Compact Snapshot Hyperspectral Imaging With Neural Dispersion-Engineered Metalens

There is a growing demand for ultra-compact, low-cost, and high-fidelity snapshot hyperspectral imaging devices across various fields. However, conventional systems struggle to fulfill all these evolving requirements. Although a single diffractive optical element offers a compact solution, its degrees of freedom in light phase modulation are limited. Here, we exploit wavelength-dependent characteristics of the point spread function (PSF) to design a dispersion-engineered metalens that serves as a key element for hyperspectral imaging. The PSF of the metalens is tailored to undergo an extra lateral shift as a function of wavelength, thereby encoding richer spectral information. An efficient differentiable computational model is developed to simulate the hyperspectral imaging process, together with a downstream spectral reconstruction network. The metalens structure and the reconstruction network are subsequently optimized in an end-to-end deep learning joint optimization framework. We fabricated the optimal metalens using two-photon grayscale lithography and built a prototype hyperspectral imaging camera. Both indoor and outdoor experimental results validated its outstanding spatial-spectral reconstruction performance, highlighting the effectiveness in practical applications.