Mineral classification using SWIR hyperspectral imaging: From classical machine learning techniques to transformers.
| dc.contributor.advisor | Rojas Norman, Alejandro José | es |
| dc.contributor.advisor | Garcés Hernández, Hugo Omar | es |
| dc.contributor.author | Cifuentes Ramírez, José Ignacio | es |
| dc.date.accessioned | 2025-08-19T18:52:19Z | |
| dc.date.available | 2025-08-19T18:52:19Z | |
| dc.date.issued | 2025 | |
| dc.description | Tesis presentada para optar al grado de Magíster en Ciencias de la Ingeniería con mención en Ingeniería Eléctrica. | es |
| dc.description.abstract | This work provides a comprehensive review of supervised models for hyperspectral imag ing, from classical machine learning techniques to state-of-the-art Transformers. Additionally, this research proposes a multispectral optical sensor that harnesses capabilities from SWIR hyperspectral to SWIR-multispectral for mineral classification using classical machine learning techniques. Furthermore, a novel 1D-RMC model was described using a solid mathematical foundation, thereby ensuring its implementation and performance viability based on previous results levering the spectral domain. Experiments were carried out using three datasets: one containing nine classes of minerals, the second containing reflectance images of 130 samples of 76 distinct minerals, and third, to ensure the 1D-RMC capabilities, the standard hyperspectral dataset "Indian Pines" and "Houston2013.” The results showed that it is feasible to build an optical sensor with five channels to obtain competitive results compared with hyperspectral data using machine learning techniques. Furthermore, the proposed deep learning models out performed the traditional machine learning algorithms by at least 3% in terms of accuracy and F1-score for the mineral dataset. Finally, 10 well-known models, including Transfomers, were compared with 1D-RMC, obtaining competitive results with state-of-the-art backbone network models, suprassing by at least 2% in terms of Overall Accuracy (OA), average accuracy (AA), and Kappa (κ) metrics. | en |
| dc.description.campus | Concepción | es |
| dc.description.departamento | Departamento de Ingeniería Eléctrica | es |
| dc.description.facultad | Facultad de Ingeniería | es |
| dc.description.sponsorship | ANID, Proyecto FONDECYT 1211184 | es |
| dc.description.sponsorship | ANID, Proyecto FONDECYT 1220903 | es |
| dc.description.sponsorship | ANID, Basal FB008 | es |
| dc.identifier.doi | https://doi.org/10.29393/TMUdeC-114CJ1MC114 | |
| dc.identifier.uri | https://repositorio.udec.cl/handle/11594/12960 | |
| dc.language.iso | en | en |
| dc.publisher | Universidad de Concepción | es |
| dc.rights | CC BY-NC-ND 4.0 DEED Attribution-NonCommercial-NoDerivs 4.0 International | en |
| dc.rights.uri | https://creativecommons.org/licenses/by-nc-nd/4.0/ | |
| dc.subject | Minerals Classification | en |
| dc.subject | Hyperspectral imaging | en |
| dc.subject | Electric transformers | en |
| dc.title | Mineral classification using SWIR hyperspectral imaging: From classical machine learning techniques to transformers. | en |
| dc.type | Thesis | en |