Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SSTD: Stripe-Like Space Target Detection Using Single-Point Weak Supervision

Zijian Zhu, Ali Zia, Xuesong Li, Bingbing Dan, Yuebo Ma, Enhai Liu, Rujin Zhao

TL;DR

This work tackles stripe-like space target detection (SSTD) under limited labeling by introducing AstroStripeSet, a large synthetic dataset with diverse stray-light scenarios and multiple label formats. It then proposes a single-point weakly supervised teacher–student framework that starts from SAM-generated pseudo-labels, refines them via a LoRA-tuned StripeSAM teacher, and iteratively trains an evolving StripeNet student to expand high-quality labels; a GeoDice loss aligned to straight stripe geometry guides learning. The approach achieves state-of-the-art or fully supervised-comparable performance and shows strong zero-shot generalization to real space imagery, validating the methodology for practical space situational awareness tasks. By publicly releasing AstroStripeSet and code, the work provides a robust benchmark and a scalable framework for weakly supervised SSTD, with potential extensions to other segmentation challenges and cross-domain transfer learning.

Abstract

Stripe-like space target detection (SSTD) plays a key role in enhancing space situational awareness and assessing spacecraft behaviour. This domain faces three challenges: the lack of publicly available datasets, interference from stray light and stars, and the variability of stripe-like targets, which makes manual labeling both inaccurate and labor-intensive. In response, we introduces `AstroStripeSet', a pioneering dataset designed for SSTD, aiming to bridge the gap in academic resources and advance research in SSTD. Furthermore, we propose a novel teacher-student label evolution framework with single-point weak supervision, providing a new solution to the challenges of manual labeling. This framework starts with generating initial pseudo-labels using the zero-shot capabilities of the Segment Anything Model (SAM) in a single-point setting. After that, the fine-tuned StripeSAM serves as the teacher and the newly developed StripeNet as the student, consistently improving segmentation performance through label evolution, which iteratively refines these labels. We also introduce `GeoDice', a new loss function customized for the linear characteristics of stripe-like targets. Extensive experiments show that our method matches fully supervised approaches, exhibits strong zero-shot generalization for diverse space-based and ground-based real-world images, and sets a new state-of-the-art (SOTA) benchmark. Our AstroStripeSet dataset and code will be made publicly available.

SSTD: Stripe-Like Space Target Detection Using Single-Point Weak Supervision

TL;DR

This work tackles stripe-like space target detection (SSTD) under limited labeling by introducing AstroStripeSet, a large synthetic dataset with diverse stray-light scenarios and multiple label formats. It then proposes a single-point weakly supervised teacher–student framework that starts from SAM-generated pseudo-labels, refines them via a LoRA-tuned StripeSAM teacher, and iteratively trains an evolving StripeNet student to expand high-quality labels; a GeoDice loss aligned to straight stripe geometry guides learning. The approach achieves state-of-the-art or fully supervised-comparable performance and shows strong zero-shot generalization to real space imagery, validating the methodology for practical space situational awareness tasks. By publicly releasing AstroStripeSet and code, the work provides a robust benchmark and a scalable framework for weakly supervised SSTD, with potential extensions to other segmentation challenges and cross-domain transfer learning.

Abstract

Stripe-like space target detection (SSTD) plays a key role in enhancing space situational awareness and assessing spacecraft behaviour. This domain faces three challenges: the lack of publicly available datasets, interference from stray light and stars, and the variability of stripe-like targets, which makes manual labeling both inaccurate and labor-intensive. In response, we introduces `AstroStripeSet', a pioneering dataset designed for SSTD, aiming to bridge the gap in academic resources and advance research in SSTD. Furthermore, we propose a novel teacher-student label evolution framework with single-point weak supervision, providing a new solution to the challenges of manual labeling. This framework starts with generating initial pseudo-labels using the zero-shot capabilities of the Segment Anything Model (SAM) in a single-point setting. After that, the fine-tuned StripeSAM serves as the teacher and the newly developed StripeNet as the student, consistently improving segmentation performance through label evolution, which iteratively refines these labels. We also introduce `GeoDice', a new loss function customized for the linear characteristics of stripe-like targets. Extensive experiments show that our method matches fully supervised approaches, exhibits strong zero-shot generalization for diverse space-based and ground-based real-world images, and sets a new state-of-the-art (SOTA) benchmark. Our AstroStripeSet dataset and code will be made publicly available.
Paper Structure (16 sections, 5 equations, 6 figures, 5 tables)

This paper contains 16 sections, 5 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Geometric Feature-Based Approaches
  4. Pattern-Based Approaches
  5. Data-Driven Approaches
  6. AstroStripeSet Dataset
  7. Teacher-Student Label Evolution Framework using Single-Point Weak Supervision
  8. Initial Pseudo-Labels from Pre-Trained SAM
  9. StripeSAM as the Fine-Tuning Teacher
  10. StripeNet as the Evolving Student
  11. GeoDice Loss Function
  12. Experiments
  13. Dataset and Evaluation Metrics
  14. Results and Discussion
  15. Ablation Study
  16. ...and 1 more sections

Figures (6)

  • Figure A: Challenges in the SSTD task. (a), (c) and (e) show space images with different stray light, while (b), (d) and (f) are the stripe-like target's ground truths.
  • Figure B: Statistics for the AstroStripeSet dataset.
  • Figure C: An overall framework for our teacher-student label evolution with single-point weak supervision.
  • Figure D: Visual comparison of our method with the SOTA fully supervised network on the upcoming AstroStripeSet. (Detected true targets, miss detections, and false alarms are highlighted in red, blue, and yellow, respectively).
  • Figure E: Results from three label evolution iterations.
  • ...and 1 more figures