Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toward Onboard AI-Enabled Solutions to Space Object Detection for Space Sustainability

Wenxuan Zhang, Peng Hu

TL;DR

The paper addresses the challenge of space object detection (SOD) for collision avoidance in congested LEO environments by exploring onboard vision-based deep learning approaches. It proposes three SE-enhanced GELAN-based detectors—GELAN-SE, GELAN-ViT-SE, and GELAN-RepViT-SE—that allocate separate local CNN and global ViT processing streams, augmented with squeeze-and-excitation (SE) blocks to recalibrate channel-wise features. On the SODv2 Unity-generated dataset, GELAN-ViT-SE achieves the highest $mAP_{50}$ (≈0.751) and competitive $mAP_{50:95}$ (≈0.274) while reducing compute and power relative to baselines, with RepViT-based variants offering favorable embedded efficiency. Jetson Orin Nano experiments show RepViT-SE as the most power-efficient option with solid accuracy, making it attractive for onboard SOD tasks under strict energy and compute constraints. Overall, the work demonstrates that SE-enhanced GELAN architectures can deliver robust, energy-conscious SOD suitable for onboard space sustainability applications, and points to RepViT-based configurations as practical candidates for embedded deployment.

Abstract

The rapid expansion of advanced low-Earth orbit (LEO) satellites in large constellations is positioning space assets as key to the future, enabling global internet access and relay systems for deep space missions. A solution to the challenge is effective space object detection (SOD) for collision assessment and avoidance. In SOD, an LEO satellite must detect other satellites and objects with high precision and minimal delay. This paper investigates the feasibility and effectiveness of employing vision sensors for SOD tasks based on deep learning (DL) models. It introduces models based on the Squeeze-and-Excitation (SE) layer, Vision Transformer (ViT), and the Generalized Efficient Layer Aggregation Network (GELAN) and evaluates their performance under SOD scenarios. Experimental results show that the proposed models achieve mean average precision at intersection over union threshold 0.5 (mAP50) scores of up to 0.751 and mean average precision averaged over intersection over union thresholds from 0.5 to 0.95 (mAP50:95) scores of up to 0.280. Compared to the baseline GELAN-t model, the proposed GELAN-ViT-SE model increases the average mAP50 from 0.721 to 0.751, improves the mAP50:95 from 0.266 to 0.274, reduces giga floating point operations (GFLOPs) from 7.3 to 5.6, and lowers peak power consumption from 2080.7 mW to 2028.7 mW by 2.5\%.

Toward Onboard AI-Enabled Solutions to Space Object Detection for Space Sustainability

TL;DR

The paper addresses the challenge of space object detection (SOD) for collision avoidance in congested LEO environments by exploring onboard vision-based deep learning approaches. It proposes three SE-enhanced GELAN-based detectors—GELAN-SE, GELAN-ViT-SE, and GELAN-RepViT-SE—that allocate separate local CNN and global ViT processing streams, augmented with squeeze-and-excitation (SE) blocks to recalibrate channel-wise features. On the SODv2 Unity-generated dataset, GELAN-ViT-SE achieves the highest (≈0.751) and competitive (≈0.274) while reducing compute and power relative to baselines, with RepViT-based variants offering favorable embedded efficiency. Jetson Orin Nano experiments show RepViT-SE as the most power-efficient option with solid accuracy, making it attractive for onboard SOD tasks under strict energy and compute constraints. Overall, the work demonstrates that SE-enhanced GELAN architectures can deliver robust, energy-conscious SOD suitable for onboard space sustainability applications, and points to RepViT-based configurations as practical candidates for embedded deployment.

Abstract

The rapid expansion of advanced low-Earth orbit (LEO) satellites in large constellations is positioning space assets as key to the future, enabling global internet access and relay systems for deep space missions. A solution to the challenge is effective space object detection (SOD) for collision assessment and avoidance. In SOD, an LEO satellite must detect other satellites and objects with high precision and minimal delay. This paper investigates the feasibility and effectiveness of employing vision sensors for SOD tasks based on deep learning (DL) models. It introduces models based on the Squeeze-and-Excitation (SE) layer, Vision Transformer (ViT), and the Generalized Efficient Layer Aggregation Network (GELAN) and evaluates their performance under SOD scenarios. Experimental results show that the proposed models achieve mean average precision at intersection over union threshold 0.5 (mAP50) scores of up to 0.751 and mean average precision averaged over intersection over union thresholds from 0.5 to 0.95 (mAP50:95) scores of up to 0.280. Compared to the baseline GELAN-t model, the proposed GELAN-ViT-SE model increases the average mAP50 from 0.721 to 0.751, improves the mAP50:95 from 0.266 to 0.274, reduces giga floating point operations (GFLOPs) from 7.3 to 5.6, and lowers peak power consumption from 2080.7 mW to 2028.7 mW by 2.5\%.
Paper Structure (19 sections, 2 equations, 4 figures, 4 tables)

This paper contains 19 sections, 2 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. CNN-based Detectors
  4. ViT-based Detectors
  5. Effect of SE block
  6. The Proposed Model
  7. Overview
  8. SE for Efficient Capacity Allocation
  9. Architecture
  10. SODv2 Generation
  11. Performance Evaluation
  12. Experimental Setup
  13. Performance Results on V100 GPU
  14. Inference Result on V100 GPU
  15. Result Discussion on V100 GPU
  16. ...and 4 more sections

Figures (4)

  • Figure 1: The architecture of the RepNCSPELAN4_SE module. The SE layer is highlighted with a distinct background color.
  • Figure 2: Architectures of the proposed GELAN-based models. The SE block is integrated into the RepNCSPELAN4_SE structure.
  • Figure 3: Bar charts of RAM usage on Jetson Orin Nano, with the 95% confidence level indicated for each model.
  • Figure 4: Bar charts of power usage on Jetson Orin Nano, with the 95% confidence level indicated for each model.