Anal. Chem. 2010, 82, 8412–8420
Multimode Imaging in the Thermal Infrared for Chemical Contrast Enhancement. Part 1: Methodology Heather Brooke, Megan R. Baranowski, Jessica N. McCutcheon, Stephen L. Morgan, and Michael L. Myrick* Department of Chemistry and Biochemistry, 631 Sumter Street, University of South Carolina, Columbia, South Carolina 29208 We combine a thermal light source with a conventional thermal infrared camera, alternating current (AC) detection methods, and chemical filtering of the infrared (IR) light to generate several imaging modalities in a simple manner. We demonstrate that digital lock-in amplifier techniques can increase the chemical contrast in an active thermal infrared image using both reflectance and thermal re-emission. We show this method is useful for visualizing thin coatings on fabrics that are invisible to the eye. We also take advantage of a “like-detects-like” chemical filtering approach to chemical selectivity for the purpose of chemical identification using a broadband thermal detector. Infrared (IR) imaging is a noninvasive, nondestructive technique used in variety of fields including forensics,1,2 medicine,3-7 agriculture,8-12 astronomy,13-17 and geology,18-25 to name a few. IR imaging can refer to measurement of thermal emission from scenes (thermography) or reflectance from samples. Thermography is a passive technique that relies on the thermal emission from the sample, which is not very sensitive to the chemical nature * To whom correspondence should be addressed. Phone: 803-777-6018. (1) Miskelly, G. A.; Wagner, J. H. Forensic Sci. Int. 2005, 155, 112–118. (2) Raymond, M. A.; Hall, R. L. Forensic Sci. Int. 1986, 31, 189–194. (3) Murthy, J. N.; van Jaarsveld, J.; Fei, J.; Pavlidis, I.; Harrykissoon, R. I.; Lucke, J. F.; Faiz, S.; Castriotta, R. J. Sleep 2009, 32, 1521–1527. (4) Chiang, M. F.; Lin, P. W.; Lin, L. F.; Chiou, H. Y.; Chien, C. W.; Chu, S. F.; Chiu, W. T. J. Formosan Med. Assoc. 2008, 107, 937–944. (5) Ovechkin, A.; Kim, K. S.; Lee, J. W.; Lee, S. M. Am. J. Chin. Med. 2003, 31, 455–466. (6) Yang, H. Q.; Xie, S. S.; Hu, X. L.; Chen, L.; Li, H. Am. J. Chin. Med. 2007, 35, 231–240. (7) Zhang, H. Y.; Kim, Y. S.; Cho, Y. E. Yonsei Med. J. 1999, 40, 401–412. (8) Bonwell, E. S.; Wetzel, D. L. J. Agric. Food Chem. 2009, 57, 10067–10072. (9) Bulanon, D. M.; Burks, T. F.; Alchanatis, V. Biosyst. Eng. 2008, 101, 161– 171. (10) Costa, L. N.; Stelletta, C.; Cannizzo, C.; Gianesella, M.; Lo Fiego, D. P.; Morgante, M. Ital. J. Anim. Sci. 2007, 6, 704–706. (11) Johnson, A. G.; Glenn, C. R.; Burnett, W. C.; Peterson, R. N.; Lucey, P. G. Geophys. Res. Lett. 2008, 35, L15606/1-L15606/6. (12) McCarty, J. L.; Loboda, T.; Trigg, S. Appl. Eng. Agric. 2008, 24, 515–527. (13) Kruse, F. A.; Perry, S. L. J. Appl. Remote Sens. 2009, 3, 22. (14) Acero, F.; Ballet, J.; Decourchelle, A.; Lemoine-Goumard, M.; Ortega, M.; Giacani, E.; Dubner, G.; Cassam-Chenai, G. Astron. Astrophys. 2009, 505, 157–167. (15) Bandfield, J. L. Icarus 2009, 202, 414–428. (16) Linz, H.; Henning, T.; Feldt, M.; Pascucci, I.; van Boekel, R.; Men’shchikov, A.; Stecklum, B.; Chesneau, O.; Ratzka, T.; Quanz, S. P.; Leinert, C.; Waters, L.; Zinnecker, H. Astron. Astrophys. 2009, 505, 655–661. (17) Sandberg, A.; Sollerman, J. Astron. Astrophys. 2009, 504, 525–U268.
8412
Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
of the sample.26,27 Measurements of reflectance can be further divided into single band or hyperspectral measurements. Hyperspectral reflectance imaging28,29 is an active technique that is chemically sensitive but often requires instrumentation that is delicate and complex and, therefore, expensive. Our laboratory has been interested in the possible application of infrared imaging to forensic-type investigations, as infrared light is absorbed by all organic substances. Would it be possible, for instance, to use this fact to develop a reasonably cost-effective means to visualize chemical contrast in a scene, one that is nondestructive and can be used in situ? If so, then thermal emission and reflectance are promising methods of choice since they require no physical interaction with potential samples. In addition, the mid-infrared spectral region appears to offer some advantages over the near-infrared because of the potential for measuring fundamental absorbances of analytes; thus, the cross sections for interactions with infrared light can exceed those of near-infrared light by several orders of magnitude. On the other hand, this wavelength selection places us firmly in the regime of strong absorption much of the time so that many of the ordinary models used to describe diffuse reflectance fail and, indeed, much of the reflectance is of a more specular than diffuse nature; it is often more appropriately described as Fresnel diffuse reflectance. (18) Danielescu, S.; MacQuarrie, K. T. B.; Faux, R. N. Hydrol. Processes 2009, 23, 2847–2859. (19) Veeder, G. J.; Davies, A. G.; Matson, D. L.; Johnson, T. V. Icarus 2009, 204, 239–253. (20) Calkins, J.; Oppenheimer, C.; Kyle, P. R. J. Volcanol. Geotherm. Res. 2008, 177, 695–704. (21) Gege, P.; Fries, J.; Haschberger, P.; Schotz, P.; Schwarzer, H.; Strobl, P.; Suhr, B.; Ulbrich, G.; Vreeling, W. J. Isprs J. Photogramm. 2009, 64, 387– 397. (22) Johnson, A. G.; Glenn, C. R.; Burnett, W. C.; Peterson, R. N.; Lucey, P. G. Geophys. Res. Lett. 2008, 35. (23) Prata, A. J.; Bernardo, C. J. Volcanol. Geotherm. Res. 2009, 186, 91–107. (24) Spampinato, L.; Calvari, S.; Oppenheimer, C.; Lodato, L. J. Volcanol. Geotherm. Res. 2008, 177, 301–312. (25) Tank, V.; Pfanz, H.; Kick, H. J. Volcanol. Geotherm. Res. 2008, 177, 515– 524. (26) Laloue, P.; Bissieux, C.; Henry, J. F.; Pron, H.; L’Ecolier, J.; Nigon, F. Int. J. Therm. Sci. 2008, 47, 249–260. (27) Ziegler, M.; Tomm, J. W.; Elsaesser, T.; Erbert, G.; Bugge, F.; Nakwaski, W.; Sarzala, R. P. Appl. Phys. Lett. 2008, 92, 3. (28) Kerekes, J. P.; Strackerjan, K. E.; Salvaggio, C. Opt. Eng. 2008, 47, 10. (29) Plaza, A.; Benediktsson, J. A.; Boardman, J. W.; Brazile, J.; Bruzzone, L.; Camps-Valls, G.; Chanussot, J.; Fauvel, M.; Gamba, P.; Gualtieri, A.; Marconcini, M.; Tilton, J. C.; Trianni, G. Remote Sens. Environ. 2009, 113, S110–S122. 10.1021/ac101109w 2010 American Chemical Society Published on Web 09/23/2010
Figure 1. Photograph of the multimode imaging system, with all key components labeled. The light source is a hot plate. The chopper was based on a modified fan bearing with a geared motor drive; the standard was used to synchronize data analysis. The filter was made from a KBr salt window coated with a film by dip coating. The camera is an 8-13 µm microbolometer-based thermal imager with 12-bit digitization, driven by LabVIEW software written in-house.
The method described here relies on a thermal imaging system adapted with a modulated light source that allows for the application of digital lock-in amplifier techniques to differentiate the responses due to diffuse reflection, thermal emission, and infrared-induced thermal re-emission. Using the alternating current (AC) response generated by the modulated light source gives an advantage over the conventional direct current (DC) response by enhancing the chemical contrast in the image, allowing for the detection and identification of a stain of interest. While this simple instrumentation is not suited to hyperspectral analysis, we discuss a “like-detects-like” chemical filtering approach described in one of our recent reports30 for the use of a broadband thermal detector for chemical identification purposes. Lodder et al. and Hieftje et al. have successfully used this approach, called molecular factor computing, for vapor detection and identification.31-33 EXPERIMENTAL SECTION Camera Setup. Figure 1 is a photograph of our instrument setup. A thermal light source is modulated by a chopper, resulting in modulated infrared irradiation of a sample. A thermal infrared camera captures light returning from the sample, and postprocessing is used to produce images corresponding to different modes of measurement. (30) Simcock, M. N.; Myrick, M. L. Appl. Spectrosc. 2006, 60, 1469–1476. (31) Dai, B.; Urbas, A.; Douglas, C. C.; Lodder, R. A. Pharm. Res. 2007, 24, 1441–1449. (32) Fong, A.; Hieftje, G. M. Appl. Spectrosc. 1995, 49, 1261–1267. (33) Fong, A.; Hieftje, G. M. Appl. Spectrosc. 1995, 49, 493–498.
To enable measurement of sample reflectance, the light source must overcome the background thermal infrared emission, which is near 465 W/m2 for a blackbody radiator at room temperature. The light source in this system is a 1000 W General Electric hot plate. Although the spectral distribution differs, the total thermal emission of the hot plate is approximately equal to the total blackbody radiation of a 2.2 m2 surface at room temperature. To ensure that only reflected light reaches the camera, the light source is surrounded by black optical cardboard, the outside of which is covered with aluminum foil to reduce its emissivity. The light source is modulated with a chopper built using the bearings of a home ceiling fan, equipped with flat blades to minimize the stirring of air. Each blade of the chopper is bright metal with the interior side (the side facing the source) coated with a black absorber. This provision reduces the stray infrared light that would have been caused by a highly reflecting metal blade, while simultaneously reducing the thermal re-emission that would have resulted from a blade that was optically dark all over. The chopper is rotated with a chain drive driven by a DC electric motor run from a variac. The chain drive reduces the rotation speed to achieve a chopping rate in the range of 1 Hz. This chopping rate is high enough to eliminate effects from the low-frequency drift of the light source but low enough that it is possible to see heating and cooling of the sample as a result of chopping. The detector is a Merlin uncooled microbolometer (Indigo Systems, Goleta, CA) with a focal plane array (FPA) consisting of a 320 × 240 matrix of microbolometer detectors. The FPA is Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
8413
in an evacuated metal dewar that is actively thermally stabilized at 313 K and incorporates internal thermal reference detectors to compensate for thermal drift and noise. The lens is germanium with an antireflective coating, limiting the spectral sensitivity of the system to 7.0-14.0 µm. The camera generates a real time 12-bit digital data stream with a frame rate of 60 Hz, which is collected and analyzed by LabVIEW (version 8.5 w/Machine Vision, National Instruments, Austin, TX) software designed in-house. The camera is connected to a dedicated computer with a Windows XP operating system via a parallel digital video connection (RS-422 line drivers, nominal rate 6.25 MHz). The samples are placed approximately 1 m away from both the light source and the detector. A 2 in. diameter gold diffuse reflectance standard (Optronic Laboratories, Orlando, FL) is placed in the sampled area to normalize the images for detector fluctuations over time and to provide a reference for calculating the phase of the data relative to the chopper position. A time study of the camera response revealed that the light source does not hold a steady temperature but instead shows a small sinusoidal oscillation with a period on the order of 2 min that we attribute to the method of thermal regulation used in the source. The effect of this slow variability in the light source is minimized by taking all data quickly and including the gold standard as a reference for source intensity.
Sample Preparation. A test sample was prepared that consisted of a chemical film on a fabric to illustrate contrast and visibility of the coating when viewed using various imaging modes. The fabric used was tricolor, unfinished acrylic obtained from a textile-related manufacturer in the Southeast. The acrylic fabric was dyed at the NC State School of Textiles pilot facility (Raleigh, NC), with three single basic dyes: CI basic violet 16, CI basic yellow 28, and CI basic blue 159. A commercially available single action airbrush (Paasche Airbrush Co., Chicago, IL) was used to apply a polymer solution of 11 g of Acryloid B67 (Rohm and Haas, Philadelphia, PA) in 100 mL of toluene on the fabric while covered by a stencil. Acryloid B67 was chosen because it is known to make thin films easily and has absorbance bands in the infrared region. A hose connected the airbrush to a laboratory nitrogen valve, and the gas pressure was adjusted to obtain a flow rate of 1 mL/min. The polymer solution was sprayed onto the surface of the sample for 4 min, in a sweeping motion. The average film coverage over the application area was calculated to be ∼18 µm, using the total mass of polymer applied (∼0.440 g), density of the polymer (1.04 g/cm3), and the total coated area of ∼230 cm2. The polymer film did not appear to saturate the fabric completely through, and it was dry only seconds after application was completed. Scanning electron microscopy (SEM) images on both faces of the coated fabric confirm that the polymer coverage is thicker
Figure 2. Example of the data block of images that are recorded by the camera. The scene is exposed to the light source, and then, the light is blocked, with several of these on/off cycles throughout the data block. The reference signal shows the on/off chopping phase of the light source. The intensity change is exaggerated in these images for clarity. 8414
Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
Figure 3. Example of the single-pixel raw data for a 15 s acquisition period. The gold standard shows sharp on/off transitions, while the fabric data show heating and cooling effects throughout the measurement. Note the scale of the gold signal is about 10× greater than the fabric signal because the reflectance of the fabric is much smaller.
on the front face of the fabric sample and too thin to be readily observed on the back. Chemical Filter Preparation. The same polymer (Acryloid B67) that was airbrushed on the fabric was used to coat one surface of a 50 mm diameter KBr window, which is transparent throughout the region of spectral sensitivity for the microbolometer. Masking tape was attached to the rim of the substrate to create a shallow well in which the polymer solution was pooled and allowed to dry to make a film of greater thickness than dip coating could produce. Approximately 25 drops of a solution of 20 g of polymer in 200 mL of toluene was pipetted onto the surface. The KBr substrate was then allowed to dry overnight in a horizontal position, allowing the solvent to evaporate. Transmission spectra of the dried film/filter were recorded using a Nexus 470 FT-IR (Thermo Nicolet, Madison, WI), and the film thickness was gravimetrically estimated to be on the order of 50 µm. RESULTS AND DISCUSSION Data Acquisition and Analysis. Before data collection begins, a real-time image is displayed on the computer monitor with active cursors. The cursors are used to designate the image area containing the gold standard to identify the portion of the image that will serve as a phase reference. Data acquisition at a frame rate of 60 Hz is then initiated to create a data block of 900 images (approximately 15 s). Data acquisition is not synchronized in any way with the position or frequency of the chopper. Consequently, the program uses the coordinates of the gold standard to generate a reference signal that is approximately in phase with the modulation of the light source. A second, sequential data block of images is collected with a chemical filter placed in front of the camera to preferentially block the wavelengths most likely to be affected by the coating that was applied to the fabric. Figure 2 shows the form of the data block; generally, one block is approximately 100-125 Mb in size.
Each pixel’s signal is similar to that shown in Figure 3, which includes multiple on/off cycles of the light source. With the chopper initially blocking the light source, it was observed that single-pixel intensities exhibited a rising baseline once the chopper was turned on. This was interpreted to mean that the source was gradually warming the sample and increasing its thermal emission. Consistent with this interpretation, the single-pixel signals from most of the image area do not show the sharp, symmetric on/off transitions one might expect from a chopped reflectance measurement. Instead, the transition from the “off” state to the “on” state shows an abrupt onset that continues to rise more slowly during the remainder of the “on” phase. The transition from “on” to “off” likewise shows an abrupt decrease in signal followed by a slower decrease that continues until the next upward transition. Initially, this decrease during the “off” state is smaller than the corresponding increase in signal during the “on” state, resulting in a rising baseline. To minimize this gradual increase in the baseline at the start of each experiment, the system is allowed to equilibrate with the chopped light source for approximately 5 min before data is collected. The shape and intensity of the on/off signals from each pixel of the image are informative about the absorptivity and emissivity of the sample. The heat deposited per unit area per unit time in each element of the image is the product of the illumination irradiance with the absorptance, taken as one minus the reflectance, of the sample: d2q ) dt dA
∫
∞
λ)0
Eλ(1 - r(λ)) dλ
(1)
where q represents heat energy, dt is an element of time, dA is an element of area, Eλ is the spectral irradiance of the light integrated over all angles of incidence, r(λ) is the total spectral reflectance of the sample, and 1 - r(λ) is the spectral absorptance of the sample. Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
8415
Figure 4. (Left) Photograph of the tricolored acrylic sample with a plastic stencil of the USC logo placed in front of it for airbrushing. (Right) Acrylic sample after airbrushing with Acryloid B67 and removal of the stencil. There is little apparent contrast between the areas of neat fabric and polymer stained fabric in the visible region.
To a first approximation, the average temperature increase of each image element is proportional to the energy absorbed per unit time. Therefore, the average temperature increase is correlated with the sample absorptance and negatively correlated with the sample reflectance. Thus, we can expect that areas of low reflectance will also be areas with strong modulation in the heating due to irradiance. Signals from these areas will have low-amplitude square-wave AC modulation due to reflectance but will also show a modulated thermal emission whose amplitude and shape will depend on the chopping frequency. At high chopping frequencies, for example, this modulation of the thermal emission would be low in amplitude and approximately sawtooth shaped. Conversely, areas of high reflectance will show large-amplitude square-wave reflectance modulation with minimal thermal modulation effects. Because heating and cooling continue throughout every “on” and “off” cycle, respectively, any thermal modulation of an image element tends to reach extrema at the instants when the states are switching, and thus, the thermal modulation is out of phase with the reflectance by e90 degrees. Since gold has a high reflectance in this spectral region, signals for pixels imaging the gold standard do not show such heating effects but instead show sharp, symmetric on/off transitions. Thus, the gold response is completely in phase with the modulation of the light source and is used as a reference signal for data analysis. This reference signal is converted to a square wave and used for various purposes in analyzing the image data. On the basis of this reference wave, for instance, the three-dimensional data blocks are truncated, in the time dimension, to render an integral number of cycles, allowing the application of digital lock-in amplifier techniques to extract the signal of interest. Lock-in amplifier techniques use synchronous demodulation to extract signals from carrier waves that have 50% duty cycles, such as a sinusoidal or square wave. To do this, the modulated carrier wave is multiplied by a reference signal that has the same frequency of modulation, creating a DC output containing only the components that are in phase with the reference signal. By changing the phase of the reference wave, it is possible to extract signals that are orthogonal to one another.34 (34) Ingle, J. D.; Crouch, S. R. Spectrochemical Analysis; Prentice Hall: Englewood Cliffs, 1988.
8416
Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
In this experimental setup, the signal that is exactly in phase with the modulation of the light source is due to the fraction of light that is diffusely reflected from the sample, corresponding to the reference signal with no phase shift (0°, hereafter called the AC-0 signal). As explained above, the thermal modulation is approximately orthogonal to the reflectance measurement in the “fast chopping” regime and is extracted by shifting the reference signal by 90° (hereafter called the AC-90 signal). The data in the truncated arrays is mean-centered (pixel-wise) and multiplied by the reference wave (with phase shifts of 0° and 90°), and each pixel is averaged through the time dimension. Mean centering eliminates the large DC offset in the single-pixel modulated data; this minimizes “noise” in the lock-in-type signal caused by having inexact synchronization between the chopping frequency and the frame frequency of the camera. Our camera does not always acquire an image at the precise moment that would give a phase shift of 90 degrees. To obtain an accurate image for this phase change, we use a MATLAB program to find the frame where the gold standard changes sign. We then calculate the “real 90” phase image by a weighted average of the frames immediately before and after the sign change. All AC-90 images reported are collected in this manner. The data is converted into an image, which is equalized, displayed on the screen, and saved as a JPG. The DC average is also calculated, as it corresponds to the response expected from traditional thermography techniques, and is used as a comparison of the imaging techniques. The truncated data blocks and their corresponding reference signals are also saved in binary format to enable postprocessing. DC, AC-0, and AC-90 IR Imaging. Figure 4 shows photographs of the sample used as a representative illustration for this detection system. To the eye, there is no visible difference in the fabric sample after the application of the polymer stain. Figure 5 shows six imaging modes of the sample using a combination of the DC, AC-0, and AC-90 measurements imaged from the scene directly or through a filter composed of the same material as the coating on the fabric. Figure 5A shows the average DC amplitude for the unfiltered response, which is analogous to the traditional IR thermographic image and can be used to detect thermal gradients in the image.1–25 There is no significant contrast between the fabric and the polymer stain in this imaging mode
Figure 5. (A) DC average of unfiltered data showing no contrast between neat and doped fabric. (B) AC-0 unfiltered image showing significant contrast between neat and doped fabric. (C) AC-90 unfiltered image showing heating/cooling effects of the sample, some contrast between neat and doped fabric is apparent. (D) DC average of data with an ∼50 µm Acryloid B67 filter in front of the camera, again no contrast between neat and doped fabric. (E) AC-0 filtered image, the decrease in contrast found compared to the unfiltered image suggests that the stain is similar to the filter material. (F) AC-90 filtered image, loss of contrast is at least partially due to the decreased intensity in the presence of the polymer filter.
because the fabric is at thermal equilibrium and behaves as an optically thick radiation source. The lack of chemical contrast in this conventional infrared image, despite the fact that the fabric and its coating are readily distinguished via their infrared spectroscopy, strongly suggests that chemical information will not be readily obtained via simple thermography. This type of DC thermograph is commonly used for detecting temperature gradients. The fundamental reason that chemical contrast is not apparent in a DC thermograph is that thermal emission of an optically thick equilibrated sample is dependent on its spectral absorptance (one minus reflectance). Often the reflectance is quite small in the 7-14 µm wavelength band sensed by the camera, approaching the specular limit at many wavelengths. While the relative spectral variability of the reflectance may be substantial and the absolute spectral variability of the absorptance may be equal to that of the reflectance, the relative spectral variability of the absorptance is necessarily small under these conditions. As an example, a 10% relative variability in a reflectance of 0.1 represents approximately a 1% relative variability in the absorptance. The AC-0 image corresponding to the same data set is shown in Figure 5B. There is now a distinct difference in the contrast between the neat and doped fabric, and the logo is clearly identifiable. There are also visibly observable contrast differences in the background plywood, suggesting a chemical difference in the grain of the wood, not just the color difference seen in the visible region. As a way of quantitatively comparing the contrast between neat and doped areas of fabric in the images, we calculate both the Weber contrast35 (WC) and Fisher contrast (FC) per pixel of the groups, which are 18.15% and 6.95, respectively, for the AC-0 image. These contrast descriptors are defined by eq 2, where ¯x is ¯ is the mean pixel the mean pixel intensity of the cropped area, X (35) Tang, J. S.; Peli, E.; Acton, S. IEEE Signal Process. Lett. 2003, 10, 289– 292.
intensity for all samples, n is the number of pixels averaged, and s is the standard deviation of the cropped area: WC )
|
|
¯xneat - ¯xdoped (100%) ¯xneat
(2a)
FC ) ¯ )2 + ndoped(x¯doped - X ¯ )2)(nneat + ndoped - 2) (nneat(x¯neat - X (nneatσneat2 + ndopedσdoped2)(nneat + ndoped - 1) (2b) The Weber contrast is a traditional method of determining the contrast value in an image and is useful for comparing the difference in the raw intensities of the samples. However, it does not take into account the pixel noise, which greatly affects the visually observable contrast difference in an image. For this reason, we include the nonconventional method of Fisher contrast, as the calculated values compare well with visually observable trends between the various image types. The amplitude of the AC-0 image is significantly smaller than that of the DC image, roughly 4% as large when calculated by comparison to the response of the gold standard. A quick backof-the-envelope calculation shows why this should be so: the DC image is dominated by blackbody radiation, which is much more intense than the radiation due to reflectance. A blackbody radiator (e.g., our sample) at 300 K emits approximately 450 W/m2 over all wavelengths, roughly 43% of which (200 W/m2) is in the wavelength region detected by the camera. Our IR source has a surface temperature near 700 K (measured with a two-band optical pyrometer (KT 19.81 II, Heitronics, Wiesbaden, Germany) and assuming an emissivity of 0.8), and if we take its 1000 W energy consumption as all in the form of radiation, less than 300 W would be produced in the camera’s band. Taking 1/2 of that as directed toward the fabric sample and Analytical Chemistry, Vol. 82, No. 20, October 15, 2010
8417
assuming it illuminates a zone with a radius of 1 m, the irradiance of the fabric would be less than 50 W/m2. Diffuse reflectance spectra of the fabric samples show that in the camera’s wavelength range the reflectance of the acrylic fabric rarely exceeds 10% reflectance. Taking that value as the average reflectance, we would expect a maximum of 5 W reflected from the sample, about 2.5% of the total detected at the camera. This result is within a factor of 2 of what is actually observed. By applying the lock-in amplifier analysis, we are able to retrieve this low absolute signal and generate an image that is less prone to drift and pixel variability than the DC image. Thus, the AC image quality often appears of higher quality than the DC image, despite being formed by only a small fraction of the total IR light intensity. The AC-0 image results mainly from reflectance, and its amplitude is independent of the chopping rate until the rate is so slow that heating and cooling of the sample gives a competitive signal (typically