The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
216 
1. INTRODUCTION 
Hyperspectral imaging pushbroom spectrometers 1,3,4 are 
currently used in several domains in order to identify the 
spectral signatures of a broad range of materials in the reflected 
solar energy spectrum. Those instruments are designed such that 
they could sense the largest variety of targets with relatively 
high performances. Scientific requirements are retrieved by 
investigating the properties of representative applications (e.g. 
agriculture, limnology, vegetation, soil, atmosphere) 5 that the 
instrument will be used for along its operation life and they are 
formulated, for instance, in terms of required spectral 
resolution 6,7 , signal-to-noise ratio (SNR), noise-equivalent delta 
radiance (NeAL) 5 , or spatial resolution. Such an approach is 
justified by the fact that both airborne and spacebome sensors 
usually sense various targets at the same time, and reasonable 
instrument performances must be granted more or less all over 
the spectral range. Therefore every instrument comes with a 
default optical and electrical configuration. 
Nevertheless scientists might be interested in a specific 
application that groups few similar targets and/or only a portion 
of the instrument spectral range; they could then request to 
optimize the instrument performances in that range. For 
instance, a scientist might want to discriminate 0.5% water 
content in an olive crop and he is interested in the spectral 
region between 300 and 800 nm. This specific application 
would generate new requirements that might be different from 
the ones upon the instrument has been designed on. 
Certain instruments permit some of their electrical (e.g. gains, 
integration time) and/or optical parameters (e.g. filters) to be 
easily modified. Special devices can change the default 
configuration as, for instance, field programmable gate array 
(FPGA) cards; few lines of code are updated before a mission 
starts. Such a technique can be used for both airborne missions 
and spacebome missions. 
The instrument variables that can be changed are (1) the 
integration time (or exposure time), (2) the frame period, and 
(3) the spectral on-chip binning (also called spectral on-chip 
averaging). Those parameters can be set in a way that the 
instrument can meet, whether it is possible, the new 
requirements dictated from the specific science application. 
Generally, the binning operation consists of summing lines up 
in a way that SNR can improve. The next sections (1) will 
explain further the tuning variable and (2) will show how an 
instrument configuration, being driven from a particular 
application, is generated. 2 
2. PROBLEM MODELING 
frame; a frame has M spatial pixels (M columns) and P spectral 
pixels (P rows). Figure 1 shows how the sensor generates such 
a cube. 
Figure 1 : Optical chain of a common hyperspectral pushbroom 
spectrometer. 
The model generates the application-driven sensor 
configurations by optimizing those two variables: 
• Integration time: the interval of time used to collect 
photons of light on a detector. The higher the 
integration time, the higher is the signal. 
• Spectral Binning (or spectral averaging or spectral 
zone mode): two or more spectral bands are summed 
up in a way that they form a unique row channel 
(Figure 2). If this summation is done by the hardware 
during image acquisition then is called on-chip 
binning, otherwise if it is performed offline by means 
of post processing algorithms is then called off-chip 
binning. In general, the higher the number of binned 
rows (bands), the higher is the spectral SNR. 
s 
p 
E 
C 
T 
R 
A 
L 
UNBINNED BINNED 
SPATIAL 
Scientists could require a hyperspectral sensor to flight for a 
specific application; it could then happen that the nominal 
instrument configuration is inadequate to such a purpose and 
must be modified in order to satisfy the scientific requirements 
of that particular application. An optimization tool is hereafter 
described. 
Generally, those spectrometers provide data under the form of 
hyperspectral cubes 8 . Such a cube has two spatial dimensions, 
i.e. the across-track dimension and the along-track dimension 
(provided by the motion of the platform), and one spectral 
dimension, representing the spectral signature of every spatial 
pixel. A hyperspectral cube with M across-track pixels, L along- 
track pixels, and P spectral bands is here considered. The plane 
formed by the across-track and the spectral dimensions is called 
Figure 2: Spectral Binning. The number of across-track spatial 
pixels is preserved whereas the bands (0,1,2) are binned to form 
band (0), bands (3,4) will form band 1 and so on. 
Spatial binning in the along-track and/or across-track direction 
can be also applied. We assume that the M x L x P cube 
dimensions refer to a sensor operated in the unbinned 
configuration; therefore the Mx P size of the frame would most 
likely correspond to the one of the detector. The main detector 
architectures taken here in account are: 
• A Charge Coupled Device (CCD). 
• A Complementary Metal Oxide Semiconductor 
(CMOS).