The goal of spectral imaging is to capture the spectral signature of a target. Traditional scanning method for spectral imaging suffers from large system volume and low image acquisition speed for large scenes. In contrast, computational spectral imaging methods have resorted to computation power for reduced system volume, but still endure long computation time for iterative spectral reconstructions. Recently, deep learning techniques are introduced into computational spectral imaging, witnessing fast reconstruction speed, great reconstruction quality, and the potential to drastically reduce the system volume. In this article, we review state-of-the-art deep-learning-empowered computational spectral imaging methods. They are further divided into amplitude-coded, phase-coded, and wavelength-coded methods, based on different light properties used for encoding. To boost future researches, we’ve also organized publicly available spectral datasets.

We demonstrate that a sharper focal spot area can be generated ($0.147{\lambda }^2$) by using an azimuthally polarized beam propagating through a vortex $0–2\pi$ phase plate than for radial polarization ($0.17{\lambda }^2$) or for linear polarization ($0.26{\lambda }^2$) under the same condition. Further research illustrates that such optimistic results can still be expected when condition limitations are liberalized. This will facilitate new approaches to get superresolution in confocal systems.

We experimentally demonstrated that the microsphere can discern the details of the object whose sizes are below the conventional diffractive limit and such super-resolution capability can be reinforced if semi-immersing the corresponding microspheres in liquid droplet, producing a sharper contrast and a comparatively smaller magnification factor. The microsphere is considered as a channel that connects the near-field evanescent wave and the transmission one in far field. A conjecture based on this is proposed to explain the mechanism of super-resolution and the corresponding phenomenon.

Many applications requiring both spectral and spatial information at high resolution benefit from spectral imaging. Although different technical methods have been developed and commercially available, computational spectral cameras represent a compact, lightweight, and inexpensive solution. However, the tradeoff between spatial and spectral resolutions, dominated by the limited data volume and environmental noise, limits the potential of these cameras. In this study, we developed a deeply learned broadband encoding stochastic hyperspectral camera. In particular, using advanced artificial intelligence in filter design and spectrum reconstruction, we achieved 7000–11,000 times faster signal processing and ~10 times improvement regarding noise tolerance. These improvements enabled us to precisely and dynamically reconstruct the spectra of the entire field of view, previously unreachable with compact computational spectral cameras.

A method using a freeform surface lens for LED secondary optic design is proposed in this paper. By Snell’s Law, the differential equations are given to build the relationship between the normal direction of a freeform surface and its input/output ray vectors. Runge–Kutta formulas are used to calculate the differential equations to design the freeform surface. Moreover, the optical model for uniform illumination is simulated and optical performance is analyzed. A practical freeform surface lens for LED uniform illumination is fabricated using an injection molding method. By the process, our system demonstrates a uniform illumination with a divergence half-angle of $6\mathrm{°}$ and an efficiency of 78.6%.