ARSRNet: accurate space object recognition using optical cross section
curves

Xia Wang; YuRong Huo; YuQiang Fang; Feng Zhang; Yifan Wu

doi:10.1364/AO.435304

Applied Optics
Vol. 60,
Issue 28,
pp. 8956-8968
(2021)
•https://doi.org/10.1364/AO.435304

ARSRNet: accurate space object recognition using optical cross section curves

Xia Wang, YuRong Huo, YuQiang Fang, Feng Zhang, and Yifan Wu

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Xia Wang, YuRong Huo, YuQiang Fang, Feng Zhang, and Yifan Wu, "ARSRNet: accurate space object recognition using optical cross section curves," Appl. Opt. 60, 8956-8968 (2021)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Related Topics
Table of Contents Category
- Surface Optics and Plasmonics
Optics & Photonics Topics
?

The topics in this list come from the Optics and Photonics Topics applied to this article.

About this Article
History
- Original Manuscript: June 30, 2021
- Revised Manuscript: September 2, 2021
- Manuscript Accepted: September 8, 2021
- Published: September 30, 2021

Abstract

Limited by the conditions and performance of ground-based optical observations, it is difficult for us to obtain a plethora of optical cross section (OCS) data for some space objects (SOs). Unevenly distributed OCS data and unclear labels will affect the performance of SOs recognition based on neural networks. Furthermore, when we need to identify a new SO or SO category using deep neural network, the trained network model may no longer be applicable. We need to retrain the network with new training data. In order to alleviate these problems and improve the generalization and training convergence speed of SOs recognition networks, a novel, to the best of our knowledge, neural network model, ARSRNet, is proposed in this paper. The ARSRNet can identify SOs and their attitude accurately using only a small quantity of training OCS data and without clear labels. And the proposed network is able to adapt to new recognition tasks. Meanwhile, we propose an AdamRprop network optimization algorithm to accelerate network training and improve recognition accuracy. Experimental results show that the recognition accuracy of ARSRNet reaches 90.60% on the test OCS dataset. Compared with mainstream network optimization algorithms, the proposed AdamRprop is more appropriate for ARSRNet and can accelerate the convergence of ARSRNet.

Full Article | PDF Article

More Like This

Classification for geosynchronous satellites with deep learning and multiple kernel learning

Yurong Huo, Zhi Li, Yuqiang Fang, and Feng Zhang
Appl. Opt. 58(21) 5830-5838 (2019)

Fast calculation method of complex space targets’ optical cross section

Yi Han, Huayan Sun, Yingchun Li, and Huichao Guo
Appl. Opt. 52(17) 4013-4019 (2013)

Residual grounding transformer network for terrain recognition on the lunar surface

Linwei Qiu, Haichao Li, Zhi Li, and Cheng Wang
Appl. Opt. 60(21) 6002-6014 (2021)

Previous Article Next Article

Data Availability

Data underlying the results presented in this paper are not publicly available at this time, but we will consider making it public after collecting more data.

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (17)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (6)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (13)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

1.	Input: OCS curves.
2.	Initialize: Training set $T = {(x_{1}, y_{1}), (x_{1}, y_{1}), \dots, (x_{N}, y_{N}) \| \forall y \in {1, 2, 3, \dots, M}}$ , where $N$ is the total number of samples
	in the training set, $M$ is the number of classes in the training set; the set $H_{k}$ is defined as a set in which all samples labeled
	$k$ , e.g., $T_{K}$ denotes a subset of $T$ where all samples labeled $k$ ; if the label $y_{a}$ of example $x_{a}$ and the label $y_{b}$ of example
	( $x_{b}, y_{b}$ ) are the same, i.e., $y_{a} = y_{b}$ , then $l_{a, b} = 1$ , otherwise $l_{a, b} = 0$ .
3.	Output: The loss $L_{φ, θ, ρ}$ for each training episode.
4.	Randomly select $C$ categories from training set $T$ ; $N_{C}$ is the number
	of classes selected.
5.	Randomly select support set: $S = {(x_{i}, y_{i})}_{i = 1}^{K * C}$ . The number of
	samples in support set is $N_{C}$ .
6.	for c in ${1, 2, 3, \dots, C}$ do
7.	Randomly select query set: $Q_{c} = {(x_{j}, y_{j})}_{j = 1}^{K^{q}}$ , where $y_{j} = c$ ; and
	the number of samples in $Q_{c}$ is $N_{Q}$ .
8.	Randomly select assist set: $A_{c} = {(x_{k}, y_{k})}_{k = 1}^{K^{a}}$ , where $y_{k} = c$ ; and
	the number of samples in $A_{c}$ is $N_{A}$ .
9.	end for
10.	Initialize loss: $L_{φ, θ, ρ} = 0$
11.	for c in ${1, 2, 3, \dots, C}$ do
12.	Compute the mean vector: $m_{c} = \frac{1}{N_{c}} \sum_{(x_{k}, y_{k}) \in A_{c}} E (x_{k}),$
13.	while $(x_{j}, y_{j})$ in $Q_{c}$ do
14.	Update loss: (5) $\begin{aligned} L_{φ, θ, ρ} & = L_{φ, θ, ρ} + \frac{1}{N_{C} N_{Q}} [l_{j, c} \log (s_{j, c}) + (1 - l_{j, c}) \log (1 - s_{j, c})] \\ + \sum_{(x_{i}, y_{i}) \in S_{c}} {(o_{i, j} - l_{i, j})}^{2} \end{aligned}$
15.	end while
16.	end for

			Size		Texture
Satellite	Oribt	Attitude	Body (volume)	Each Panel (length)	Body	Panel	Antenna	The rest
SAT 1	LEO	three-axis stabilization	$1 m^{2}$	5 m	Kapton	Gaas	Al	Lambert-plate
SAT 2	LEO	three-axis stabilization	$3 m^{2}$	4 m	Kapton	Gaas	Al	Lambert-plate
SAT 3	LEO	three-axis stabilization	$1 m^{2}$	2 m	Kapton	Gaas	Al	Lambert-plate
SAT 4	LEO	three-axis stabilization	$0.5 m^{2}$	1 m	Kapton	Gaas	Al	Lambert-plate
SAT 5	LEO	three-axis stabilization	$3 m^{2}$	5 m	Kapton	Gaas	Al	Lambert-plate

Material	${\bar{R}}_{d i f f}$	${\bar{R}}_{s p e c}$	$n_{u} (n_{v})$
Kapton	0.1418	0.8234	12455.8892
Gaas	0.2746	0.6427	2233.3101
Lambert-plate	0.635	0.0220	0.0719
Al	0.3685	0.5746	596.9783

Model	Accuracy	Accuracy of Ten-Way, One-Shot
SimReNets	90.60%	91.82%
Matching Nets [31]	90.13%	91.60%
MANN [28]	83.22%	88.01%
Siamese Nets with Memory [43]	88.44%	88.58%
Convolutional Siamese Nets [30]	85.70%	84.13%
MetaNets [29]	90.35%	91.23%
Prototypical Nets [32]	90.08%	91.59%

No.	Category	Total Number of Samples	Proportion
1	S1A1	237	89.2%
2	S1A2	237	90.0%
3	S1A3	237	99.7%
4	S2A1	237	81.3%
5	S2A2	237	77.4%
6	S2A3	237	88.5%
7	S3A1	237	83.4%
8	S3A2	237	69.0%
9	S3A3	237	80.1%
10	S4A1	237	99.9%
11	S4A2	237	88.7%
12	S4A3	237	66.0%
13	S5A1	237	74.9%
14	S5A2	237	55.3%
15	S5A3	237	80.2%

Abstract

Data Availability

Cited By

Figures (17)

Tables (6)

Equations (13)

Applied Optics