Skip to main content

Hyperspectral imaging on mobile devices

Resource type
Thesis type
(Thesis) Ph.D.
Date created
2025-12-10
Authors/Contributors
Author: Sharma, Neha
Abstract
Hyperspectral imaging systems capture information in multiple wavelength bands across the electromagnetic spectrum. These bands provide substantial details based on the optical properties of the materials present in the captured scene. The high cost of hyperspectral cameras and their strict illumination requirements make the technology out of reach for end-user and small-scale commercial applications. In this thesis, we propose multiple ideas to bring hyperspectral features to smartphones and enable many new mobile applications that were not possible before. Enabling hyperspectral imaging on smartphones, however, faces several research challenges. The first challenge is the lack of information beyond the visible range. Hyperspectral applications rely on the optical properties of the materials, especially in the near infrared (NIR) range. However, regular smartphone cameras use cutoff filters to truncate all signals in that range. We propose four camera design solutions for obtaining NIR signals, which utilize various depth-sensing technologies in modern smartphones. The second challenge is the limited number of bands captured by smartphone cameras, which are insufficient for computing accurate spectral signatures needed for identifying different materials and the subtle differences in their compositions and structures. We design a deep learning model to convert images captured by regular smartphone cameras to hyperspectral bands across the visible and NIR range. The third challenge is the wide diversity of illuminations in the environments in which smartphones operate, which introduces significant inconsistencies and distortions in the captured images. We propose image normalization methods to mitigate the adverse effects of the mixed and diverse illumination sources commonly found in everyday environments such as homes and grocery stores. We integrate all proposed components in a mobile hyperspectral imaging system, called MobiSpectral, and we deploy it on multiple smartphones to demonstrate its practical feasibility. We conduct extensive experimental studies to show the accuracy of MobiSpectral and compare its performance against a high-end hyperspectral camera that costs tens of thousands of dollars. To demonstrate the versatility and potential of MobiSpectral, we design multiple novel applications on smartphones to address problems of critical importance in society, including detecting fraud in labeling organic foods, monitoring blood glucose non-invasively, and localizing and visualizing human veins for accurate needle insertion. Through objective and ethics-approved subjective studies involving diverse participants and utilizing various mobile devices and illumination sources, we demonstrate the accuracy, robustness, and generality of the proposed applications.
File
Extent
127 pages.
Identifier
etd24145
Copyright statement
Copyright is held by the author(s).
Permissions
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Academic Supervisor
Thesis advisor: Hefeeda, Mohamed
Language
English
Member of collection
Download file Size
etd24145.pdf 115.95 MB