研究目的
To overcome the trade-off between high resolving power and large field-of-view in digital in-line holographic microscopy by proposing a deep learning-based upscaling method.
研究成果
The deep learning-based upscaling method successfully enhances the resolution of digital in-line holographic microscopy while maintaining a large field-of-view, overcoming the traditional trade-off. It provides higher quality reconstructions and detailed 3D information, making it suitable for monitoring and analyzing particle and cell dynamics. Future work could focus on generalizing the network to various sample types and optimizing computational efficiency.
研究不足
The network is trained with simulated holograms of specific particle shapes, which may limit generalization to other sample types. The upscaling factor is fixed, requiring retraining for different factors, and computational time increases with image size and filter parameters. Experimental noise can distort high-frequency information, and the method relies on specific equipment and conditions.
1：Experimental Design and Method Selection:
The study employs digital in-line holographic microscopy (DIHM) combined with a deep convolutional neural network (SRCNN) for upscaling hologram images. The network is trained to map low-resolution holograms to high-resolution ones, enhancing spatial resolution while maintaining a large FOV. Theoretical models include the Fresnel-Kirchhoff diffraction formula for hologram reconstruction and the angular spectrum method.
2：Sample Selection and Data Sources:
Samples include polystyrene spherical particles (15 μm diameter), prolate particles (fabricated from polystyrene), and red blood cells (RBCs) from rats. Holograms are generated through computer simulations and experimental setups involving flow in tubes and static placements.
3：List of Experimental Equipment and Materials:
Equipment includes a collimated green laser (λ = 532 nm, 100 mW; Crystal Laser, USA), water immersion-type objective lenses (20×/NA=0.50 and 60×/NA=1.00; Nikon, Japan), a high-speed camera (Ultima-APX; Photron, Japan, pixel size: 17 μm), a syringe pump, a rotating device, and a personal computer with GPU (NVIDIA GeForce GTX 1070Ti). Materials include polystyrene particles, polyvinyl alcohol, glycerol, isopropanol, PDMS, NaCl solution, PBS, and refractive index matching fluids.
4：50 and 60×/NA=00; Nikon, Japan), a high-speed camera (Ultima-APX; Photron, Japan, pixel size:
4. Experimental Procedures and Operational Workflow: Holograms are recorded using the DIHM setup. For particles, they are placed or flowed in tubes, and holograms are captured. For RBCs, blood is diluted and flowed in micro-tubes. Hologram signals are extracted by subtracting background images. Reconstruction is performed using numerical methods based on diffraction formulas. The neural network is trained with simulated holograms and applied to experimental data for upscaling.
5：Experimental Procedures and Operational Workflow:
5. Data Analysis Methods: Performance is evaluated using peak signal-to-noise ratio (PSNR) and structural similarity. Data analysis involves comparing upscaled holograms with ground truth high-resolution images, assessing reconstruction quality at various depths, and analyzing light scattering patterns.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
