Journal of Engineering and Applied Sciences
ISSN: Online 1818-7803ISSN: Print 1816-949x
107
Views
0
Downloads
Abstract
The following study presents the validation of an application developed in MATLAB® for data augmentation which allows to improve the training of convolutional neural networks. The validation is done by comparing the accuracy percentages in the prediction of a trained convolutional neural network with five databases augmented in a different way which allows to determine the characteristics of the training images that produce an increase in network recognition capacities. Each network trained was evaluated by confusion matrices and compared their activations against a test image where it was found that the network with the greatest recognition capacity depends on the changes generated by the data augmentation in the original images (rotations, crops, background changes) as well as the ratio of augmented images and the number of original images used by the data augmentation algorithm developed to produce a new database.
How to cite this article:
Robinson Jimenez Moreno, Javier Pinzon Arenas and Paula Catalina Useche M.. Implementation of a Data Augmentation Algorithm Validated by Means of the
Accuracy of a Convolutional Neural Network.
DOI: https://doi.org/10.36478/jeasci.2017.5323.5331
URL: https://www.makhillpublications.co/view-article/1816-949x/jeasci.2017.5323.5331