Banca de DEFESA: PÂMELLA REGINA FERNANDES DA COSTA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : PÂMELLA REGINA FERNANDES DA COSTA
DATE: 23/12/2022
TIME: 09:00
LOCAL: videoconferência https://www.youtube.com/c/PPGGUFRNTransmiss%C3%A3odeBancas
TITLE:

APPLIED NEURAL NETWORKS FOR TO CREATE PREDICTIVE GPR SECTIONS

KEY WORDS:

Predictive GPR Section; Deep Convolutional Neural Networks; Style Transfer; Video Frame Interpolation.

PAGES: 107
BIG AREA: Ciências Exatas e da Terra
AREA: Geociências
SUMMARY:

In the sedimentary geobodies geometries acquired with Ground Penetrating Radar (GPR) some limitations interfere with a more accurate reconstruction of geological bodies in the subsurface. Increasing the amount of GPR data and a consequent improvement in geometries identification, Deep Convolutional Neural Networks (DCNNs) techniques were applied: Style Transfer and Frame Interpolation, independently, as well as the two techniques association. Based on this methodology, given a set of three consecutive GPR sections, parallel and equidistant from each other, A, B and C, intermediate reflections were created from A and C GPR sections, here called predictive GPR sections, which seek to represent the geometries found in the B GPR section sections for the three distinct depositional geometries studied: (i) montiform geometry found in carbonate rock outcrops, Irecê Basin, Salitre Formation, located at Fazenda Arrecife (Bahia); (ii) siliciclastic rocks, Parnaíba Basin, Piauí Formation, located in Serra das Araras (Piauí) whose geometries vary from tabular to wedge-shaped; (iii) and the recent washover deposits located around Lagoa Mirim in Vila Taim (Rio Grande do Sul) that exhibit wedge-shaped geometries. From the difference in depth (deviations) obtained from the overlap between the real geometries and the predictive geometries, standard deviations and mean errors were calculated. Predictive GPR sections were created for all geological scenarios and techniques applied. However, it was only possible to fully identify the interest geometries applying Style Transfer and the two techniques association. Except for the outcrop of carbonate rocks, whose limits of the montiform geometries are in abrupt contact with the surrounding geometries which allowed the delimitation of the microbial body for the Frame Interpolation technique. GPR intermediary sections were also created, from the pixel-by-pixel average between A and C GPR sections. Designated them as “lower bound” and it was ensured that the geometries obtained with the predictive GPR sections created better results. Fortuitously, the predictive geometries obtained with Style Transfer were able to exhibit smaller differences, in relation to the geometries of the B sections, for each of the geological scenarios studied.

COMMITTEE MEMBERS:
Externo à Instituição - DIEGO DA COSTA MIRANDA - UFPA
Externo ao Programa - 347628 - ADRIAO DUARTE DORIA NETO - nullPresidente - 1161652 - FRANCISCO PINHEIRO LIMA FILHO