Non-Parametric Spatial Spectral Band Selection methods
| dc.contributor.advisor | Yuen, Peter W. T. | |
| dc.contributor.advisor | James, David | |
| dc.contributor.author | Torres, Ruben M | |
| dc.date.accessioned | 2023-05-31T09:04:44Z | |
| dc.date.available | 2023-05-31T09:04:44Z | |
| dc.date.issued | 2021-05 | |
| dc.description | © Cranfield University 2021. All rights reserved. No part of this publication may be reproduced without the written permission of the copyright owner | en_UK |
| dc.description.abstract | This project is about the development of band selection (BS) techniques for better target detection and classification in remote sensing and hyperspectral imaging (HSI). Conventionally, this is achieved just by using the spectral features for guiding the band compression. However, this project develops a BS method which uses both spatial and spectral features to allow a handful of crucial spectral bands to be selected for enhancing the target detection and classification performances. This thesis firstly outlines the fundamental concepts and background of remote sensing and HSI, followed by the theories of different atmospheric correction algorithms — in order to assess the reflectance conversion for band selection — and BS techniques, with a detailed explanation of the Hughes principle, which postulates the fundamental drawback for having high-dimensional data in HSI. Subsequently, the thesis highlights the performances of some advanced BS techniques and to point out their deficiencies. Most of the existing BS work in the field have exhibited maximal classification accuracy when more spectral bands have been utilized for classification; this apparently disagrees with the theoretical model of the Hughes phenomenon. The thesis then presents a spatial spectral mutual information (SSMI) BS scheme which utilizes a spatial feature extraction technique as a pre-processing step, followed by the clustering of the mutual information (MI) of spectral bands for enhancing the BS efficiency. Through this BS scheme, a sharp ’bell’-shaped accuracy-dimensionality characteristic has been observed, peaking at about 20 bands. The performance of the proposed SSMI BS scheme has been validated through 6 HSI datasets, and its classification accuracy is shown to be ~10% better than 7 state-of-the-art BS algorithms. These results confirm that the high efficiency of the BS scheme is essentially important to observe, and to validate, the Hughes phenomenon at band selection through experiments for the first time. | en_UK |
| dc.description.coursename | PHD | en_UK |
| dc.identifier.uri | https://dspace.lib.cranfield.ac.uk/handle/1826/19747 | |
| dc.language.iso | en | en_UK |
| dc.relation.ispartofseries | PHD;PHD-21-TORRES | |
| dc.rights | © Cranfield University, 2015. All rights reserved. No part of this publication may be reproduced without the written permission of the copyright holder. | |
| dc.subject | Hyperspectral imaging | en_UK |
| dc.subject | Remote sensing | en_UK |
| dc.subject | Band clustering | en_UK |
| dc.subject | Atmospheric correction | en_UK |
| dc.title | Non-Parametric Spatial Spectral Band Selection methods | en_UK |
| dc.type | Thesis | en_UK |