Portable deep learning singlet multi-spectral microscope. (February 2021)

• Record Type: Journal Article
• Authors: Gao, Jinming; Shen, Hua; Cui, Xiangqun; Zhu, Rihong
• Is Part Of: Optics and lasers in engineering. Volume 137(2021)
• Publication Date: 2021-02
• Subjects: Microscopy -- Optical design -- Computational imaging -- Portable imaging device; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2020.106378
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: A novel portable multi-spectral microscope is proposed by cooperating the aberration controlled singlet lens and deep learning computational imaging method. Propose a singlet lens design method with only rotational symmetric aberrations remaining to keep linear signal properties across a large FOV and reduce singlet lens fabrication difficulty. The trained deep learning neutral network is used to eliminate the controlled aberrations for improving resolution. Present a compact and cost-effective multi-spectral microscope with the designed singlet lens and trained neutral network (spectral channels: 7, resolution: 1.38 μm, size: 40 cm*20 cm*20 cm, and weight: ~500 g). Abstract: Compared with monochromatic or RGB imaging technologies, multi-spectral imaging technology is beneficial for obtaining more information details from images of different wavelengths with better contrast. A novel portable and cost-effective multispectral microscope is proposed, whose highlights are firstly imaging by a special designed singlet lens with only controlled rotational symmetric aberrations and then removing the controlled aberrations by deep learning computational imaging method to improve the resolution. Our method helps to reduce the extreme difficulties of singlet lens fabrication due to its simple surface produced by the unnecessarily supreme aberration optimization, while ensuring the resolution. In this manuscript, we introduce singlet lens design methods to connect Zernike polynomial coefficients with singlet lens parameters through wavefront aberrations. Then, the deep learning networks, parameters and working environments are provided to computationally enhance the resolution. By imaging a gold standard resolution pattern and typical bio-samples, experiment results demonstrate that our method and demo-setups achieve multi-spectral microscopy well and cost-effectively. It is believable that our portable deep learning singlet multi-spectral microscope would be an alternate to the conventional approaches and applied in biology, chemistry, environmental science, especially in resource-limited areas.