Enabling Wide Adoption of Hyperspectral Imaging

Abstract

Hyperspectral imaging systems capture information in multiple wavelength bands across the electromagnetic spectrum, providing substantial details of the materials present in the captured scene. However, the high cost of hyperspectral cameras makes the technology out of reach for end-user and small-scale commercial applications. The goal of my research is to enable hyperspectral imaging on mobile devices. In this extended abstract, I present the direction of research that I have followed during the first half of my PhD, along with ideas and work in progress for the second half. I propose a new system, called MobiSpectral, that turns a mobile device to a simple (hyper) spectral camera by extending its spectral sensitivity. I design new APIs for developers to write hyperspectral mobile applications. My main API is based on a deep-learning model to convert the captured images to hyperspectral images with multiple bands across the entire visible and near-infrared spectral range, revealing hidden information and enabling a myriad of new applications on mobile devices. My method is robust and can work in different illumination conditions.