Performance Analysis of Improved Particle Swarm Optimization for Fabric Defect Detection

In the textile industries, automatic defect detection in fabric plays a vital role for checking the quality of the product. The main objective of this work is to develop a Modified Particle Swarm Optimization (MPSO) based classifier to improve the performance of the classifier for detecting defect in fabrics. The entire fabric defect detection system is developed in two phases namely feature extraction phase and feature selection phase followed by a classifier. In feature extraction phase, Discrete Wavelet Transform (DWT) based texture features are obtained from the pre-processed image. Gabor filters are used to enhance the quality of the input image and also reduce the noise. In the feature selection phase, Binary Particle Swarm Optimization (BPSO) and its modified versions are used to select the most predominant features among all the features. Finally the performance of the selected feature subset is evaluated using back propagation neural network classifier. The Modified version of PSO is developed by incorporating different functions for generating random numbers. Five bench mark functions such as Scaling, Square, Cone, Cube, and Cylinder are utilized for generating random numbers in MPSO. The performance of both the techniques is compared using the performance metrics: Accuracy, Error rate, Sensitivity, Specificity and F1 score. With the proposed method of fabric defect detection system 94% of accuracy is achieved with reduced number of features.