Standoff Chemical Detection System ,” in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas , Eds., Proc. SPIE 5617 , 179 – 189 ( 2004 ). [CrossRef]

]. The CDM architecture, similarly in an optical wireless communication system, allows the ease of network scalability and management.

System noise is a key figure-of-merit. The noises are incurred in the analog parts of the system, including laser RIN, scanner mechanical noise, detector noise, transimpedance amplifier (TIA) noise, and ADC noise. The typical laser pulsed RIN power spectral density (PSD) is -100 dB/Hz from 10Hz-20 kHz. The calculated noise-equivalent power (NEP) of laser RIN was typically 5×1011W/Hzfor strong signals, and much less for most signals. This was comparable or smaller than the detector NEP, which are 3×1012Hz the InSb and 3.3×1010W/Hz for the MCT. The TIA noise is ~1×1011W/Hz. Other noises were relatively smaller than the detector or laser noises and can be ignored. Both measured laser and detector noises showed white Gaussian noise distribution. The typical laser energy used to obtain a single spectral image in this study is around 6.4 μJ to 17 μJ. The image resolution and acquisition rate were limited only by the laser power.

4. Experimental results of MIR multi-spectral imaging

4.1 Overall description

The experiments are aimed to be phenomenological rather systematic by selecting targets based on familiarity rather than spectroscopy. The system can generate a large number of images, but the paper aims only to expose the basic features of this technology by discussing a few cases that represent both challenges and opportunities. One is selected for detailed discussion. The results in this section include: i) targets of common materials, ii) chemical contamination, iii) miscellaneous household, natural objects, and foods, iv) target at longer distance, and v) multi-spectral and 3-dimentional (3-D) image combination. Most measurements were obtained with ~mW laser average power at ~13-m distance, except for one at 40 m.

For each target scene, multi-spectral images were acquired, processed, and analyzed. The processing aims to produce false color images (FCI) for human perception and discussion. However, they are not the true output; rather the information content that can be used for automatic object discrimination and classification is the main output. An important aspect of the analysis is to evaluate the spectral information capability of the system. The analysis methodology includes comparison of the object spectral signatures in the images with their known FTIR spectra, statistical analysis of the spectral contrast, and comparison with IR focal plane array images when applicable. Most images are 2-D for simplicity, but an example of 3D multi-spectral imaging is also included to show the system more general capability.

4.2 Experimental results: multispectral imaging of a variety of targets

4.2.1 Target of common materials

This example is selected for a detailed exposition of the analysis approach.

Fig. 2. (a) Visible image of material targets taped on a wall at 13 m away; (b) MIR raw spectral images at 3.3, 4.9, 7.2, and 9.65 μm plotted individually; (c) IR FPA camera images of the targets 2 m away.

4.2.1.1 MIR multi-spectral images

Figure 2 shows: (a) a visible image of a number of common materials as indicated, which were taped to an indoor wall; (b) the 200×200 pixel spectral images at 3.3, 4.9, 7.2 and 9.65 μm; and (c) IR FPA camera images for reference, which were taken from 2 m away with thermal illumination in addition to the ambient light. The 3.3-μm image appears more blurred horizontally than the others since the beam is ~20 times diffraction limited and has a large spread in that dimension. Each spectral intensity image was plotted with its own contrast adjustment, thus the absolute spectroscopic information is lost and the spectral differences between objects are not obvious. Their multi-spectral image processing will be discussed in Section 4.2.1.3.

This result shows examples of the geometrical scattering effects, as evidenced with pronounced speckles in the long-λ images. As mentioned in Section 2.2, the wall exhibits diffusive scattering behavior for short-λ and specular reflection for long-λ. Some other dark targets exhibit an opposite long-λ drop-off behavior, but not noticeable on this intensity scale. The electrical tape (#3) also exhibits an opposite behavior with strong specular-like reflection at 3.3 μm but moderate at long λ, possibly because of absorption, which serves as an example of combined geometrical scattering and spectroscopic effects. Detailed analysis of these features however would be beyond the length of this paper.

4.2.1.2 FTIR spectra of targets

A relevant question is whether these targets have meaningful spectral signatures at the 4 wavelengths employed. Figures 3 show the FTIR spectra at 45° reflection of the various materials in Fig. 2. Although this angle is different from the near-normal angle in laser measurements, the spectra should provide a reasonable correlation. The four vertical lines in the figures indicate the 4 laser wavelengths. The circles mark the expected spectral response at those wavelengths for some materials. As shown, most targets have little spectral features at those wavelengths. Some have significant signatures in the 8–12 μm fingerprint region, but with only the 9.6 μm available in this region, most targets appear to have a similar, nearly “flat” response at the 4 wavelengths. There are total intensity differences, which would explain different levels of darkness (or brightness). The results in Figs. 2 show the challenge for the laser measurements, which clearly must have sufficient radiometric accuracy and signal-to-noise to discern the minute spectral differences at these 4 wavelengths.

Fig. 3. FTIR spectra of some of the target materials used in Fig. 2. The vertical lines mark the laser wavelengths used in spectral imaging. Some dissimilar materials per chance have very similar spectra at the sampling wavelengths.

4.2.1.3 Image processing with spectral contrast

Although image processing can significantly clean up and improve the spatial features in Fig. 2(b), such as speckle filtering, spatial smoothing and deconvolution, edge sharpening,… they are not employed since this work focuses only on the spectral aspect. All spatial features are left as-is. As mentioned above, target discrimination and classification are strictly based on the statistical property of the spectral features. However, it is useful for human perception to combine the spectral images in Fig. 2(b) into false color images. This is particularly relevant in real-time systems designed directly for humans (i. e. the human operator is in the loop of the target recognition process). The key criterion is that it should not produce artifacts, i. e. presenting more information than a preferred statistical classifier of raw data permits.

There are a number of algorithms to generate FCI. The results are displayed in Figs. 4. Figure 4(b) shows a FCI generated from a RGB combination of any 3 λ-images in Fig. 2(b). With arbitrary coefficients, the resulting FCI loses initial spectral information. Figure 4(c) shows the RGB combination using a PCT (principal component transform) algorithm [16

16 . A. Mackiewicz and W. Ratajczak , “ Principal Components Analysis (PCA)”, Computers & Geosciences 19 , 303 – 342 ( 1993 ).

]. The PCT analysis applied to individual spectral images shows two components with strong S/N ratios, a third with a fair S/N, and the last with mostly noise. Both intensity contrast adjustment and RGB combination were applied to the 3 orthogonal PCT components for best “human perception.” Another algorithm, ISODATA (Iterative Self-Organizing Data Analysis Technique Algorithm), which is based on unsupervised classification [17

17 . J. R. Jensen , Introductory Digital Image Processing: A Remote Sensing Perspective, 2 nd ed. ( Prentice Hall, Inc . 1996 ).

] produces the FCI in Fig. 4(d). Lastly, Fig. 4(e) shows the FCI from a spectral contrast algorithm used in this work; the details are given in the Appendix.

All FCI’s show the some degrees of spectral discrimination. All glass and quartz pieces have the same color as expected, and are most distinctive owing to the 9.65-μm signature. CaF2 appears transparent (same color as the wall) as expected since it has no absorption. For the rest, different algorithms show different degrees of discrimination. The plain RGB and PCT FCI in Figs. 4(b) and 4(c) do not show strong discrimination for some materials, which appears black because the spectral information is lost in the process of individual intensity adjustment of the spectral image that is uncorrelated with each other. The ISODATA image however has the opposite problem of over-discrimination as it even makes distinction of different speckle patterns on the wall. While this is mathematically rational, it serves as an example of purely phenomenological approaches that ignore physical meaning. The spectral contrast image 4(e) mathematically makes 4 other discriminations: 1: electrical tape (red-purple); 2: packing foam, asphalt-2, and plexiglass (yellow-greenish); 3: polymer-1 (blue-purplish); and 4: polymer-2 (pink) with a film-stress shadow (dark) not detected in the visible spectrum. Lastly, the Scotch tape pieces appear as blue stripes, and the wall appears as a single color (green). Different color specks in the rest (cardboard and asphalt-1) represent individual pixel spectral noise; no smoothing or noise reduction was applied.

Fig. 4. (a) A visible image of the target. (b-d) False color images from the same IR spectral images with different phenomenological algorithms (see text). The algorithms for (b) and (c) lose or distort some spectral information, and some objects become dark. The algorithm for (d) over-classifies and makes more color distinction than physically meaningful. The algorithm for (e) is designed to preserve laser spectral data in lieu of intensity. The resulting false color image has reasonable correlation with the object IR spectra shown in Fig. 3. Notice that colorless objects (black or transparent) in (a) have “colorful” mid-IR signatures in (e).

This result should be considered in the context of spectra in Figs. 3. Various materials happen to have very similar spectra at the 4 wavelengths, and thus should have the same color as expected. This shows the consistency and reliability of the technology. A few more wavelengths between 3-5 and 8-10 μm can make further spectral discrimination.

4.2.1.4 Discussion on spectral image processing

It is clear from Fig. 4 that algorithm is an integral part of laser multi-spectral imaging. As discussed in Section 2.3, an algorithm can be spatial- or spectral-centric. The FTIR spectra in Fig.3 show the minute spectral differences between the materials at the 4 wavelengths, indicating that spectral, not spatial features are critical. Spatial-centric algorithms that process each normalized intensity image on its own may lose critical spectral information. The spectral contrast algorithm used for Fig. 4(e) produced results that are consistent to the physics reality.

Mathematically, the distinctive colors obtained from the spectral contrast algorithm represent the distinguishable modes of spectral-difference distribution. This algorithm is purely spectral, ignoring spatial contiguousness and does not use any spatial filtering. Consequently, the FCI suffers some “color noise” as shown on the dark targets. However, this should be considered in the context that these material targets are nearly black, i. e. they have very weak signals. The fact that some colors were obtained underscores the system spectral sensitivity and accuracy as well as the algorithm efficiency. If spatial correlation is used to allow many-pixel color averaging, the resulting color would have a much better S/N (which scales as 1/√P, P = pixel number), and even the noisiest targets, asphalt-2 and cardboard would be distinguishable. (Notice that even being noisy, they are distinguishable from the wall background). More generally, although it is outside the scope of this paper, higher level algorithms can combine both spectral and spatial features to achieve object classification and recognition.

From the information theory point-of-view, a useful metric of the system capability to measure spectral diversity for a given image is the spectral information content (intensity is not of interest). This can be done by evaluating the entropy in Eq. (1), with pr being the empirical probability density of normalized spectra {λm}, m=1,..N. But a simpler measure is the entropy of the spectral amplitude difference only, not the full spectrum, which is defined as:

S˜({λm}m=1N;ri)S˜({λm}m=1N;rj)σT[mwmS˜(λm;ri)S˜(λm;rj)γ]1/γNoise(numerator)forij
(2)

where wm is a weight function, which was chosen uniform, and γwas chosen = 2 in this case. This quantity makes no distinction between colors but only their distance, and is a more relevant metric for color range and resolution, and less dependent on the number of λ. The division by the expected noise σT means that any spectral contrast less than the noise are treated as meaningless. A noisier laser-based system would have a worse spectral resolution. The resulting distributions of a random pixel population from images in Fig. 2(b) with different number of λ’s are shown in Fig. 5. Log-normal appears to be the best fit. A purely white-noise spectrum would have Chi-distribution p(s;n)= [21-n/2/Γ(n/2)] s n-1 exp(-s 2/2) for color distance s, n is the degree of freedom. Clearly, at low n or low number of wavelengths, the spectral-distance distribution is not Chi, i. e. not purely random. As more wavelengths are used, the spectral diversity among objects is higher, and the distribution becomes similar to Chi, indicating that the system has no systematic bias.

Fig. 5. Distributions of spectral distance defined in Eq. (2) of a random population of pixels in the spectral images in Fig. 3(b). (a)-(c): increasing the number of wavelengths used in the statistical analysis. The ~7.3-bit entropy is a measure of the “color” amplitude dynamic range and resolution (not color diversity) of these images. It should be taken in the context of how little spectral differences there are among the target objects in Fig. 3. Comparison with log-normal and chi-distribution are also shown.

4.2.2 Chemical contamination

Figures 6 and 7 illustrate the cases of chemical contamination. Figure 6 shows dry sand with patches of oil and water contamination. For the visible image in Fig. 6(a), the contamination appears as dark patches, but the distinction is based on intensity, not color as the RGB decompositions in Fig. 6(b) of the three marked spots for oil, water, and sand appear nearly the same. Their IR spectra in Fig. 6(d) are truly different, which reflect in the markedly different colors in the FCI of Fig. 6(c), suggesting different chemicals.

In another example, Fig. 7(a) shows the visible image of an aluminum plate contaminated with four different oils, two are bio-organic and two are petroleum hydrocarbons. The hydrocarbons have distinctive C-H signatures around 3.4 μm. The oils were allowed to run off under gravity, and the adsorbed film thicknesses were estimated to be ≪100 μm. The IR multi-spectral FCI at 13 m away is shown in Fig. 7(b). The oil patches appear as green/blue, and the metal appears red/yellow. This case is also an example of the scattering problem. The Al plate had highly non-uniform scratches that caused extremely irregular speckle patterns. It had mirror-like reflection at long λ’s that overwhelmed the receiver dynamic range, rendering invisible all other features, including the normally bright wall background. The spectral distinction between the two types of oil was not pronounced; nevertheless the real significance is that these thin films were detectable under highly difficult scattering problem. More advanced system with auto gain control and a large dynamic range may solve this problem.

Fig. 6. (a): Visible image of sand contaminated with oil and water. (b) Colorimetric decomposition shows that the difference between the three marked square spots is not spectral (color) but only intensity (brightness). (c): The MIR multi-spectral false color image makes clear spectral discrimination, and not just intensity discrimination between the spots. The reason is (d): they have distinctive MIR spectra.
Fig. 7. Left: visible image of an aluminum plate contaminated with 4 thin-film stripes of oils. Right: the multi-spectral MIR false-color image showing the oil films as green/blue, the metal as red/yellow. Petrochemical cutting fluid displays a bluish hue that is statistically distinguishable from organic oils.

4.2.3 Miscellaneous other objects

Soil, vegetation, minerals, etc. are typical outdoor natural targets; household/industrial objects are typical indoor targets. A number of multi-spectral images for these types of target are shown in Fig. 8. Figure 8(a) shows a soil collection; the FCI shows distinction among various types. Figures 8(b) and (c) show sand, humus soil, and leaves. It is an example about measurement condition. The FCI in Fig. 8(b) was the result of low signal levels with little returns from the leaves, which appear dark except for a spot with specular reflection. There were little spectral signatures. The FCI in Fig. 8(c) was for the same objects (with a slight rearrangement), but the signal level was stronger and the beam incident angle was apparently different. A remarkable result is not the fact that the leaves have definitive color (which is coincidentally green), but that a part of a leave that barely appears yellowish in the visible becomes pronounced in the IR FCI. A similar result was also observed in Fig. 8(d), in which dried leaves appear as light green, compared with black for green leaves. Various objects appear “shiny,” which caused color degradation of dark objects because of the dynamic range problem as discussed above for the target in Fig. 7.

Figure 9 shows the result for a variety of food. The 9.6-μm laser, which is the most important fingerprint wavelength, degraded during the course of the experiments, resulting in noisy signals for normally already-low signals from food objects. The FCI in 9(b) has maximum spectral-contrast enhancement showing color noises for some vegetables, although distinction between hams and vegetables was possible. In the FCI of Fig. 9(c), the spectral contrast was modified with intensity, where color distinction between cardboard (yellow), concrete (pink-gray), bok choy (blue), and the back screen (green) is at a more typical S/N level.

Fig. 8. (a) Mineral collection; top: visible image; bottom: multi-spectral false-color image (FCI). Sand (quartz) is red, humus soil and woods are brownish/dark green, asphalts are bluish. The beam was ~2.5 cm, and > most pebbles. (b) Sandy soil, humus soil, and leaves. Top: visible; bottom: IR FCI. Sandy and humus soils have different colors. (c) FCI of the same target in (b) under a slightly different arrangement. The green false color of the leaves was coincidental. A barely discernible yellowish spot of a leave became very pronounced in the IR FCI. (d) Household objects. Dried leaves are distinctive from green leaves (black because of weak signals). A piece of wood appears yellow; and concrete appears gray. Other shiny objects with specular reflection cause the dynamic range problem as spectra of weak signals (dark region) are lost.
Fig. 9. (a) Visible image of a variety of food. (b) and (c) False color images (FCI) with slightly different processing algorithms. In (b), a spectral-enhanced IR algorithm discards intensity data, showing some distinction between vegetables and non-vegetables (ham, bread). Shadows on background screen were evidenced. The spectral noise was a result of weak signals. (c) The FCI from an algorithm with intensity included, showing cardboard and concrete distinction, but food appear dark because of low signals.

4.2.4 Stand-off multi-spectral imaging for a target at 40 m

Fig. 10. (a) Visible image of a scene from 40 m away (~1/2 beam Rayleigh range). (b) False color image (FCI) generated from the 3.3- and 4.9-μm wavelength images in (e) and (f). (c) and (d): Passive infrared focal plane array camera images. The FCI in (b) shows some spectral discrimination (wall, cardboard box, and plastic container) even with only 2 wavelengths.

4.2.5 Three-dimensional multi-spectral laser imaging

Laser ranging can be added to this system for 3-D multi-spectral imaging. A 25-W-peak, 0.87-μm laser diode was coupled to the X-Y scanner, and a 2-ns rise-time PIN Si diode was used for the receiver. The range finder was optically time-shared with the MIR multi-spectral imager, since the system did not have a near-IR/MIR coupler. All else was the same including the CDM electronics. Range was determined from the time-of-flight from the pseudo-noise code autocorrelation. The spatial resolution was ≤3 cm for uniformly flat object. Figure 11 shows a result of 3-D multi-spectral laser imaging. Figure 11(a) shows the arrangement of three objects of different materials at various distances. The ranging image is shown in Fig. 11(b). The MIR multi-spectral FCI is shown in Fig. 11(c), in which the foam appears red, the cardboard is brown/dark green, and anodized aluminum optical plate is blue-green. The superposition of both images is shown in Fig. 11(d). The 3-D ranging image shows a slope-off error at the edge the front object because of the diode laser large beam spread in the vertical direction (its low-beam-quality dimension) and the beam simultaneously struck both the front and rear objects. Nevertheless, Fig. 11 (d) clearly shows the system capability to resolve the target both 3D spatially and spectrally.

Fig. 11. (a) Visible image showing an arrangement of three objects, a foam cushion, a cardboard box, and an anodized aluminum optical breadboard. (b) Laser ranging 3D image. (c) False color image (FCI) from MIR multi-spectral images. (d) 3-D image combination of range and FCI. The slope-off at the foam upper edge was due to the large beam spread that struck both the foam and the breadboard.

5. Summary and conclusion

This paper describes a multi-spectral mid-IR laser imaging study including system engineering, experiments, image processing and analysis. Even with just 4 wavelengths, which are quite modest compared with passive hyperspectral imaging, the results for a number of common objects and materials indicate significant potential for target discrimination. Unlike spectroscopic lidar works, neither the 4 wavelengths nor the diverse targets here were possibly designed to be optimal for each others for spectral discrimination. Yet from very bright (specular scattering) to nearly “black” targets, the system was able to distinguish minute spectral differences. Various image processing algorithms are used, and the study shows an algorithm, which emphasizes on spectral contrast is particularly effective for the laser-based system, whose capability for accurate and sensitive measurements is essential to the spectral discrimination capability. The system here can still be improved with larger signal dynamic range. The design and architecture is highly suitable for semiconductor-laser-based compact systems. A logical target-specific design [18

18 . R. Hardie , M. Vaidyanathan , and P. F. McManamon , “ Spectral band selection and classifier design for a multis-pectral imaging laser radar ,” Opt. Eng. 37 752 – 762 ( 1998 ). [CrossRef]

] with more selected laser lines, e. g. 10–20, and higher power, e. g. ~ 1 W, can be very a powerful tool for stand-off imaging sensing at the range of ~ 100’s m.

This work was supported in part by the DARPA/ARO under contract DAAD-190010361, the US DARPA/AFRL under contract F33615-02-C-1139, the Air Force Research Lab under contract F29601-00-2-0058, and the State of Texas through the THECB and the Texas Center for Superconductivity and Advanced Materials. We thank C. Peng and A. Seetharaman for assistance.

Appendix: Spectral contrast algorithm for false color imaging

Each spectrum is separated into color (normalized spectrum) and intensity as discussed in Section 2.3. Without any processing, raw colors can be classified using some discriminant function operating on their statistical distribution p[{λm};m = 1,⋯ N]. However, for false color image generation, some arbitrary combination of RGB (or similar human color perception scheme) is needed. A spectral contrast algorithm is used to emphasize the spectral differences between pixels as mentioned in Sections 2.3 and 4.2, which involves the subtraction of an average background:

Sa˜({λm}m=1N;rp)=S˜({λm}m=1N;rp)S˜({λm}m=1N;rp)
(A.1)

to yield a “difference spectrum.” An offset q rp is added to make a positive, and which is again normalized:

Sb˜({λm}m=1N;rp)=Sa˜({λm}m=1N;rp)+qrp/m(Sa˜({λm}m=1N;rp)+qrp)
(A.2)

The spectral images b({λm }m=1N;rp) can be RGB-combined using the same approach commonly used, which is to find three linear combinations of all spectra with the criterion that it must yield the number of hues for human perception that are consistent with some preferred statistical classification:

SVisible(R,G,B;rp)=CSb˜({λm}m=1N;rp);
(A.3)

where C is a function, which can be as simple as a linear matrix multiplication mapping an N-dimensional vector into a RGB vector. The original all-spectra intensity image can also be combined to modify the absolute RGB image:

SFalseColor(R,G,B;rp)=fA(rp)SVisible(R,G,B;rp)
(A.4)

where A(rp) is the original pixel amplitude,f can be any intensity function, e. g. A(rp) itself, or a sigmoid of A(rp), to allow distinction of pixels of similar hue but different intensity. It should be noted that this algorithm is sufficiently simple to implement with FPGA or DSP for real-time display in systems designed directly for human-in-the-loop assisted target recognition as mentioned in Section 4.2.

References and links

1 .

See e. g. J. B. Campbell , Introduction to Remote Sensing , 2nd ed. ( Guilford Press. 1996 ), Chap. 14.

2 .

T. P. Jannson , P. I. Shnitser , A. A. Kostrzewski , I. P. Agurok , W. Wang , A. Goldsmith , R. M. Kurtz , S. A. Kupiec , G. D. Savant , and J. L. Jannson , “ HWIL LIDAR imaging sensor, 3D synthetic and natural environment, and temporal ATR ,” in Technologies for Synthetic Environments: Hardware-in-the-Loop Testing VII, R. L. Murrer , ed., Proc. SPIE 4717 , 68 – 76 ( 2002 ). [CrossRef]

3 .

J. Massa , G. Buller , A. Walker , G. Smith , S. Cova , M. Umasuthan , and A. Wallace , “ Optical design and evaluation of a three-dimensional imaging and ranging system based on time-correlated single-photon counting ,” Appl. Opt. 41 , 1063 – 1070 ( 2002 ). [CrossRef] [PubMed]

4 .

A. D. Gleckler , A. Gelbart , and J. M. Bowden , “ Multispectral and hyperspectral 3D imaging lidar based upon the multiple slit streak tube imaging lidar ,” in Laser Radar Technology and Applications VI, G. W. Kamerman , ed., Proc. SPIE 4377 , 32 – 335 ( 2001 ).

5 .

C. R. Swim , “ Review of active chem-bio sensing ,” in Chemical and Biological Sensing V, P. J. Gardner , ed., Proc. SPIE 5 416 , 178 – 185 ( 2004 ). [CrossRef]

6 .

A. Achey , J. Bufton , J. Dawson , W. Huang , S. Lee , N. Mehta , and C. R. Prasad , “ An enhanced multiwavelength ultraviolet biological trigger lidar ,” in Optically Based Biological and Chemical Sensing for Defence, J. C. Car-rano and A. Zukauskas , eds., Proc. SPIE 5617 , 87 – 91 ( 2004 ). [CrossRef]

7 .

J. R. Roadcap , P. D. Dao , and P. J. McNicholl , “ Case study of multiple-wavelength lidar backscatter from aerosols ,” in Laser Systems Technology, W. E. Thompson and P. H. Merritt , eds., Proc. SPIE 5087 , 156 – 166 ( 2003 ). [CrossRef]

8 .

See e. g., IEEE Signal Processing Magazine 19 , No 1 Jan ( 2002 ).

9 .

Y. Wang , C. Peng , H. Zhang , and H. Q. Le , “ Wavelength modulation imaging with tunable mid-infrared semiconductor laser: spectroscopic and geometrical effects ,” Opt. Express 12 , 5243 – 5257 ( 2004 ), http://www.opticsexpress.org/abstract.cfm?URI=OPEX-12-21-5243. [CrossRef] [PubMed]

10 .

Y. Wang , Y. W., C. Peng , H. Zhang , A. Seetheraman , and H. Q. Le , “ Concepts for Scalable, CDMA-Networked, M/LWIR Semiconductor Laser Standoff Chemical Detection System ,” in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas , Eds., Proc. SPIE 5617 , 179 – 189 ( 2004 ). [CrossRef]

11 .

J.S. Yu , S. Slivken , A. Evans , J. David , and M. Razeghi , “ Very High Average Power at Room Temperature from λ ~ 5.9 μm Quantum Cascade Lasers ,” Appl. Phys. Lett. , 82 , 3397 – 3399 ( 2003 ). [CrossRef]

12 .

Z. Morbi , D. B. Ho , H.-W. Ren , H. Q. Le , and S. S. Pei , “ Short-range remote spectral sensor using mid-infrared semiconductor lasers with orthogonal code-division multiplexing approach ,” Opt. Eng. 41 , 2321 – 2337 ( 2002 ). [CrossRef]

13 .

Y. Wang , C. Peng , H. Zhang , and H. Q. Le , “ Remote spectral imaging with multi-wavelength and tunable, wavelength-modulation lasers ,” in Proceedings of the Conference on Laser and Electro-Optics 1 , 3 ( 2004 ).

14 .

See e. g. A. Ishimaru “ Wave Propagation and Scattering in Random Media ,” Academic Press, San Diego, CA ( 1978 ).

15 .

See e. g., A. A. Kokhanovsky , Optics of Light Scattering Media, 2nd ed. ( Praxis Publishing , 2001 ); T. A. Ger-mer, “Model Integrated Scattering Tool,” http://physics.nist.gov/Divisions/Div844/facilities/MIST/mist.htm

16 .

A. Mackiewicz and W. Ratajczak , “ Principal Components Analysis (PCA)”, Computers & Geosciences 19 , 303 – 342 ( 1993 ).

17 .

J. R. Jensen , Introductory Digital Image Processing: A Remote Sensing Perspective, 2 nd ed. ( Prentice Hall, Inc . 1996 ).

18 .

R. Hardie , M. Vaidyanathan , and P. F. McManamon , “ Spectral band selection and classifier design for a multis-pectral imaging laser radar ,” Opt. Eng. 37 752 – 762 ( 1998 ). [CrossRef]

OCIS Codes
(100.0100) Image processing : Image processing
(110.0110) Imaging systems : Imaging systems
(110.3080) Imaging systems : Infrared imaging
(120.0280) Instrumentation, measurement, and metrology : Remote sensing and sensors
(140.3070) Lasers and laser optics : Infrared and far-infrared lasers
(140.5960) Lasers and laser optics : Semiconductor lasers
(150.6910) Machine vision : Three-dimensional sensing
(280.3420) Remote sensing and sensors : Laser sensors
(280.3640) Remote sensing and sensors : Lidar
(300.6340) Spectroscopy : Spectroscopy, infrared
(330.6180) Vision, color, and visual optics : Spectral discrimination

ToC Category:
Research Papers

History
Original Manuscript: June 6, 2005
Revised Manuscript: August 12, 2005
Published: August 22, 2005

Citation
Yi Wang, Yang Wang, and Han Le, "Multi-spectral mid-infrared laser stand-off imaging," Opt. Express 13, 6572-6586 (2005)
http://www.opticsinfobase.org/oe/abstract.cfm?URI=oe-13-17-6572


Sort:  Journal  |  Reset  

References

  1. See e. g. J. B. Campbell, Introduction to Remote Sensing, 2nd ed. (Guilford Press. 1996), Chap. 14.
  2. T. P. Jannson, P. I. Shnitser, A. A. Kostrzewski, I. P. Agurok, W. Wang, A. Goldsmith, R. M. Kurtz, S. A. Kupiec, G. D. Savant, and J. L. Jannson, �??HWIL LIDAR imaging sensor, 3D synthetic and natural environment, and temporal ATR,�?? in Technologies for Synthetic Environments: Hardware-in-the-Loop Testing VII, R. L. Murrer, ed., Proc. SPIE 4717, 68-76 (2002). [CrossRef]
  3. J. Massa, G. Buller, A. Walker, G. Smith, S. Cova, M. Umasuthan, and A. Wallace, "Optical design and evaluation of a three-dimensional imaging and ranging system based on time-correlated single-photon counting," Appl. Opt. 41, 1063-1070 (2002). [CrossRef] [PubMed]
  4. A. D. Gleckler, A. Gelbart and J. M. Bowden, �??Multispectral and hyperspectral 3D imaging lidar based upon the multiple slit streak tube imaging lidar,�?? in Laser Radar Technology and Applications VI, G. W. Kamerman, ed., Proc. SPIE 4377, 32-335 (2001).
  5. C. R. Swim, �??Review of active chem-bio sensing,�?? in Chemical and Biological Sensing V, P. J. Gardner, ed., Proc. SPIE 5416, 178-185 (2004). [CrossRef]
  6. A. Achey, J. Bufton, J. Dawson, W. Huang, S. Lee, N. Mehta, and C. R. Prasad, �??An enhanced multiwavelength ultraviolet biological trigger lidar,�?? in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas, eds., Proc. SPIE 5617, 87-91 (2004). [CrossRef]
  7. J. R. Roadcap, P. D. Dao, P. J. McNicholl, �??Case study of multiple-wavelength lidar backscatter from aerosols,�?? in Laser Systems Technology, W. E. Thompson and P. H. Merritt, eds., Proc. SPIE 5087, 156-166 (2003). [CrossRef]
  8. See e. g., IEEE Signal Processing Magazine 19, No 1 Jan (2002).
  9. Y. Wang, C. Peng, H. Zhang, and H. Q. Le, "Wavelength modulation imaging with tunable mid-infrared semiconductor laser: spectroscopic and geometrical effects," Opt. Express 12, 5243-5257 (2004), <a href="http://www.opticsexpress.org/abstract.cfm?URI=OPEX-12-21-5243">http://www.opticsexpress.org/abstract.cfm?URI=OPEX-12-21-5243</a>. [CrossRef] [PubMed]
  10. Y. Wang, Y. W., C. Peng, H. Zhang, A. Seetheraman and H. Q. Le, �??Concepts for Scalable, CDMA-Networked, M/LWIR Semiconductor Laser Standoff Chemical Detection System,�?? in Optically Based Biological and Chemical Sensing for Defence, J. C. Carrano and A. Zukauskas, Eds., Proc. SPIE 5617, 179-189 (2004). [CrossRef]
  11. J.S. Yu, S. Slivken, A. Evans, J. David and M. Razeghi, �??Very High Average Power at Room Temperature from λ ~ 5.9 µm Quantum Cascade Lasers,�?? Appl. Phys. Lett., 82, 3397-3399 (2003). [CrossRef]
  12. Z. Morbi, D. B. Ho, H.-W. Ren, H. Q. Le, and S. S. Pei, �??Short-range remote spectral sensor using mid-infrared semiconductor lasers with orthogonal code-division multiplexing approach,�?? Opt. Eng. 41, 2321-2337 (2002). [CrossRef]
  13. Y. Wang, C. Peng., H. Zhang, and H. Q. Le, �??Remote spectral imaging with multi-wavelength and tunable, wavelength-modulation lasers,�?? in Proceedings of the Conference on Laser and Electro-Optics 1, 3 (2004).
  14. See e. g. A. Ishimaru �??Wave Propagation and Scattering in Random Media,�?? Academic Press, San Diego, CA (1978).
  15. See e. g., A. A. Kokhanovsky, Optics of Light Scattering Media, 2nd ed. (Praxis Publishing, 2001); T. A. Germer, �??Model Integrated Scattering Tool,�?? <a href="http://physics.nist.gov/Divisions/Div844/facilities/MIST/mist.htm">http://physics.nist.gov/Divisions/Div844/facilities/MIST/mist.htm</a>.
  16. A. Mackiewicz and W. Ratajczak, �??Principal Components Analysis (PCA)�??, Computers & Geosciences 19, 303-342 (1993).
  17. J. R. Jensen, Introductory Digital Image Processing: A Remote Sensing Perspective, 2nd ed. (Prentice Hall, Inc. 1996).
  18. R. Hardie, M. Vaidyanathan and P. F. McManamon, �??Spectral band selection and classifier design for a multispectral imaging laser radar,�?? Opt. Eng. 37, 752-762 (1998). [CrossRef]

Cited By

Alert me when this paper is cited

OSA is able to provide readers links to articles that cite this paper by participating in CrossRef's Cited-By Linking service. CrossRef includes content from more than 3000 publishers and societies. In addition to listing OSA journal articles that cite this paper, citing articles from other participating publishers will also be listed.


« Previous Article  |  Next Article »

OSA is a member of CrossRef.

CrossCheck Deposited