3137 Sensors and Materials, Vol. 32, No. 10 (2020) 3137–3155
MYU Tokyo
S & M 2329
*
Corresponding author: e-mail: cjlin@ncut.edu.tw
https://doi.org/10.18494/SAM.2020.2771
ISSN 0914-4935 © MYU K.K.
https://myukk.org/
Uniform Experimental Design for Optimizing the Parameters
of Multi-input Convolutional Neural Networks
Cheng-Jian Lin,
1*
Chen-Hsien Wu,
1
Chi-Chia Sun,
2,3
and Cheng-Hsien Lin
4
1
Department of Computer Science and Information Engineering, National Chin-Yi University of Technology,
Taichung 411, Taiwan
2
Department of Electrical Engineering, National Formosa University, Yunlin 632, Taiwan
3
Smart Machinery and Intelligent Manufacturing Research Center, National Formosa University,
Yunlin 632, Taiwan
4
Department of Electrical Engineering, National Chung Hsing University, Taichung City 402, Taiwan
(Received January 17, 2020; accepted May 25, 2020)
Keywords: image recognition, gender classification, convolutional neural network, uniform
experimental design
In this paper, a multi-input convolutional neural network (CNN) based on a uniform
experimental design (UED) is proposed for gender classification applications. The proposed
multi-input CNN uses multiple CNNs to obtain output results through individual training
and concatenation. In addition, to avoid using trial and error for determining the architecture
parameters of the multi-input CNN, a UED was used in this study. The experimental results
confirmed that the dual-input CNN with a UED achieved accuracies of 99.68 and 99.06% for
the CIA and MORPH datasets, respectively. The accuracy of the proposed CNN increased
significantly when increasing the number of inputs.
1. Introduction
In traditional machine learning methods, image features must be defined and captured by
the user in advance.
(1,2)
Recently, convolutional neural networks (CNNs) have been used to
automatically capture features for overcoming the aforementioned problem. Therefore, CNNs
have been widely and successfully applied in image recognition,
(3–5)
speech recognition,
(6)
colorimetric models,
(7)
and face recognition.
(8,9)
CNNs are the most commonly used
architecture for deep learning, and they exhibit superior performance in image recognition.
In 1998, LeCun et al. proposed the first CNN architecture called LeNet-5
(10)
and applied
this architecture to handwriting recognition. However, owing to problems such as excessive
parameters, gradients, and lack of hardware equipment, the costs of the architecture exceeded
its benefits. Deep learning was not popular with users in 1989. Krizhevsky et al. proposed
the AlexNet
(11)
architecture and introduced the dropout method
(12)
to prevent the network from
falling into overfitting. Many researchers have proposed deep CNN architectures. Popular
CNNs, such as GoogLeNet, have been proposed by Szegedy et al.
(13)
and Simonyan and
Zisserman.
(14)