Suppression of stray light based on energy
information mining
TING SUN,
1,
*FEI XING,
2,4
JINGYU BAO,
2
SONGSONG JI,
1
AND JIN LI
3
1
Joint International Research Laboratory of Advanced Photonics and Electronics, Beijing Information Science & Technology University,
Beijing 100192, China
2
Department of Precision Instruments, Tsinghua University, Beijing 100084, China
3
Photonics and Sensors Group, Department of Engineering, University of Cambridge, 9 JJ Thomson Avenue, Cambridge CB3 0FA, UK
4
e-mail: xingfei@mail.tsinghua.edu.cn
*Corresponding author: sunting09@tsinghua.org.cn
Received 4 May 2018; revised 13 September 2018; accepted 28 September 2018; posted 2 October 2018 (Doc. ID 330989);
published 25 October 2018
The star tracker plays a critical role in precision aerospace missions due to its high accuracy, absolute attitude
output, and low power consumption. For an optical sensor, the problem of stray light is always an important
research issue. A star energy information mining method for stray light suppression is proposed in this study. The
gray-level co-occurrence matrix and k-nearest neighbor algorithm are adopted to identify the types of stray light
that enter the optical system. Effective recognition of the stray light types is an important premise for the follow-
ing steps. Then the parameters are optimized during background estimation. When star spots are extracted, the
local differential encoding combined with Levenshtein distance filtering is conducted to eliminate the interference
noise spots. The proposed algorithm can achieve accurate star spot extraction even when stray light exists in real
night sky observation experiments.
© 2018 Optical Society of America
https://doi.org/10.1364/AO.57.009239
1. INTRODUCTION
An attitude determination system is crucial for a spacecraft to
perform precise space missions [1–5]. Nowadays, requirements
for star trackers, especially miniaturized ones with high accu-
racy and high dynamic performance are increasing [6–11].
Since the star image is the only data source for the star tracker,
the presence of stray light is fatal for its performance. Therefore,
researches on the suppression of stray light for the star image are
urgent. Effective distinction among star spots, the background,
and the noise can provide essential information for subsequent
star identification and attitude determination.
The studies on the suppression of stray light of star trackers
mainly include the following: (1) design and optimization of
lens hood [12–15]—a well-designed lens hood that considers
the working orbit and exact space task can help the star tracker
overcome some stray light problems; (2) data fusion between a
star tracker and other sensors such as a gyroscope, and thus the
invalid star tracker data can be compensated [16,17]; (3) star
image processing—this is an important approach because the
star image is the only information source of the star tracker.
However, threshold optimization and noise elimination in
the processing of the star image with stray light still lack special
study and are crucial issues.
This work mainly focuses on a star extraction method
based on energy information mining. Effective extraction and
accurate centroid determination of the star spots cannot be ob-
tained without a well-optimized processing algorithm. Star im-
ages are initially classified based on the gray-level co-occurrence
matrix (GLCM) and k-nearest neighbor algorithm, with which
the parameters in the background estimation and the threshold
setting operation can be optimized. Then the local differential
encoding, combined with Levenshtein distance (L-Distance)
filtering, is adopted to distinguish the interference noise. The
entire flow chart of the proposed method is described in Fig. 1.
Through night sky experiments and multiframe verification,
the recognition results of the extracted star spots by the pro-
posed method can be validated.
2. CLASSIFICATION OF STAR IMAGE WITH
COMPLICATED BACKGROUND
A. Description of General Star Image
A typical star image has dozens of bright star spots and a dark
background. The signal-to-noise ratio (SNR) is generally be-
tween 20 dB and 50 dB. The star spot principally exists with
a Gaussian distribution within a window between 3 pixels ×
3 pixels to 5 pixels × 5 pixels, as shown in Fig. 2. The noise of
an active pixel sensor (APS) CMOS image sensor mainly in-
cludes thermal noise, photon shot noise, 1∕f noise, fix pattern
noise, dark current noise, nonuniformity noise, etc., which may
bring the change in gray distribution of the star image and form
Research Article
Vol. 57, No. 31 / 1 November 2018 / Applied Optics 9239
1559-128X/18/319239-07 Journal © 2018 Optical Society of America
Provided under the terms of the OSA Open Access Publishing Agreement