Home // DATA ANALYTICS 2020, The Ninth International Conference on Data Analytics // View article

DCGAN-Based Data Augmentation for Enhanced Performance of Convolution Neural Networks

Authors:
Christian Reser
Christoph Reich

Keywords: Convolutional Generative Adversarial Network; Steel Surface Damage; Augmentation; Image Classification; Neural Network; Industry 4.0

Abstract:
The quality of steel is essential for many products. Unfortunately, during the production process of steel, surface defects (scratches, inclusions, etc.) occur, resulting in financial losses for steel producers. Therefore, to find and classify surface damage at the earliest stage of the steel production process to take actions for mitigating quality is preferred. Recently neural networks show the usefulness of image classification. Prerequisite is a large data set. But to collect a large data set often takes too long and is too expensive. This paper investigates how to handle smaller data sets, generate artificial data by augmentation and evaluate their efficiency. Of special interest is the augmentation of images by Deep Convolutional Generative Adversarial Networks (DCGANs). A detailed evaluation and comparision with other augmention techniques show that DCGAN augmentation outperformed other augmentations in accuracy and loss, but it is no replacement for a large data set.

Pages: 47 to 53

Copyright: Copyright (c) IARIA, 2020

Publication date: October 25, 2020

Published in: conference

ISSN: 2308-4464

ISBN: 978-1-61208-816-7

Location: Nice, France

Dates: from October 25, 2020 to October 29, 2020