Classification of woven fabrics using principal components analysis (PCA).

Abstract

Intelligent Analysis Systems (IAS) are gaining recognition in developing countries, Uganda inclusive. This report describes the various components of the development of a woven fabric classification system. Multi-dimensional set of data of fabric physical properties from 15 fabric samples was used as the input data to the system. The data was analyzed using PCA extract the selected four PCs. Using the extracted PCs, k-means clustering was performed to obtain three fabric clusters (k=3) and (has fabric classes (A, B and C). The system was developed using a single layer feed forward backpropagation NN in the names of a perceptron. Training was done for 467 epochs and there after r-square was calculated to determine the performance of the modelling network. Linear regression was used for comparison of performance. Also the performance of linear regression was determined using r-square. The intelligent classification system has the potential of making the public textile market well balanced because the buyers/customers will not be exploited since they will be correct fabric classes to be purchased.