Abstract—Counter Propagation Neural Network (CPN) is a hybrid neural network because it makes use of the advantages of supervised and unsupervised training methodologies. CPN has a reputation for high accuracy and short training time. In this paper, a variant of CPN, namely preprocessed Counter Propagation Neural Network is proposed. We propose that if some preprocessing can be introduced to assign weights instead of random weight assignment during CPN training, it will result in good classification accuracy, very short training time and simple model complexity. The preprocessed CPN has promising applicability in a number of domains, among which textile defect classification is a prominent one. Textile sector is the most prospective export sector in Bangladesh. We demonstrate the utility and capability of our preprocessed CPN classifier in automated textile defect classification in the context of Bangladesh. We have found very good results.

Index Terms—Textile Defect, Automated Inspection, Defect Classification, Counterpropagation Neural Network (CPN), Preprocessed CPN, Centroid, Accuracy

Cite: Mokhlesur Rahman and Tarek Habib, "A Preprocessed Counterpropagation Neural Network Classifier for Automated Textile Defect Classification," Journal of Industrial and Intelligent Information, Vol. 4, No. 3, pp. 209-217, May 2016. doi: 10.18178/jiii.4.3.209-217