Procesamiento eficiente y profundo de imágenes hiperespectrales de la observación remota de la Tierra y aplicaciones en tareas de clasificación

Abstract

Advances in computing technology and remote sensing field have fostered the development of powerful spectrometers that are able to collect large volumes of hyperspectral data. These data are characterized by their high spectral resolution, recording solar radiation and absorption of surface materials by measuring different wavelengths, along the electromagnetic spectrum. As a result, hyperspectral imaging (HSI) is a hot topic in remote sensing data analysis due to the vast amount of information comprised by this kind of images, which allows for a better characterization and exploitation of the Earth surface by combining rich spectral and spatial information. However, HSI poses major challenges for classification methods due to the high dimensionality of the data and the limited availability of training samples. In this context, deep learning (DL) methods arise as an interesting solution to enhance the HSI data processing and classification, reaching promising results in a wide range of applications within computer vision tasks. This thesis focus its efforts in the development of new and eficient DL approaches for HSI classification, processing not only spectral information but also spatial and spectral-spatial features from original HSI data cube and providing more robust solutions to overfitting, high dimensionality, data anomalies and data variability. To illustrate the advantages and benefits of the implemented proposals in comparison with the current state-of-the-art in HSI land cover classification, several experiments have been conducted considering real HSI scenes and performing the corresponding comparison with the available processing methods in the literature.