Complex-Domain Semantic Segmentation of Spacecraft Directly from ISAR Echoes

Area of Science:

• Spacecraft engineering
• Radar signal processing
• Computer vision

Background:

• Semantic segmentation of Inverse Synthetic Aperture Radar (ISAR) images is vital for on-orbit spacecraft maintenance.
• Conventional methods lose scattering information and require extensive manual annotation.
• Existing approaches inadequately utilize spacecraft structural priors for accurate segmentation.

Purpose of the Study:

• To develop a novel complex-domain semantic segmentation framework for ISAR echoes.
• To address limitations of existing ISAR semantic segmentation methods, including information loss and high annotation costs.
• To improve the accuracy and efficiency of spacecraft semantic segmentation using ISAR data.

Main Methods:

• Proposed a One-Stop Segmentation (OSS) framework utilizing ISAR echoes directly.
• Introduced an Automatic ISAR Labeling (AIL) method based on ISAR scattering characteristics.
• Developed the One-Stop Segmentation Network (OSSNet) with a Domain Alignment Module (DAM) and Multi-Perspective Attention (MPA) framework (including SCA and SBA modules).

Main Results:

• The OSS framework achieved a mean Intersection over Union (mIoU) of 92.13%.
• The framework obtained a mean F1-score of 95.75% in ISAR spacecraft semantic segmentation.
• Demonstrated superior performance compared to existing methods on a simulated ground-based radar dataset.

Conclusions:

• The proposed OSS framework effectively performs semantic segmentation directly on ISAR echoes, preserving crucial information.
• OSSNet, with its DAM and MPA modules, significantly enhances segmentation accuracy by leveraging structural priors.
• The framework offers a more efficient and accurate solution for intelligent safety maintenance of on-orbit spacecraft.