The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part BI. Beijing 2008
2.4 In-flight normalization method
The difficulty is to acquire in flight images to measure for
each detector the response to the same input radiances.
After a pre-processing that globally shifts each column of the
raw image, we get an image that contains all needed
information as shown in Fig 7. This means that every row
contains the set of detectors response to the same landscape.
The first approach is to image uniform snowy expanses located
in Greenland and Antarctic at different radiances depending on
the latitude and the season. These areas are used to compute the
linear normalization coefficients of the SPOT family satellites
[1]. However, they suffer such a cloudy weather that a typical
normalization campaign lasts 2 to 3 weeks to get useable cloud-
free scenes at a single radiance level. Adding a constraint to get
several radiance levels by means of selecting areas according to
the local sun elevation would lead to very tedious operations.
The new approach is to make all the detectors view the same
points of the scene, following the same trajectory on the ground,
one after the other. The region of interest is therefore no longer
a uniform region, but rather a non-uniform varied region to
measure the response to several input radiances. The
equalization is thus performed in a single pass. The so-called
AMETHIST method [2] is based on a specific guidance also
called normalization steered viewing mode.
2.5 “Rotated retina” guidance
As the sensor consists of rectilinear arrays, a ‘rotated retina’
guidance is required and was defined [3] thanks to a 90° yaw
angle, with acquisition performed in the direction of the rows.
In the classical push-broom viewing mode (Fig 5) successive
lines of the image correspond to successive part of the scene.
Detector projection
Ground
velocity
Classical push-broom viewing mode
Figure 5 : The push-broom principle
In the normalization steered viewing mode (Fig 6) successive
lines of the image correspond to the same part of the scene with
a translation of one detector ground projection. This induces a
deformation of the raw image, with the useful data distributed
according to series of diagonal lines (Fig 7).
Detector projection f
Ground
velocity
Normalization steered viewing mode
Figure 6 : Rotated retina guidance
Figure 7 : raw image acquired with a 90° yaw angle and the
corresponding useful area
PA and XS lines of sight projection
X in meter
Figure 8 : PA and XS sensors projection
AMETHIST guidance law is optimized to superimpose on the
ground several predefined detectors (for instance, the centre of
each array). Because of the focal plane architecture and camera
distortion effects, the only way to make the consecutive
projections of the line of sight superimposed is to perform two
quasi-circular trajectories, one for the PA retina and one for the
XS retina,. Nevertheless, the PA arrays tilt causes geometrical
residuals of about 4 PA pixels. Hence, the whole detectors do
not perfectly see the same landscape, but the TDI device
average the acquired data and compensate most of these
residuals effects.
2.6 Computing normalization parameters from steered
viewing mode images
Regardless of geometrical disturbances, we may use each single
row of the shifted image as a measurement of the detectors
responses to the same input radiance L. This approach would
lead to a great sensitivity to radiometric noise and mis
registration. This is why an histogram matching method is
preferred, because it will not be sensitive to a single pixel
location : the only hypothesis is to put in front of each detector
the same collection of radiance levels. If all detectors behaved
the same way, all histograms hj[Z]computed on each column j
would be identical. Differences between column histograms are
due to relative sensitivities among detectors (Fig 9). After
normalization, all histograms should be identical.
15