LINEAR VARIABLE FILTERS – A CAMERA SYSTEM REQUIREMENT ANALYSIS FOR

HYPERSPECTRAL IMAGING SENSORS ONBOARD SMALL REMOTELY PILOTED

AIRCRAFT SYSTEMS

Philippe Serruys1, Aleksandra Sima1, Stefan Livens1, Bavo Delauré1,

Klaas Tack2, Bert Geelen2, Andy Lambrechts2
1
Flemish Institute for Technological Research - VITO NV, Mol, Belgium
2
IMEC, Leuven, Belgium

ABSTRACT multispectral or hyperspectral imagery offers vastly

enhanced possibilities for these applications.
Improving the spectral detail of earth observation imaging
from Remotely Piloted Aircraft Systems (RPAS) can greatly To fit the payload mass and volume constraints of small
expand its potential for use in vegetation monitoring and RPAS with a hyperspectral instrument, variable interference
specifically in precision agriculture. Spatially variable filters are an optimal choice. The filters can be placed very
interference filters which can be placed very close to the close to the sensor, which allows building a compact and
image sensor offer an excellent opportunity for reducing the lightweight instrument. The LVF-technology has also been
size, mass and complexity of hyperspectral imagers, proposed for hyperspectral instruments for small satellites
allowing them to be mounted onboard small RPAS. Recent [1, 2].
advances in filter deposition techniques allow to directly
deposit interference filters on an image sensor. The Besides their low mass, the optical filter technology
monolithic integration of optical hyperspectral filters on top developed by IMEC offer an additional advantage: the
of a standard CMOS image sensor has been demonstrated ability to design the spatially variable filter and optimize it
by IMEC. Compared to the more conventional deposition of for specific applications. For example, a camera with
filters onto an external glass substrate, this new approach hyperspectral filters from the red towards the red-edge and
offers advantages in terms of cost, alignment accuracy, near-infrared part of the electromagnetic spectrum are
straylight , etc.. recommended for vegetation monitoring applications, such
A hyperspectral camera prototype compatible with small as precision agriculture, habitat monitoring and
RPAS has been developed by VITO to demonstrate the environmental impact studies.
potential of LVF-based compact spectral camera’s. Whereas
application of the filter technology offers major advantages However, hyperspectral acquisitions onboard RPAS still
for RPAS systems, it still faces some important challenges. face some major challenges, even when employing LVF
The prototype system specifications need to fit a fixed wing technology. The camera system needs to be very
RPAS platform that is able to cover several km² in a single lightweight, but should acquire images at high frame rates,
flight with hyperspectral geo-information. It remains with according data write speed and data storage capacity.
challenging to make a sufficiently compact camera system, The integration time, being a compromise between motion
achieve precise spectral band registration, handle the blur and signal to noise ratio, is another important challenge.
amount of data to be processed and cope with limited Acquiring narrow spectral band images inherently implies
integration times possible during acquisition. that the amount of light per band is limited, which lowers
the signal to noise ratio.
Index Terms— hyperspectral, multispectral, camera,
RPAS, precision agriculture, linear variable filter, This paper provides a comparison of different camera
configurations equipped with LVF technology and their
1. INTRODUCTION impact on the system specification requirements for an
integration on a RPAS fixed wing platform. Frame rate
Using RPAS based imaging sensors capable of capturing specifications are analyzed, as well as data write speed, data
data in several spectral bands in the visible and near-infrared storage, integration time and motion blur.
part of the electromagnetic spectrum allows capturing
highly accurate image data at a high spatial resolution,
making them useful in many types of remote sensing
applications. Increasing the number of spectral bands to

Authorized licensed use limited to: TU Delft Library. Downloaded on October 25,2023 at 21:12:02 UTC from IEEE Xplore. Restrictions apply.
 2. LINEAR VARIABLE FILTERS

LVFs are Fabry-Pérot interference filters in which the

thickness of the cavity varying linearly along one
dimension, transmitting light of specific wavelengths to a
2D array of detectors. Typically, the filters are deposited on
a glass substrate which is then mounted onto the sensor.

Recent advances at IMEC enable such filters to be

deposited directly on top of the image sensors. The direct
deposition eliminates alignment issues between filter and
sensor. This can yield a higher quality imager and also
allows a more flexible design. The thickness of the cavity
can be chosen per sensor pixel, eq.g. multiple sensor lines or
groups of pixels can be designed to have the same spectral
response. [3, 4, 5]. This is illustrated in Figure 1.

Figure 2 : Acquisition principle with LVF

To reconstruct the spectrum, a very accurate

coregistration of the spectral bands (and thus the image
series) is essential, which requires accurate knowledge of
position and attitude. Any inaccuracy in the georeferencing
between the images will result in a reduced performance of
Figure 1: A linear variable filter configuration the LVF imaging system. An illustration on how inaccurate
monolithically integrated onto an image sensor coregistration can affect the resulting spectrum is given in
Figure 3.
Different configurations of spatially variable filters are
possible, apart from the present LVF configuration, other
configuration like a snapshot imager also possible [5].

A spectral imaging system based on LVF technology can

be very compact because the spectral imager does not rely
on any additional optics. However, the acquisition of the
spectral channels is fundamentally different compared to
traditional prism or grating based line scanning
hyperspectral imagers.

An LVF imager employs an acquisition principle that is

schematically shown in Figure 2. As the thickness of the
cavity is varying linearly along one dimension on the sensor,
the different lines of the sensor receive light of different
narrow spectral bands. One snapshot acquires a 2-D scene,
but every line (in the across flight direction) corresponds to
a different spectral band. As a consequence a scanning
motion is required to retrieve the complete reflectance
spectrum of a certain point on the earth.
Figure 3: Top: acquisition: with perfect co-registration
Bottom: imperfect co-registration causing false spectrum.

The top drawing shows the ideal case in which the

measurement of the different spectral channels effectively

Authorized licensed use limited to: TU Delft Library. Downloaded on October 25,2023 at 21:12:02 UTC from IEEE Xplore. Restrictions apply.
comes from the same location (red dot). In the bottom Test flights with the prototype comprising the two
drawing, the measurements comes from different locations. sensors are currently being performed at VITO onboard of
If the spectral reflectance at these locations is different, the an octocopter.
combined samples will result in an incorrect spectrum.
4. CAMERA SYSTEM SPECIFICATIONS
3. SPECTRAL CAMERA PROTOTYPE DESIGN
We aim to integrate the spectral camera on fixed wing
In order to demonstrate the potential of the spectral platforms to cover several km² in a single flight. Therefore a
imager with direct filter deposition for integration onboard detailed analysis of impact of the camera system
RPAS a compatible hyperspectral payload prototype has specifications has been carried out.
been developed at VITO. The camera prototype is based on
a 2048 by 1088 pixels sensor (pixel pitch of 5.5μm) with an A number of different instrument scenarios have been
LVF filter developed by IMEC deposited directly on the considered. All use a zone of 2048 x 1088 pixels covered
sensor surface. The current filter consists of 99 narrow with spectral filters. The number of spectral bands and the
spectral bands (at 8 detector lines per band) covering the optics focal length are varied, as shown in Table 1.
spectral range of 600-900 nm with a FWHM less than 15
nm. The system is equipped with an 18.5 mm hyperspectral Scenario 1 2 3 4
lens.
# spectral bands 136 30 20 20
The prototype hyperspectral payload is designed to be # detector lines / band 8 36 54 54
compact and lightweight (1kg including power distribution
focal length [mm] 18 18 18 9
unit, GPS receiver and inertial measurement unit) and to
provide high resolution data from a small multi-rotor RPAS. Table 1 : Scenario specifications
When equipped with a 18.5 mm lens and flying at 150m
altitude results in a ground sampling distance of 4.5 cm and Calculations were performed to obtain the required
a swath of 91 m. camera system specifications for the different scenarios.
Following parameters have been calculated: required frame
rate, the amount of required images per ha, the data storage
per km², the data write speed, the integration time and its
influence on the motion blur due to platform movement.
For all calculations, a cruise speed of 21 m/s has been
assumed, which is an average speed for fixed wing mini-
RPAS platforms able to fly several km² per flight. The
platform attitude also influences camera system
requirements, when assuming that no attitude correction
mechanism (such as a gimbal) can be loaded onto the mini-
RPAS. A realistic attitude rate was set to 4 degrees per
second, which is 2 standard deviations of the pitch motion
Figure 4: Octocopter equipped with prototype of a mean mini-RPAS platform flying at 21 m/s. Finally, a
hyperspectral payload standard operational altitude of 150 m above ground level
has been adopted. This altitude will very likely be the
The spectral camera prototype has been integrated with maximum allowed in several European countries.
an additional camera containing a panchromatic 2048 x
2048 pixels sensor (pixel pitch of 5.5μm), including 4.1. Frame rate and data storage
simultaneous triggering. The panchromatic spectral
broadband camera can be used to determine image exterior The required frame rate of the spectral imager can be
orientation and derive the digital surface model as an input calculated as follows:
for the geo-referencing of the image data of the spectral ݂ ‫݈ ݈ܽݐ݋ݐ‬௦
camera. This functionality of both cameras can be integrated ݂‫ = ݎ‬൫‫ݒ‬௖௥௨௜௦௘ + 2ߪ(‫ݒ‬௣௜௧௖௛ )൯
‫ܸ ܣ‬௦ ݈௦ ‫ܾ݀݊ܽ ݎ݁݌‬
in a single unit combining both the panchromatic and
spectral filter zone on one image sensor chip (the geo-
Where fr is the required frame rate, vcruise the mini-RPAS
spectral camera concept) [6].
platform cruise speed, ı Ypitch) the standard deviation of the
platform’s pitch rate, A the operation altitude above ground
level, Vs the vertical length of the sensor, f the lens focal

Authorized licensed use limited to: TU Delft Library. Downloaded on October 25,2023 at 21:12:02 UTC from IEEE Xplore. Restrictions apply.
length, ls the number of detector lines on the sensor. Table 2 5. CONCLUSION
provides the results of the calculation for the different
scenarios. LVF is a promising technology which allows to equip
mini-RPAS with spectral sensors. This opens up new
Scenario 1 2 3 4 opportunities to deliver improved hyperspectral and
multispectral data for vegetation monitoring and other earth
Frames / s 63 19 13 6
observation applications at low cost.
Images / ha 1452 320 214 53 The camera prototype at VITO has been designed to be
GB/km² @16bit 603 133 88 22 operated from an octocopter RPAS. First test flights are
GSD [cm] 4,6 4,6 4,6 9,2 being conducted with this configuration. As the aim is to
Swath [m] 94 94 94 183 integrate the spectral camera on fixed wing platforms which
Table 2 : Calculation results per scenario can cover several km² in a single flight.
Detailed analysis of the camera system specification
In the first scenario, the aim is to cover the entire area impact has been achieved. Required frame rates have been
with a hypercube of 136 bands. As the LVF divides the calculated, as well their impact on the write speed and data
sensor bands of each 8 detector rows, the frame rate is high storage in order to get clear camera system requirements.
and the data storage demanding. The following scenarios Influence of data reduction techniques such as pixel binning
consider less bands, 30 and 20 bands spread over the sensor have been considered. Finally, the integration time and its
assuming a multispectral filter design. This decreases the impact on the motion blur on a fast moving mini-RPAS
required frame rate and the needed data storage. In the last platform has been evaluated.
scenario the 18 mm lens is replaced by a 9 mm lens,
resulting in a further decrease of the required frame rate to 6 6. ACKNOWLEDGMENTS
frames /s and the required data storage to 22 GB/km².
The authors would like to acknowledge the support of
4.2. Integration time and motion blur IWT SBO-project 100021 “CHAMELEON”.

Choosing a suitable integration time is always a 7. REFERENCES

compromise between two conflicting goals. On one hand
one wants to limit the motion blur by choosing a short time, [1] L. Maresi, M. Taccola, M. Kohling, and S. Livens, “Phyto-
whereas on the other hand one intends to maximize the Mapper - Compact Hyperspectral Wide Field of View Instrument,”
in Proc. 7th IAA Symposium on Small Satellites for Earth
signal with a long time (especially for narrow bands).
Observation, May 04 - 08, 2009, Berlin, Germany, 2009, pp. 1–6.
[2] Moreau V., DeClerq, C. Lousberg G., Maresi L., Delauré B.
“Development of a compact hyperspectral / panchromatic imager
for management of natural resources”, Proceedings of the 4S
symposium on small satellites, services and systems, 4-8 June 2012
Portoroz, Slovenia
[3] N. Tack, A. Lambrechts, P. Soussan, and L. Haspeslagh, “A
compact, high-speed, and low-cost hyperspectral imager,” in Proc.
SPIE 8186, Electro-Optical Remote Sensing, Photonic
Technologies, and Applications V, 81860Y, 2011, pp. 1–6.
[4] B. Geelen, N. Tack and A. Lambrechts, “A snapshot
multispectral imager with integrated tiled filters and optical
duplication,” Proc. SPIE 8613, Advanced Fabrication
Technologies for Micro/Nano Optics and Photonics VI, (2013).
Figure 5 : Motion blur in relation to integration time [5] B. Geelen, N. Tack and A. Lambrechts, “A compact snapshot
multispectral imager with a monolithically integrated, per-pixel
Figure 4 provides the relation between sensor integration filter mosaic,” Photonics West, (2014, submitted).
time and the resulting motion blur for a fixed wing mini- [6] Delauré B, Michiels B., Biesemans J., Livens S. and Van
RPAS at cruise speed of 21 m/s with pitch attitude motion Achteren T., “The geospectral camera: a compact and
of 4 degrees/s. geometrically precise hyperspectral and high spatial resolution
If a maximal motion blur of 2 pixels is tolerated, the imager”, International Archives of the Photogrammetry, Remote
Sensing and Spatial Information Sciences, Volume XL-1/W1,
integration time should not exceed 2.5 ms. At this ISPRS Hannover Workshop 2013, 21 – 24 May 2013, Hannover,
integration time, the challenge is to achieve a satisfying Germany
signal to noise ratio. Several different strategies are
followed at IMEC and VITO to address this.

Authorized licensed use limited to: TU Delft Library. Downloaded on October 25,2023 at 21:12:02 UTC from IEEE Xplore. Restrictions apply.