DART-Net: Dual-domain Attention-Transformer Network for Synergistic Weak Component Enhancement and Noise Suppression in ISAR Imaging

作者：Jiale Huang, Xiaoyong Li, Lei Liu, Xiaoran Shi, Xueru Bai, Feng Zhou · 年份：2025 · DOI：10.1109/icspcc66825.2025.11194341 · 被引用次数：1 · 研究领域：Spectroscopy Techniques in Biomedical and Chemical Research、Advanced Optical Sensing Technologies、Fault Detection and Control Systems

Inverse Synthetic Aperture Radar (ISAR) imaging plays a crucial role in reconnaissance and target monitoring, yet faces significant challenges: critical components of observed targets often exhibit poor visibility in imaging results due to various uncertainties, while strong noise contamination further obscures weak scattering components. To address these issues, we proposes a Dual-domain Attention-Transformer Reconstruction Network (DART-Net) that simultaneously enhances weak target components and suppresses noise. Existing methods struggle to achieve optimal balance between noise suppression and weak component enhancement. Our approach integrates spatial-channel dual-domain attention mechanisms with Transformer’s global modeling capability, demonstrating remarkable advantages in synergistic optimization: 1) The spatial attention module precisely locates key weak component regions, while the channel attention dynamically adjusts feature channel weights to suppress noise-dominated information, effectively resolving the : "enhancement-induced perturbation versus denoising-induced distortion" dilemma in conventional methods; 2) The multi-head self-attention mechanism captures long-range dependencies to maintain structural consistency and texture fidelity under complex noise scenarios; 3) The end-to-end framework directly processes raw ISAR complex-valued data, preserving phase information to provide high-fidelity input. Experimental results on electromagnetic simulation data de...