![Institutional Repository [SANDBOX]](images/dias.logo.png "Institutional Repository [SANDBOX]"){kind=logo}
Institutional Repository [SANDBOX]
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/4EA08039-4B00-4DD1-837C-1FB4258724CA | ||
| Year | 2018 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Stergiani Vastaroucha, "Smart spectral imaging for material identification", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2018 https://doi.org/10.26233/heallink.tuc.78983 | ||
| Appears in Collections |
Hyperspectral Imaging is a powerful analytical tool that enables the acquisition of a series of images in narrow spectral bands. This technique makes it possible to extract both spatial and spectral information about the scene under investigation. Therefore, it is widely used for nondestructive and non-invasive analysis in a variety of fields, ranging from food quality assessment to biomedical applications. Material Identification is the key to all these applications, which is achieved by using a library of spectral signatures of materials of interest and Spectral Similarity Measurements. Finding the right Spectral Similarity Measure is an important step and many studies have been conducted for their evaluation in terms of accuracy. However, in these studies the evaluation is made using members of the libraries (labeled data) and time performance is never assessed. This study proposes a series of steps that should be followed in Spectral Similarity Measures evaluation, including both labeled and unlabeled data for comparison. More specifically, reflectance measurements of various materials were gathered from online public spectral libraries to construct a database of spectral signatures. Furthermore, a series of Spectral Similarity algorithms were implemented and tested for their speed and accuracy. The accuracy of the algorithms has been assessed on the basis of their ability to produce a right match when an unknown spectrum is compared against the reference spectra of the database. The Spectral Similarity algorithms tested in this study are: a) SAM, b) ED, c) SID, d) SCA, e) SGA, f) SID-SAM, g) SID-SCA, h) AWN, i) SSS, j) NS3, k) SSD and l) SPM. Finally, hyperspectral measurements of skin lesions have been used as a test case for the final evaluation of the implemented comparison methods. The combination of a spectral database of references with the right Spectral Similarity algorithms can provide a valuable tool for material identification, with applications in a variety of scientific and industrial fields.
Copyright © DIAS 2013
