摘要
数字全息显微术中相位恢复面临噪声导致的解包裹误差与像差引入的相位畸变双重挑战,现有分步处理方法存在计算效率低与误差传递的固有局限,亟需发展同步求解新范式。为此,提出了一种基于多光谱通道U-Net(MScU-Net)的数字全息显微相位恢复方法。该网络在ResU-Net架构中引入Transformer增强的多光谱通道注意力机制,通过端到端非线性映射,直接从含噪、失真的包裹相位数据中实现相位解包裹与像差补偿的同步求解。实验结果表明,MScUNet在相位恢复中表现出显著优势,其中,峰值信噪比PSNR达29.82 dB,结构相似度SSIM保持0.96;背景区域噪声标准差(0.0004~0.0030 rad)较最优对比方法降低82.6%~95.8%;实测台阶高度系统偏差≤3.74%,测量局部波动达到亚纳米级(s=0.42 nm),该方法为微纳尺度的三维形貌测量提供了有效解决方案。
Phase retrieval in digital holographic microscopy contends with unwrapping errors induced by noise and phase distortions arising from aberrations.Traditional sequential approaches are constrained by limited computational efficiency and error propagation,motivating the development of synchronous solution paradigms.A multi-spectral channel U-Net(MScU-Net)is proposed for phase retrieval in digital holographic microscopy.The network embeds a Transformer-enhanced multi-spectral channel attention mechanism within a ResU-Net framework,enabling end-to-end nonlinear mapping that simultaneously performs phase unwrapping and aberration compensation directly from noisy,distorted wrapped phase maps.Experimental results demonstrate substantial improvements:peak signal-to-noise ratio(PSNR)attains 29.82 dB,structural similarity index measure(SSIM)is maintained at 0.96,and background noise standard deviation(0.0004-0.0030 rad)is reduced by 82.6%-95.8%relative to a state-of-the-art method.Step-height measurements show a systematic deviation of≤3.74%from white-light interferometry references,with local measurement fluctuations reaching sub-nanometer precision(s=0.42 nm).These findings indicate that MScU-Net offers an effective synchronous solution for accurate micro and nanoscale 3D topography measurement.
作者
严利平
苗小喆
黄柳
YAN Liping;MIAO Xiaozhe;HUANG Liu(School of Information Science and Engineering(School of Cyber Science and Technology),Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《光学精密工程》
北大核心
2025年第15期2441-2454,共14页
Optics and Precision Engineering
基金
浙江省自然科学基金资助项目(No.LQ23E050020)
浙江省领雁计划资助项目(No.2024C01174)
浙江省科技创新领军人才资助项目(No.2022R52052)。
关键词
数字全息显微
相位解包裹
像差补偿
多光谱通道注意力机制
digital holographic microscopy
phase unwrapping
aberration correction
multi-spectral channel attention mechanism
