Um Novo Método de Fatoração Não Negativa em Modelos Bilineares: Aplicação em Separação de Fontes Espectrais

Alejandra J Inga Quezada, Yannick Deville, Leonardo Tomazeli Duarte

DOI: 10.14209/sbrt.2024.1571037391

Evento: XLII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2024)

Keywords:

Abstract

In hyperspectral remote sensing, sensors collect data that contain mixed pixels, which have mixed spectra of more than one pure material (endmembers). Recovering the spectra of these endmembers can be formulated as a blind source separation problem, for which non-negative matrix factorization (NMF) techniques have been used extensively. Non-linear extensions of NMF have been shown to be capable of dealing with non-linear aspects in certain spectral mixing processes (imaging urban areas). This article proposes a new non-negative factorization method for bilinear mixture models. Our proposal considers a projection operator that allows non-negativity to be maintained while preserving as much information as possible. Numerical experiments with two sets of synthetic hyperspectral data indicate that the proposal performs better than other methods in the literature.

Download