Related Experiment Video
Updated: Oct 3, 2025
14:23
Imaging Intermediate Filaments and Microtubules with 2-dimensional Direct Stochastic Optical Reconstruction Microscopy
Published on: March 6, 2018
11.0K
Deep Multi-Feature Transfer Network for Fourier Ptychographic Microscopy Imaging Reconstruction.
Xiaoli Wang1,2, Yan Piao1, Jinyang Yu2
1Information and Communication Engineering, Electronics Information Engineering College, Changchun University of Science and Technology, Changchun 130022, China.
Sensors (Basel, Switzerland)
|February 15, 2022
Summary
Fourier ptychographic microscopy (FPM) reconstruction is improved using a novel deep learning method. This technique enhances anti-noise performance and reduces data redundancy for high-resolution imaging.
Area of Science:
- Optics and Photonics
- Computational Imaging
- Machine Learning for Microscopy
Background:
- Fourier ptychographic microscopy (FPM) offers wide field-of-view, high-resolution, and quantitative phase imaging.
- Current FPM reconstruction methods struggle with noise and data redundancy, limiting performance.
- Advanced reconstruction techniques are crucial for maximizing FPM's potential.
Purpose of the Study:
- To develop a novel deep learning-based reconstruction method for Fourier ptychographic microscopy.
- To improve the anti-noise performance and reduce data redundancy in FPM imaging.
- To achieve high-resolution reconstruction with enhanced image quality.
Main Methods:
- A deep multi-feature transfer network utilizing ResNet50, Xception, and DenseNet121 for feature extraction.
- Cascaded feature fusion strategy for channel merging to enhance reconstruction quality.
- Pre-upsampling incorporated into the network for improved high-resolution image texture details.
Main Results:
- The proposed method demonstrates robust performance against noise and blurred images.
- Achieved superior reconstruction results with reduced acquisition time and lower resolution data.
- Validated effectiveness through both simulation and experimental data.
Conclusions:
- The deep multi-feature transfer network offers a powerful approach for FPM image reconstruction.
- The end-to-end neural network mapping provides a new perspective for solving FPM reconstruction challenges.
- This method enhances FPM's practical applicability by improving robustness and efficiency.
