Adaptive hyperspectral imaging with a MEMS-based full-frame programmable spectral filter

Abstract

Rapidly programmable spatial light modulation devices based on MEMS technology have opened an exciting new arena in spectral imaging: rapidly reprogrammable, high spectral resolution, multi-band spectral filters that enable hyperspectral processing directly in the optical hardware of an imaging sensor. Implemented as a multiplexing spectral selector, a digital micro-mirror device (DMD) can independently choose or reject dozens or hundreds of spectral bands and present them simultaneously to an imaging sensor, forming a complete 2D image. The result is a high-speed, highresolution, programmable spectral filter that gives the user complete control over the spectral content of the image formed at the sensor. This technology enables a wide variety of rapidly reprogrammable operational capabilities within the same sensor including broadband, color, false color, multispectral, hyperspectral and target specific, matched filter imaging. Of particular interest is the ability to implement target-specific hyperspectral matched filters directly into the optical train of the sensor, producing an image highlighting a target within a spectrally cluttered scene in real time without further processing. By performing the hyperspectral image processing at the sensor, such a system can operate with high performance, greatly reduced data volume, and at a fraction of the cost of traditional push broom hyperspectral instruments. Examples of color, false color and target-specific matched-filter images recorded with our visible-spectrum prototype will be displayed, and extensions to other spectral regions will be discussed.

Calibration of programmable spectral imager with dual disperser architecture

“…For example, one can perform Hadamard transform spectroscopy by applying a Hadamard mask on the DMD, improving SNR via the Fellgett advantage [8,9,10,11,12]. By making adjustments to the system (using a telecentric lens system), it is possible to perform near-snapshot spectral filtering [13,8,14,15,16], identifying target chemicals with a fast frame rate. There is also interest in reducing the number of acquisitions required per datacube; recent work has measured the spectrum using a Fourier basis, which can reduce the number of acquisitions, as long as there are no sharp spectral features [17].…”