Yazar "Bounail, Khalid" seçeneğine göre listele

Listeleniyor 1 - 1 / 1
  • Küçük Resim
    Öğe
    Increasing spectral resolution of hyper spectral images while decreasing spectral variability using deep generative model
    (İstanbul Gelişim Üniversitesi Lisansüstü Eğitim Enstitüsü, 2023) Bounail, Khalid
    Hyperspectral imaging has become an indispensable tool for analyzing Earth's surface materials and extracting valuable information in various applications, including environmental monitoring, agriculture, and remote sensing. One fundamental task in hyperspectral analysis is spectral unmixing, which aims to decompose mixed pixel spectra into their constituent endmember spectra and corresponding abundances. Traditional spectral unmixing methods typically rely on linear models, assuming that the mixed pixel spectra are linear combinations of the pure spectral signatures. in this thesis, an approach to enhance the spectral resolution of hyperspectral images while simultaneously reducing spectral variability is explored by proposing a technique called Deep Generative Endmember Modeling (DGEM) applied to unsupervised spectral unmixing. The Linear Mixing Model (LMM) is commonly used for spectral unmixing, which aims to decompose the mixed spectral information in an image into individual pure spectral signatures called endmembers. However, traditional LMM approaches struggle with limited spectral resolution and high spectral variability, which can hinder accurate unmixing results. To address these challenges, a Deep Generative EM is proposed, which leverages deep generative models to learn the underlying structure of the endmembers. By employing a deep neural network architecture which is capable of capturing intricate relationships and generating high-resolution endmembers with reduced spectral variability. The proposed Model framework enhances the spectral resolution of hyperspectral images by estimating high-resolution spectral bands based on the learned endmember representations. Additionally, the generated endmembers exhibit reduced spectral variability, resulting in improved unmixing performance.