Evolutionary structure optimization of hierarchical neural network for image recognition

Satoru Suzuki, Yasue Mitsukura

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

The purpose of this paper is to optimize the structure of hierarchical neural networks. In this paper, structure optimization is used to represent a neural network by the minimum number of nodes and connections, and is performed by eliminating unnecessary connections from a trained neural network by means of a genetic algorithm. We focus on a neural network specialized for image recognition problems. The flow of the proposed method is as follows. First, the Walsh-Hadamard transform is applied to images for feature extraction. Second, the neural network is trained with the extracted features based on a back-propagation algorithm. After neural network training, unnecessary connections are eliminated from the trained neural network by means of a genetic algorithm. Finally, the neural network is retrained to recover from the degradation caused by connection elimination. In order to validate the usefulness of the proposed method, face recognition and texture classification examples are used. The experimental results indicate that a compact neural network was generated, maintaining the generalization performance by the proposed method.

Original languageEnglish
Pages (from-to)28-36
Number of pages9
JournalElectronics and Communications in Japan
Volume95
Issue number3
DOIs
Publication statusPublished - 2012 Mar

Fingerprint

Structure Optimization
Hierarchical Networks
Image recognition
Image Recognition
Evolutionary Optimization
Neural Networks
Neural networks
optimization
genetic algorithms
Genetic algorithms
Genetic Algorithm
Walsh Transform
Walsh transforms
Hadamard Transform
Hadamard transforms
Texture Classification
Backpropagation algorithms
Back-propagation Algorithm
Face recognition
Face Recognition

Keywords

  • face recognition
  • genetic algorithm
  • neural network
  • texture classification

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Computer Networks and Communications
  • Physics and Astronomy(all)
  • Signal Processing
  • Applied Mathematics

Cite this

Evolutionary structure optimization of hierarchical neural network for image recognition. / Suzuki, Satoru; Mitsukura, Yasue.

In: Electronics and Communications in Japan, Vol. 95, No. 3, 03.2012, p. 28-36.

Research output: Contribution to journalArticle

@article{31926591e20046d89815719f7325d637,
title = "Evolutionary structure optimization of hierarchical neural network for image recognition",
abstract = "The purpose of this paper is to optimize the structure of hierarchical neural networks. In this paper, structure optimization is used to represent a neural network by the minimum number of nodes and connections, and is performed by eliminating unnecessary connections from a trained neural network by means of a genetic algorithm. We focus on a neural network specialized for image recognition problems. The flow of the proposed method is as follows. First, the Walsh-Hadamard transform is applied to images for feature extraction. Second, the neural network is trained with the extracted features based on a back-propagation algorithm. After neural network training, unnecessary connections are eliminated from the trained neural network by means of a genetic algorithm. Finally, the neural network is retrained to recover from the degradation caused by connection elimination. In order to validate the usefulness of the proposed method, face recognition and texture classification examples are used. The experimental results indicate that a compact neural network was generated, maintaining the generalization performance by the proposed method.",
keywords = "face recognition, genetic algorithm, neural network, texture classification",
author = "Satoru Suzuki and Yasue Mitsukura",
year = "2012",
month = "3",
doi = "10.1002/ecj.10384",
language = "English",
volume = "95",
pages = "28--36",
journal = "Electronics and Communications in Japan",
issn = "1942-9533",
publisher = "Scripta Technica",
number = "3",

}

TY - JOUR

T1 - Evolutionary structure optimization of hierarchical neural network for image recognition

AU - Suzuki, Satoru

AU - Mitsukura, Yasue

PY - 2012/3

Y1 - 2012/3

N2 - The purpose of this paper is to optimize the structure of hierarchical neural networks. In this paper, structure optimization is used to represent a neural network by the minimum number of nodes and connections, and is performed by eliminating unnecessary connections from a trained neural network by means of a genetic algorithm. We focus on a neural network specialized for image recognition problems. The flow of the proposed method is as follows. First, the Walsh-Hadamard transform is applied to images for feature extraction. Second, the neural network is trained with the extracted features based on a back-propagation algorithm. After neural network training, unnecessary connections are eliminated from the trained neural network by means of a genetic algorithm. Finally, the neural network is retrained to recover from the degradation caused by connection elimination. In order to validate the usefulness of the proposed method, face recognition and texture classification examples are used. The experimental results indicate that a compact neural network was generated, maintaining the generalization performance by the proposed method.

AB - The purpose of this paper is to optimize the structure of hierarchical neural networks. In this paper, structure optimization is used to represent a neural network by the minimum number of nodes and connections, and is performed by eliminating unnecessary connections from a trained neural network by means of a genetic algorithm. We focus on a neural network specialized for image recognition problems. The flow of the proposed method is as follows. First, the Walsh-Hadamard transform is applied to images for feature extraction. Second, the neural network is trained with the extracted features based on a back-propagation algorithm. After neural network training, unnecessary connections are eliminated from the trained neural network by means of a genetic algorithm. Finally, the neural network is retrained to recover from the degradation caused by connection elimination. In order to validate the usefulness of the proposed method, face recognition and texture classification examples are used. The experimental results indicate that a compact neural network was generated, maintaining the generalization performance by the proposed method.

KW - face recognition

KW - genetic algorithm

KW - neural network

KW - texture classification

UR - http://www.scopus.com/inward/record.url?scp=84863403856&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84863403856&partnerID=8YFLogxK

U2 - 10.1002/ecj.10384

DO - 10.1002/ecj.10384

M3 - Article

AN - SCOPUS:84863403856

VL - 95

SP - 28

EP - 36

JO - Electronics and Communications in Japan

JF - Electronics and Communications in Japan

SN - 1942-9533

IS - 3

ER -