Scholarlink

<p>Achieving robust well-focused ISAR imaging in diverse observation scenarios is critical for practical applications. However, the available iterative algorithms require elaborate parameter tuning, while model-driven networks necessitate model retraining for different observation conditions, limiting their practicality. To tackle these issues, this paper proposes a novel sparse Bayesian learning network, dubbed robust gamma process Laplace network (RGaPLN), for ISAR imaging in complex environments. Firstly, our previously proposed 2D inverse-free gamma process Laplace (2D-IFGaPL) algorithm is unfolded into a deep network to eliminate the need for parameter tuning. Then, a convolutional neural network (CNN) is integrated into the unfolded network to enhance robustness against variations in signal-to-noise ratio (SNR). Furthermore, a hypernetwork is designed to dynamically generate optimal parameters for different data missing rate (DMR), enabling ISAR imaging without model retraining under varying SNR and DMR conditions. Experimental results have demonstrated the effectiveness and superiority of the proposed method under various SNR and DMR conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

High-resolution ISAR imaging based on robust gamma process Laplace network

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Yujie Zhang,
  • Xueru Bai,
  • Feng Zhou

摘要

Achieving robust well-focused ISAR imaging in diverse observation scenarios is critical for practical applications. However, the available iterative algorithms require elaborate parameter tuning, while model-driven networks necessitate model retraining for different observation conditions, limiting their practicality. To tackle these issues, this paper proposes a novel sparse Bayesian learning network, dubbed robust gamma process Laplace network (RGaPLN), for ISAR imaging in complex environments. Firstly, our previously proposed 2D inverse-free gamma process Laplace (2D-IFGaPL) algorithm is unfolded into a deep network to eliminate the need for parameter tuning. Then, a convolutional neural network (CNN) is integrated into the unfolded network to enhance robustness against variations in signal-to-noise ratio (SNR). Furthermore, a hypernetwork is designed to dynamically generate optimal parameters for different data missing rate (DMR), enabling ISAR imaging without model retraining under varying SNR and DMR conditions. Experimental results have demonstrated the effectiveness and superiority of the proposed method under various SNR and DMR conditions.