17:15 〜 18:30
[MIS02-P07] Track Image Clustering for Tanpopo Mission
キーワード：クラスタリング、たんぽぽ計画、宇宙生物学
Introduction
Tanpopo mission [1] is a series of astrobiological experiments conducted by JAXA on the International Space Station (ISS). The purpose of this series of experiments is to investigate the origin of life on Earth. One of the methods for this is collecting fine particles such as cosmic dust and space debris in the ISS orbit.
This capture experiment uses ultra-low density aerogel panels to capture particles [2]. The mission team analyzes the panels on the ground. In this study, we use the surface images of the penetration holes (in after referred to as "Tracks") on the aerogel panels taken in the initial analysis.
These tracks have different shapes depending on the penetrating material and velocity. The shape type is essential for sample cutting and estimating the penetrating material and serves as a label when stored in the sample database. However, since researchers classified them into shape types by qualitative judgment, the classification type may change as the analyses proceed. Also, they manage the data manually. This classification creates a gap between the new data and the past data, making data management labor. The classification must be realized based on the quantitative judgment through a machine rather than qualitative judgment to solve this problem. When adding new data, it will be possible to re-clustering all the previous data, saving time and effort in management.
Methods
The method used in this research is divided into the following steps.
1) Selection of the images
The camera takes the surface on the panel, and it captures from directly above with changing heights. So they have three-dimensional variable-length data. This step selects three images from the top, middle, and bottom for conversion to fixed-length.
2) Image processing
Since the images have low contrast in common, this step coordinates contrast. Moreover, since there is a difference in the number of tracks between classifications, augmentation and downsampling are performed.
3) Learning
This step use supervised and unsupervised learning for classification or clustering.
The results and discussion of these methods in this study are presented.
Reference
[1] A. Yamagishi et al., "Tanpopo: Astrobiology exposure and micrometeoroid capture experiments – proposed experiments at the exposure facility of ISSJEM ", Trans. JSASS Aerospace Tech. Jpn. vol. 12, No. ists29, p. Tk_49-Tk_55, 2014
[2] M. Tabata et al., "Ultralow-density double-layer silica aerogel fabrication for the intact capture of cosmic dust in low-Earth orbits", Journal of Sol-Gel Science and Technology volume 77, pages325-334(2016)
Tanpopo mission [1] is a series of astrobiological experiments conducted by JAXA on the International Space Station (ISS). The purpose of this series of experiments is to investigate the origin of life on Earth. One of the methods for this is collecting fine particles such as cosmic dust and space debris in the ISS orbit.
This capture experiment uses ultra-low density aerogel panels to capture particles [2]. The mission team analyzes the panels on the ground. In this study, we use the surface images of the penetration holes (in after referred to as "Tracks") on the aerogel panels taken in the initial analysis.
These tracks have different shapes depending on the penetrating material and velocity. The shape type is essential for sample cutting and estimating the penetrating material and serves as a label when stored in the sample database. However, since researchers classified them into shape types by qualitative judgment, the classification type may change as the analyses proceed. Also, they manage the data manually. This classification creates a gap between the new data and the past data, making data management labor. The classification must be realized based on the quantitative judgment through a machine rather than qualitative judgment to solve this problem. When adding new data, it will be possible to re-clustering all the previous data, saving time and effort in management.
Methods
The method used in this research is divided into the following steps.
1) Selection of the images
The camera takes the surface on the panel, and it captures from directly above with changing heights. So they have three-dimensional variable-length data. This step selects three images from the top, middle, and bottom for conversion to fixed-length.
2) Image processing
Since the images have low contrast in common, this step coordinates contrast. Moreover, since there is a difference in the number of tracks between classifications, augmentation and downsampling are performed.
3) Learning
This step use supervised and unsupervised learning for classification or clustering.
The results and discussion of these methods in this study are presented.
Reference
[1] A. Yamagishi et al., "Tanpopo: Astrobiology exposure and micrometeoroid capture experiments – proposed experiments at the exposure facility of ISSJEM ", Trans. JSASS Aerospace Tech. Jpn. vol. 12, No. ists29, p. Tk_49-Tk_55, 2014
[2] M. Tabata et al., "Ultralow-density double-layer silica aerogel fabrication for the intact capture of cosmic dust in low-Earth orbits", Journal of Sol-Gel Science and Technology volume 77, pages325-334(2016)