Real-time hyperspectral video processing via artificial intelligent optoelectronic hardware

Abstract

Hyperspectral imaging has attracted significant attention to identify spectral signatures for image classification and automated pattern recognition in computer vision. State-of-the-art implementations of snapshot hyperspectral imaging rely on bulky, non-integrated, and expensive optical elements, which do not allow fast data processing, e.g., real-time and high-resolution videos. This work introduces Hyplextm, a CMOS-compatible, fast hyperspectral camera that replaces bulk optics with suitably designed artificial intelligent optical hardware components. Hyplextm does not require spectrometers but uses conventional monochrome cameras, allowing real-time and high-resolution hyperspectral imaging at inexpensive costs. In this invited talk, we discuss the design, implementation, and real-world applications of Hyplextm.