Banca de DEFESA: EDILTON TORRES DE ANDRADE

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : EDILTON TORRES DE ANDRADE
DATA : 03/05/2022
HORA: 14:00
LOCAL: Videoconferência
TÍTULO:
PÁGINAS: 85
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Elétrica
SUBÁREA: Telecomunicações
RESUMO:

The accuracy of Image Recognition through Convolutional Neural Networks depends on the number and diversity of scenes of the images used in the network training, thus often obtaining several images of the same class is a costly task for properly training these networks.

In this work, we evaluate the use of synthetic images generated from 3D Models for network training. For this purpose, 3D models of dogs of different breeds were used and, using a three-dimensional virtual environment, we obtained, in a synthetic way, images in different environments and camera positions.

The simulations performed showed that synthetic images can be used to recognize real objects, being an alternative for obtaining samples for training Convolutional Neural Networks. Also, by applying Machine Learning Interpretability techniques we can measure how the object of interest against the background or region of the image was considered in the classification and in this way diversify the image base to obtain better accuracy in the classification.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 2334927 - ANDRE KAZUO TAKAHATA
Membro Titular - Examinador(a) Interno ao Programa - 1964056 - JOAO HENRIQUE RANHEL RIBEIRO
Membro Titular - Examinador(a) Externo ao Programa - 072.241.308-42 - JOAO CARLOS LOPES FERNANDES - FATEC-SCS
Membro Suplente - Examinador(a) Interno ao Programa - 1544392 - ALINE DE OLIVEIRA NEVES PANAZIO
Membro Suplente - Examinador(a) Externo à Instituição - RAFAEL FERRARI - UNICAMP