Fast and adaptive hyperspectral imaging for biological and biochemical characterization

• Main Author: Kubal Sharvaj
• Other Authors: Dalton Tay Chor Yong
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Materials
• Online Access: https://hdl.handle.net/10356/138798
• Institution: Nanyang Technological University

Hyperspectral images comprise of light intensity information resolved into two spatial dimensions and a spectral (wavelength) dimension. Hyperspectral imaging (HSI) provides much more information than regular optical microscopy and spectroscopy, making it valuable for non-invasive materials characterization in areas such as cell monitoring and food safety. However, HSI is challenging because of the large amount of data that has to be acquired. Traditional scanning methods suffer large measurement times, and possibly, data storage costs. Hence there is a need for subsampling approaches to HSI, such that measurements can be made more efficient without losing critical information. Standard compressive sensing approaches to hyperspectral imaging can achieve this, albeit subject to tradeoffs between image reconstruction accuracy, speed and generalizability to different samples. A promising approach to compressive HSI is adaptive basis scan, which overcomes these tradeoffs by achieving high-accuracy, generalizable imaging and fast reconstruction. However, existing adaptive methods are developed for imaging architectures that are inherently slow – the single spectrometer pixel camera, which can measure only a single spectrum at once. Here, we develop two methods to integrate multi-track spectral measurement with adaptive basis scan algorithms. We design and employ compound patterns on a DMD (digital micromirror device), which together with a multitrack acquisition architecture, can sample multiple wavelet coefficients at once. Simulation results show that the methods developed here are significantly faster than non-adaptive compressive HSI and full sampling HSI, without compromise to reconstruction accuracies across the different sample images tested.