Hyperspectral imaging, also called imaging spectroscopy, is a method of obtaining the spectral content of each pixel in a 2D image. This technology divides the image data, pixel by pixel, into very narrow wavelength (color) bands. The resulting 3D data cube (x, y, λ) allows materials to be identified by their pixel spectral content in addition to their spatial characteristics (see Figure 1). This high-resolution spectral data can effectively recognize unresolved features in the image. It can identify chemical compounds, distinguish camouflage paint from tree canopy, tag recently disturbed earth, and provide data beyond the spatial resolution of the imaging system. Consequently, hyperspectral imaging is a technique that can significantly benefit the military for reconnaissance, surveillance, and material identification and characterization. In manufacturing, it can be used for process and quality control, in mining and geology for ore mapping, and in biological and chemical research for chemical imaging.1 Ideally, hyperspectral instruments would be able to obtain all the components of a data set at the same time, something that current systems have not achieved.
Figure 1. Graphical representation of the data cube, with position along two axes and wavelength along the third. Many scanning systems measure only one slice through the cube at any given instant. Wavelength- scanning systems measure a slice of the cube at a fixed wavelength (vertical plane), whereas slit-scanning systems measure at a fixed position in one spatial dimension (horizontal plane). The HPATM system is unique in that a parallel optical processor measures the entire data cube simultaneously and continuously distributes it onto a detector array. (Figure © Bodkin Design and Engineering)
Most hyperspectral instruments use a 2D detector array and are scanned over time to acquire the third dimension of data. Existing systems are based on one of three common techniques: a spatially-scanning slit spectrometer, such as a push-broom imager; a wavelength-tuned spectral filter,2 such as a tunable etalon or a wheel with multiple Fabry–Pérot filters; or a two-dimensional Fourier transform imager.3 Bodkin Design and Engineering has developed systems based on all of these techniques but discovered that they suffer from the same limitation: they are inherently inefficient because they must scan one of the three dimensions of the data cube in order to develop the entire data set. This is time-consuming and only allows short detector integration times while the data set is being built up. In the case of low-light observations, and particularly in the infrared, this situation leads to observations that are photon-starved.
An additional instrument based on tomography (computed tomography imaging spectrometer, or CTIS)4 captures multiple spectra for each image point, overlapped on a focal plane array. An iterative algorithm recovers the content for each image point. This device offers nonscanned hyperspectral data cubes but requires time-consuming digital computation. Moreover, it has a reduced signal-to-noise ratio (SNR) owing to the many spectra.
Bodkin Design and Engineering has developed a unique approach to the hyperspectral instrument problem based on its patent-pending HyperPixel ArrayTM (HPA) technology. This approach creates a camera whose focal plane consists not of pixels but HyperPixelsTM, each of which detects the spectral content of the light striking it, up to hundreds of color bands (see Figure 2). This technique has three chief advantages: it can function as a conventional focal plane with interchangeable lenses and thus be attached to a standard telescope or a microscope; it provides hyperspectral data cubes at video frame rates, enabling the characterization of phenomena that change quickly; and it is a staring sensor, which means that its SNR is superior to scanned hyperspectral devices.
Figure 2. The miniature staring HPA imager. This device captures all three dimensions of the data cube in one video frame. It runs off a standard USB port, using a 1280×1024 focal plane array. The device has an interchangeable lens to provide various fields of view or attachment to a microscope.
The HPATM is a massively parallel system that collects the full 3D hyperspectral data cube without scanning. Incident photons from a surveillance image are collected by a two-staged optical processor. This device manipulates the data set prior to any electronic detection or software processing, operating at the speed of light. No computer algorithm can process data faster.
The 2D HPA reformats the entire image into a series of discrete spectra and relays them onto a conventional 2D detector array. Thus, each HyperPixel measures the full spectrum and its spatial location in every frame of data. Because this device uses a staring focal plane array, very high SNRs can be achieved. Furthermore, because the proprietary HPA design uses an optical parallel processor, it allows thousands of HyperPixels to be processed simultaneously. This leads to the nearly instantaneous creation of the data cube. The output of the processor is coupled to a conventional detector array whose speed is limited only by its digital clock. The instrument can capture tens to hundreds of data cubes per second.
The HPA technology is a significant improvement over conventional hyperspectral imaging, not least because it eliminates long scanning times and mitigates the associated stability requirements. The system can measure transient signals and rapidly occurring events from handheld or unstable platforms. It has no moving parts and is inherently rugged, compact, and inexpensive. The HPA can benefit users in numerous markets and opens a new range of problems to a hyperspectral solution.
Bodkin Design and Engineering, LLC
Andrew Bodkin has over 20 years of experience with mechanical, optical, and electro-optic systems. In 1998, he founded Bodkin Design and Engineering LLC, based in Newton, MA. The firm has developed numerous patented devices and overseen the successful introduction of a diverse spectrum of products, from miniature infrared cameras to dental imagers to spectroscopic instruments for drug discovery. He holds a BA in physics from Amherst College and an MS in electro-optics from Tufts University.
2. J. T. Daly, A. Bodkin, William J. Schneller, Robert B. Kerr, John Noto, Raymond Haren, Michael T. Eismann, Barry K. Karch, Tunable narrow-band filter for LWIR hyperspectral imaging, Proc. SPIE 3948, pp. 104-105, 2000. doi:10.1117/12.382145
3.Michael T. Eismann, John H. Seldin, Craig R. Schwartz, James R. Maxwell, Kenneth K. Ellis, Jack N. Cederquist, Alan D. Stocker, Ara Oshagan, Ray O. Johnson, William A. Shaffer, Marc R. Surette, Martin J. McHugh, Alan P. Schaum, Larry B. Stotts, Target detection in desert backgrounds: infrared hyperspectral measurements and analysis, Proc. SPIE 2561, pp. 80-97, 1997. doi:10.1117/12.217672