Análise de Imagens Hiperespectrais para reconhecimento de depósitos de minério de ferro utilizando aprendizado não supervisionado

  • Érica S. Pinto Departamento de Ciências da Computação, Universidade Federal de Ouro Preto, MG
  • Gustavo Pessin Instituto Tecnológico Vale, Ouro Preto, MG
  • Alan K. Rêgo Segundo Departamento de Controle e Automação, Universidade Federal de Ouro Preto, MG
Keywords: Hyperspectral Images, Remote Sensing, Iron ore, Hyperion, Unsupervised learning, K-Means

Abstract

Hyperspectral Imaging Systems (HSI) collect spectral data and spatial images simultaneously and continuously in hundreds of narrow spectral bands to generate a data cube. Specifically in mineralogy, remote sensing techniques using hyperspectral images are becoming increasingly popular on remote mapping of mineral formations. Datasets containing hyperspectral images from sensors such as Hyperion installed on the NASA’s Earth Observing satellite 1 (EO-1) can be found online. Considering these facts, we propose in this work the use of analysis and pattern recognition techniques in data cubes from HSI systems, more specifically from EO-1 Hyperion satellite. The hyperspectral image analysed includes the Carajás mine (Pará, BR) for the purpose of reconnaissance and identification of iron ore deposits using unsupervised learning and K-Means clustering method. The results show that about 6.72% of the analysed area is composed of iron ore, which corresponds to approximately 51km2.
Published
2022-10-19
Section
Articles