945
CONSTRUCTION OF STAR CATALOGUE BASED ON SVM
ZHANG Rui a , JIANG Ting 3 ’*
3 Zhengzhou Institute of Surveying and Mapping,, 450052,
China - j effor@ 163. com
3 Institute of Surveying and Mapping, Information Engineering University, Zhengzhou, 450052, China - tjiang@371.net
Commission I, WG1/6 - Small Satellites
KEY WORDS: Navigation-star catalogue, Satellite attitude, Machine-learning, SVM
ABSTRACT:
The method of constructing navigation star catalogue always based on Magnitude Filtering Method (MFM). But it did not work well
because of two typical disadvantages. On one side it would extract so many stars that there was redundancy in the catalogue. And on
the other side it would generate “hole” in some area of celestial sphere. In this article, Support Vector Machine (SVM) was
introduced into extracting navigation stars from basic catalogue. After using the new method on SAO catalogue, it was proved that
taking SVM as the method of extracting navigation-stars has good prospection.
1. PREFACE
Because of its lightness, economy and high precision in
confirmation of satellite’s attitude, star sensor is becoming
mainly instruments for measuring attitude on space probes. The
essential of measuring satellite’s attitude based on star sensor is
confirming the instantaneous orientation of star sensor axes in
celestial sphere coordinate system based on starlight vectors,
and then ascertaining the attitude of satellite. In above processes
the star recognition is the key process. So far, except the
recognition method based on neural-network, all kinds of
methods promoting by scholars need to consider the
construction of star catalogues. At present the popular way of
constructing star catalogues is filtering based on star’s light,
which only reserving the stars lighter than the giving threshold.
In some sense, it is only a kind of simple linearity classification
method, which cannot ensure that there are plenty of stars in
every watch-field because the classification only relay on a
giving threshold. In some fields there must be so many stars
which makes the recognition redundancy and in some fields
there is even none stars which appear a hole. So how to select
navigation stars and construct a self-contained and even star
catalogues based on the distributing of stars become the basis of
star recognition.
2. CONSTRUCTION PRINCIPLE OF STAR
CATALOGUES
2.1 Research Objective
In star catalogue, except for the coordinates in celestial sphere,
it includes the anniversary proper motion, anniversary and
long-term movement of all the stars. At present, the star
catalogues which used widely in space applications are as
bellows: SAO, FK4 and Goffin catalogues etc. In the SAO
catalogue, it records more than 260,000 stars’ information,
including position, brightness and spectrum. Specially, the
position precision of stars can achieve 10 8 , but brightness
precision is only 0.1. Because the position precision is more
important than brightness precision in determining attitude of
satellite, we select the SAO catalogue as basic catalogue.
In order to construct a self-contained and even star catalogues, it
is necessary to find out the distributing rules of all the stars
firstly. With the increasing of the threshold, the number of stars
appeared in every FOV (Field Of View) increase obviously.
Given the threshold of star sensor as 7.0, only when every FOV
include more than 4 stars, can the catalogue be self-contained,
and there will be more than 16,000 stars in the catalogue. It is
difficult for storing the catalogue and matching star pairs
effectively. Generically, star sensor can detect stars lighter than
6.0-6.5, so we can design appropriate FOV to make
self-contained star catalogue. To assure that there are more than
4 stars in every FOV, the FOV of star sensor should be larger
than 14°xl4°. According to normally star sensor, which the
FOV is 8°x8°, there will be 11% FOVs that cannot complete
star recognition, because the numbers of stars in them are lack
of 4P1.
2.2 Distributing Rule of Stars
From the discussion above, if we want to determine attitude of
satellites based on star sensors, there may be only two choices:
using the star sensors which have large FOV or increasing the
number of stars that storing in the satellites. But both of the two
methods are low efficient. So we have to resolve the problem by
different ways. Given the FOV of star sensor is 8°x8°, if in
every FOV it must have stars more than some given number (N),
so we can determine that in every FOV what their thresholds are,
and what the distributing rule of thresholds in the whole
celestial sphere is. When N=4 the distributing rule can be shown
by figure. 1.
From the above figure, the distributing rule of thresholds
actually is a spacial irregular plane. If we consider the
distributing rule of thresholds as optimized plane that it can
devise all stars into two types: navigation stars or
non-navigation stars, so the problem of selection of navigation
stars can switch to the problem of seeking non-linearity
optimized classified plane. It is very difficult for some linearity
classified methods, but it is advantage of SVM.