Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.676562
Title: Random access spectral imaging
Author: Kelleher, Patrick
ISNI:       0000 0004 5372 9807
Awarding Body: University of Glasgow
Current Institution: University of Glasgow
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
A salient goal of spectral imaging is to record a so-called hyperspectral data-cube, consisting of two spatial and one spectral dimension. Traditional approaches are based on either time-sequential scanning in either the spatial or spectral dimension: spatial scanning involves passing a fixed aperture over a scene in the manner of a raster scan and spectral scanning is generally based on the use of a tuneable filter, where typically a series of narrow-band images of a fixed field of view are recorded and assembled into the data-cube. Such techniques are suitable only when the scene in question is static or changes slower than the scan rate. When considering dynamic scenes a time-resolved (snapshot) spectral imaging technique is required. Such techniques acquire the whole data-cube in a single measurement, but require a trade-off in spatial and spectral resolution. These trade-offs prevent current snapshot spectral imaging techniques from achieving resolutions on par with time-sequential techniques. Any snapshot device needs to have an optical architecture that allows it to gather light from the scene and map it to the detector in a way that allows the spatial and spectral components can be de-multiplexed to reconstruct the data-cube. This process results in the decreased resolution of snapshot devices as it becomes a problem of mapping a 3D data-cube onto a 2D detector. The sheer volume of data present in the data-cube also presents a processing challenge, particularly in the case of real-time processing. This thesis describes a prototype snapshot spectral imaging device that employs a random-spatial-access technique to record spectra only from the regions of interest in the scene, thus enabling maximisation of integration time and minimisation of data volume and recording rate. The aim of this prototype is to demonstrate how a particular optical architecture will allow for the effect of some of the above mentioned bottlenecks to be removed. Underpinning the basic concept is the fact that in all practical scenes most of the spectrally interesting information is contained in relatively few pixels. The prototype system uses random-spatial-access to multiple points in the scene considered to be of greatest interest. This enables time-resolved high resolution spectrometry to be made simultaneously at points across the full field of view. The enabling technology for the prototype was a digital micromirror device (DMD), which is an array of switchable mirrors that was used to create a two channel system. One channel was to a conventional imaging camera, while the other was to a spectrometer. The DMD acted as a dynamic aperture to the spectrometer and could be used to open and close slits in any part of the spectrometer aperture. The imaging channel was used to guide the selection of points of interest from the scene. An extensive geometric calibration was performed to determine the relationships between the DMD and two channels of the system. Two demonstrations of the prototype are given in this thesis: a dynamic biological scene and a static scene sampled using statistical sampling methods enabled by the dynamic aperture of the system. The dynamic scene consisted of red blood cells in motion and also undergoing a process of de-oxygenation which resulted in a change in the spectrum. Ten red blood cells were tracked across the scene and the expected change in spectrum was observed. For the second example the prototype was modified for Raman spectroscopy by adding laser illumination, a mineral sample was scanned and used to test statistical sampling methods. These methods exploited the re-configurable aperture of the system to sample the scene using blind random sampling and a grid based sampling approach. Other spectral imaging systems have a fixed aperture and cannot operate such sampling schemes.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.676562  DOI: Not available
Keywords: QC Physics
Share: