[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

Journal of Information Science and Engineering, Vol. 37 No. 3, pp. 653-678

Performance Analysis of Improved Swarm Intelligence Based Classifier for Fabric Defect Detection

1Department of Electronics and Communication Engineering
Bannari Amman Institute of Technology
Tamil Nadu, 638401 India

2Department of Electronics and Communication Engineering
PSG College of Technology
Tamil Nadu, 641004 India
E-mail: gnanam1987@gmail.com

Automatic defect detection in fabrics is one of the most essential systems used in the textile industry to check the quality of the fabric. In most of the existing systems, a learning-based approach is implemented for defect detection in simple patterned fabrics. In this paper, swarm intelligence-based Backpropagation Neural Network (BPNN) classifiers are implemented for defect detection in complex patterned fabrics. But the problem with existing Binary Particle Swarm Optimization (BPSO) based BPNN classifier is premature convergence. To offset this problem an evolutionary state-based greedy reset is proposed to promote an effective and efficient search of the particles in the search space of the BPSO algorithm. The proposed system comprises of feature extraction phase followed by a performance evaluation phase. The combinations of features namely (i) Gray Level Co-occurrence Matrix (GLCM); (ii) Discrete Wavelet Transform (DWT) and GLCM (WGLCM); (iii) DWT, Local Binary Pattern (LBP), and GLCM (WL-GLCM) are extracted from the complex patterned fabrics and their performances are evaluated by employing swarm intelligence-based Backpropagation Neural Network (BPNN) classifier. The proposed system is validated with fabric datasets taken from the TILDA fabric database. From the results, it is observed that proposed system classification accuracy is 99.75% and it is better than the existing work with 77% reduced features.

Keywords: gabor filter, discrete wavelet transform, local binary pattern, binary particle swarm optimization, backpropagation neural networks

  Retrieve PDF document (JISE_202103_10.pdf)