The pixel-scale patterning of optical filters directly onto sensors has the potential to drastically simplify and scale down the cost of real-time multispectral imaging. Without moving parts, a more robust and compact system can be deployed into harsher environments than have been previously accessible for multispectral imagers.

Figure 1: Multispectral imaging has many applications, including inspection, authentication, and sorting. The above image includes six vials of various white powders and two vials of clear liquids, demonstrating different properties as viewed in the infrared region with a PixelCam-SWIR. Money can be authenticated with this imaging system by utilizing the infrared-blocking ink that is applied in different patterns to bills. Coffee beans with impurities, in petri dish, demonstrate inspection and sorting capabilities.
Current multispectral imaging systems provide data about specific spectral regions. Similar to the common spectrometer, information about an object or scene can be obtained across UV-VIS-IR wavelengths. Imaging techniques, however, offer decided advantages over point spectral readings because of the ability to utilize the full capacity of an advanced imaging sensor. There is a broad range of applications where multispectral imaging and sensing techniques are utilized, including agriculture, machine vision/sorting, chemical and biological detection, art conservation, art and document authentication, and biomedical sciences (see Figure 1). The primary drawback with current systems is that they are not as portable or compact as would be desired for many applications. With patterned pixel sensors, those concerns can be addressed in a more cost-effective package.

Optical Coatings

Dielectric filters allow for unparalleled flexibility in spectral selection. A dielectric filter that passes only a narrow band of red laser light while blocking the rest of the visible spectrum, for example, is just as feasible to fabricate as one that passes blue starlight from hot stars while blocking light pollution from broad-spectrum home lighting. The most commonly encountered dielectric filter is the anti-reflective (AR) coating found in vision-correction glasses, home windows, and cell-phone cameras. For cameras and glasses, clarity is improved by eliminating glare. Cameras also acquire images more quickly, as more light reaches the sensor. Home windows can be coated to reflect the invisible infrared radiation that brings unwanted heat, while allowing more visible light through. Additionally, all of these coatings can be scratch- and chip-resistant when made with hardened materials. More complex coatings produce the assortment of application-specific filters found in biological imaging, chemical identification, and laser physics, ranging from filters passing only light produced from green fluorescent proteins to the specialized laser-line selectors (see Figure 2).

With dielectric filters placed prior to cameras, inexpensive monochrome imaging sensors can now be “sensitized” to only specific spectral bands allowed through the filter. Traditionally, dielectric filters are deposited onto a rigid, transparent substrate like glass or hard thermoplastics. Dielectric filters operate on the principle of interference: Light travels through media as waves, and by adjusting the phase of a reflected wave relative to its forward traveling counterpart, the amplitude of a specific frequency (or wavelength) at the output of the filter is modulated. The filter material is deposited in sub-wavelength stacks, the composition and arrangement of which dictate the spectral transmission properties. The aforementioned AR coating stack results in constructive interference across the visible spectrum. An improved, averaged index gradient between air and glass lessens the discontinuity in the refractive index, reducing the amount of light reflected at the air-glass interface. Spectral bands can be likewise eliminated by tuning the stack for destructive interference across the desired spectral subset.

Ultimately, dielectric filters improve imaging contrast by rejecting light outside of the spectral band(s) of interest, as well as improve the transmission of light within the band of interest. The transmission improvement is only possible with dielectric filters, and not with the dye filters typically employed in the ubiquitous RGB sensor.

Deposition on Sensors

Figure 2: Scanning Electron Micrograph (SEM) view of two individual, pixel-scale optical coatings
In principle, depositing interference filters onto substrates other than glass only requires accounting for the different refractive index of the substrate. Of particular interest is the direct application of a dielectric filter to the surface of an active device. Coated glass can be utilized as well, and placed in front of the sensor. Direct application, however, has a significant advantage in that there is one less optical surface in the light path. Additionally, internal Pixelteq researchers have calculated a 3-10% theoretical increase in light throughput to the sensor when the filter is deposited directly, compared to being on intermediary glass. The features of the device that make it active — such as bond pads on photodiodes — are the very source of the challenge in successfully applying a patterned filter onto a sensor. These sensitive areas must be protected from the bombardment of high-energy atoms in a coating chamber.

Adaptation of lithography techniques from the semiconductor world enable Pixelteq to coat dielectric filters directly onto the active portion of a device, while leaving sensitive areas free of material. The entire sensor, however, is not necessarily coated with a single filter; an imaging sensor can be patterned with an arbitrary number of filters in any predetermined geometric pattern. While the Bayer pattern with absorptive dye filters is the most common arrangement to generate an RGB image from a monochromatic sensor, use of dielectric filters allows for detection of a greater number of bands, each with adjustable transmission and blocking properties during fabrication.

In this manner, a monochrome sensor can become sensitive to multiple spectral bands without the need for external filter switching — as in the case of wheel-based multispectral cameras — or scanning over space, as in the case of hyperspectral imagers. Eliminating moving parts removes the costly temporal delay associated with transitioning between filters or sampling different locations in space, enabling true video-rate multispectral imaging. Additionally, removing the moving parts enables deployment into harsh environments where the rate of failure increases dramatically. Weight, power consumption, and design costs are all reduced with a directly-coated sensor.

Applying a patterned filter onto glass, then bonding the glass to a sensor with optical-grade epoxy, would seem to accomplish the same goal. There are two distinct drawbacks, however, to the approach. First, the index match between sensor and glass substrate is imperfect, and results in some degree of light loss. The loss is further compounded by the already poor efficiency of silicon sensors, even with AR coatings. Secondly, the dielectric coating — which was previously stated to be sensitive to angle of incidence — is further removed from the sensor plane by the thickness of the glass substrate, requiring the delivery of collimated light to the surface of the glass, not the image plane, potentially introducing aberrations that reduce image quality.

Applications of Patterned Pixel Devices

Figure 3: Image output from the PixelCam, demonstrating four individual channels representative of the four spectral filters applied to this sensor and the integrated, pseudocolor composite image
The improvement in frame-rate and the elimination of moving parts comes, however, with a cost. The functional resolution of the system is reduced as the number of filters applied increases. A nine-band camera, for example, would produce nine simultaneous images, but each with a resolution oneninth of the full frame. That information is not fully lost, and adapting image interpolation algorithms can aid in restoring the perception of resolution, as is currently performed in common RGB cameras.

Video-rate multispectral imaging with coated sensors opens up a previously inaccessible application space. Aerial inspection techniques benefit greatly from a system lacking both moving parts and environmentally sensitive optical epoxy. The imager benefits primarily from improved vibration resistance, and could be mounted onto any number of manned or unmanned observation platforms. Operational longevity is also improved, as the coated sensor platform is lighter and smaller than existing multispectral tools. Applications in the visible wavelengths include crop inspection, illicit drug enforcement, and a wide variety of biomedical research and clinical tools. Each application relies on the improved image contrast produced from spectral differentiation, as well as rapid image acquisition.

Figure 4: The PixelCam-SWIR imaging platform with pen for size comparison
For all multispectral imaging, spectral channels are highly application-specific and may be functionally defined by use of a flexible platform (interchangeable filter channels) such as the Pixelteq SpectroCam (see Figure 3). A siliconbased imaging sensor, for example, may be customized to acquire three specific visible wavelength regions, as well as a near infrared channel for a particular IR-emitting fluorophore.

Indocyanine green (ICG) is one such biologically relevant dye that has significant clinical and research applications due to it being FDA approved. ICG binds tightly to proteins in blood vessel walls and, as such, can be a very powerful tool for visualizing blood flow in a variety of biomedical use cases.

Applications within the security and defense space extend the spectral range from the visible and near infrared (NIR) into the short-wave infrared (SWIR) band of 900nm-1.7 microns (see Figure 4). The applications range from color night-vision to aerial surveillance and remote detection. Simultaneous acquisition of multiple spectral channels across the SWIR wavelength range is a unique ability facilitated by depositing selective bandpass filters at the pixel level on a SWIR (InGaAs) sensor. Additionally, multispectral imaging can prove a powerful tool in authentication applications, allowing innovative new methods of analysis for documents, artwork, antiquities, currency, pharmaceuticals, uniforms, and many others.

Conclusion

In the past few years, multispectral imaging has developed a great deal as a technique. For most applications, wavebands of interest can be narrowed to less than ten, which makes data acquisition and analysis much more manageable than with hyperspectral systems. Current innovations in coating techniques allow deposition of filters down to the single pixel level, and thusly, high-speed imaging simultaneously using multiple specific spectral channels is born. With proper selection of application-specific filters, pixel-scale patterned sensors will deliver multispectral imaging to many new users.

This article was written by Steve Smith, PhD, Product Manager, PIXELTEQ (Golden, CO). For more information visit http://info.hotims.com/45604-141 .