A low-cost integrated hyperspectral imaging sensor with full temporal and spatial resolution at VIS-NIR wide range

Hyperspectral imaging provides optical information with high-dimensional spatial-temporal-spectral data cubes revealing intrinsic matter characteristics. It has been widely applied in various intelligent inspection applications. The existing hyperspectral imaging systems employ individual optical prism or grating to separate different spectra, which require complicated optical design that takes a heavy load for integration. The emerging multispectral imaging sensor integrates narrow-band filters on each pixel element, wasting most light throughput and trapped in the rigorous tradeoff between spatial and spectral resolution. In this work, we report an on-chip computational hyperspectral imaging framework with full spatial and temporal resolution. By integrating different broadband filtering materials on the imaging sensor, the target spectral information is non-uniformly and intrinsically coupled on each pixel with high light throughput. Using intelligent reconstruction algorithms, multi-channel images can be recovered from each frame, realizing real-time hyperspectral imaging. Following such a framework, we for the first time fabricated a broadband (400-1700 nm) hyperspectral imaging sensor with an average light throughput of 71.8$\%$ (enabling 32.5 dB peak signal-to-noise ratio of spectral reconstruction on ColorChecker Classic chart) and 89 wavelength channels (10 nm interval within 400-1000 nm and 25 nm interval within 1000-1700 nm). The average spectral resolution achieves 2.65 nm at 400-1000 nm and 8.59 nm at 1000-1700 nm. The demonstrated spatial resolution is 2048x2048 pixels at 47 fps, with $\sim$3.43 arc minute resolving ability at a 39-degree field of view. We employed the prototype sensor to collect a large-scale hyperspectral image dataset (210 scenes over 8 categories), and demonstrated the sensor's wide application potentials on a series of experiments, including chlorophyll and sugar quantification for agriculture growth, blood oxygen and water quality monitoring for human health, and face recognition for social security. The integrated hyperspectral imaging sensor is only 33.9 grams in weight without an imaging lens, and can be assembled on various resource-limited platforms such as unmanned aerial vehicles and nanosatellites, or equipped with off-the-shelf optical systems such as a microscope for direct real-time hyperspectral imaging. The technique integrates innovations from multiple fields of material, integrated circuits, computer science, and optics. It transforms the general challenge of high-dimensional imaging from one that is coupled to the physical limitations of high-cost optics manufacture and complex system design to one that is solvable through agile computation.