结合位置校正的神经网络傅里叶叠层成像

谢新平; 李娟; 王红强

文章摘要

谢新平,李娟,王红强.结合位置校正的神经网络傅里叶叠层成像[J].安徽建筑大学学报,2025,33(4):56-64

结合位置校正的神经网络傅里叶叠层成像

Neural Network Model for Fourier Ptychographic Microscopy Combined with Position Correction

DOI：

中文关键词: 傅里叶叠层成像神经网络超分成像位置校正

英文关键词: Fourier ptychographic microscopy neural network super-resolution imaging position correction

基金项目:安徽省高校省级自然科学研究项目（2024AH050232）；安徽建筑大学中青年教师培养行动项目（JNFX2024029、YQYB2023011）；安徽建筑大学校科研平台开放课题项目（YCSJ2024ZR02）

作者	单位
谢新平	School of Mathematics and Physics，Anhui Jianzhu University，Hefei 230601，China
李娟	School of Mathematics and Physics，Anhui Jianzhu University，Hefei 230601，China
王红强	Institute of Intelligent Machines，Hefei Institutes of Physical Science，Chinese Academy of Sciences，Hefei 230031，China

摘要点击次数: 47

全文下载次数: 0

中文摘要:

傅里叶叠层显微成像技术（FPM）是一种结合了相位恢复、叠层成像、合成孔径的超分成像技术，但其重建质量易受LED位置偏差影响。针对现有先进的无需训练的深度神经网络FPM算法（FPMUP）忽略位置校正、固定小卷积核网络设计限制高频信息恢复等问题，提出了一种结合LED位置校正的神经网络FPM算法，即基于物理的超分成像技术（PBSR-PC）。PBSR-PC在 FPMUP框架基础上，首先构建位置校正模块，计算校正波矢，消除频域拼接误差；其次设计层级化递减卷积核网络结构，通过逐层缩小的感受野实现从低频全局特征到高频超分辨率细节的重建。PBSR-PC实现物理误差校正与超分辨率重建的协同优化。与其他 FPM重构方法相比，PBSR-PC不仅能校正位置，还提高了重构图像的质量。通过仿真和实验验证了该方法的有效性。

英文摘要:

Fourier ptychographic microscopy（FPM）is a super-resolution imaging technique that combines phase retrieval, ptychographic imaging, and synthetic aperture. However, its reconstruction quality is susceptible to LED position misalignment. To address the issues of existing advanced untrained deep neural network FPM algorithms（FPMUP）, such as ignoring position correction and the limitations of fixed small convolutional kernel network design in recovering high-frequency information, a neural network FPM algorithm combining LED position correction, namely physical-based super-resolution imaging technology（PBSR-PC）, is proposed. PBSR-PC is based on the FPMUP framework. Firstly, a position correction module is constructed to calculate the corrected wave vector, eliminating the frequency domain stitching error. Secondly, it designs a hierarchical decreasing convolutional kernel network structure, achieving reconstruction from low-frequency global features to high-frequency super-resolution details through progressively reduced receptive fields. PBSR-PC achieves the collaborative optimization of physical error correction and super-resolution reconstruction. Compared with other FPM reconstruction methods, PBSR-PC not only corrects the position but also improves the quality of the reconstructed image. The effectiveness of this method is verified through simulation and experiments.

查看全文查看/发表评论下载PDF阅读器

关闭