Scholarlink

<p>Laser phase retrieval plays a crucial role in the evaluation of laser beam quality, yet model-based approaches are often limited by their sensitivity to initial conditions and susceptibility to local minima. To address these challenges, we propose AttentionPD-ResUNet, a phase retrieval framework that integrates the Phase Diversity (PD) method with attention mechanisms. Specifically, focused and defocused intensity images acquired via PD are employed as inputs to the network, which incorporates SE channel recalibration, ASPP-based multi-scale sampling, and spatial attention modules. This design enables the establishment of an end-to-end nonlinear mapping from the measured intensity distributions to the underlying wavefront phase. In comparative experiments, the proposed method achieves an RMSE of 0.068<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\lambda \)</EquationSource> </InlineEquation> and an MAE of 0.041<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\lambda \)</EquationSource> </InlineEquation> relative to the ground truth, with an average inference time of 0.41 s, thereby presenting a promising approach for reliable laser beam quality assessment.</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

Attentionpd-resunet: a laser phase retrieval network based on attention phase diversity

  • 期刊论文

  • 发表日期：

    2025年12月23日

查看全文
  • Chen Li,
  • Wenbo Jing,
  • JiaHe Meng,
  • Haili Zhao,
  • Haoyang Bai,
  • Mingzhe Song,
  • Zilong Di

摘要

Laser phase retrieval plays a crucial role in the evaluation of laser beam quality, yet model-based approaches are often limited by their sensitivity to initial conditions and susceptibility to local minima. To address these challenges, we propose AttentionPD-ResUNet, a phase retrieval framework that integrates the Phase Diversity (PD) method with attention mechanisms. Specifically, focused and defocused intensity images acquired via PD are employed as inputs to the network, which incorporates SE channel recalibration, ASPP-based multi-scale sampling, and spatial attention modules. This design enables the establishment of an end-to-end nonlinear mapping from the measured intensity distributions to the underlying wavefront phase. In comparative experiments, the proposed method achieves an RMSE of 0.068 \(\lambda \) and an MAE of 0.041 \(\lambda \) relative to the ground truth, with an average inference time of 0.41 s, thereby presenting a promising approach for reliable laser beam quality assessment.