Agustín Lobo took part in the second edition of the workshop " Hyperspectral Sensing meets Machine Learning and Pattern Analysis (HyperMLPA)" which was held online during March 24-26 as a part of the scheduled scientific program of the Spectro Expo 2021. Lobo gave the talk entitled "Machine-Learning Classification Of Proximal Conventional And Hyper-Spectral Imagery" in which the researcher explained the work developed in collaboration with the company Lithica for the iTarg3T project.
The experience presented by Lobo is aimed to demonstrate the feasibility of a dynamic 2D mapping system of the mineralization and its ore grade in a gallery mine through hyperspectral images acquired as the excavation of the front progresses. These images are analyzed afterward using machine-learning techniques to identify the rocks and minerals.
"Our approach uses hyperspectral images in which each image layer corresponds with the quantity of light reflected by the material in a narrow wavelength interval. Thus, instead of having only the values of the light reflected in the red, green, and blue for each pixel as in conventional photography, we have a continuous spectrum between the 450 and 1500 nm", said Agusti Lobo.
"We have analyzed hand-samples of the rocks in the laboratory and we have conducted simulations in a dark room, reproducing the light conditions of the mine gallery. The hyperspectral camera was mounted on a rotor on the top of a tripod and scanned a panel with the rock mine samples from a similar distance of that would be taken in the mine", said Agustí Lobo.
"Having these maps from the excavation front will allow us to enhance the evaluation of the quantity and distribution of minerals. We will also be able to obtain a tomography which can be compared with the 3D model obtained by using other geophysical techniques", concludes Agustín Lobo.
Más información
Lobo, A.,García, E., Barroso, G., Martí, D., Fernández-Turiel, JL., Ibáñez Insa, J. (2021)Machine-Learning classification of proximal conventional and hyper-spectral imagery.