Neural Network based classification of car seat fabrics

Indexed by SCOPUS Abstract —Car seat fabrics are uniquely fashioned textiles. A number of them is branded by a sponged-like appearance, characterized by spots and slightly discoloured areas. Their surface anisotropy is considered to be a relevant aesthetic feature since it has a strong impact on customer perceived quality. A first-rate car seat fabric requires a ―small‖ quantity of spots and discoloured areas while fabrics characterized either by a large number or by a low number of spots, are considered to be of lower quality. Therefore, car seat fabric quality grading is a relevant issue to be dealt with downstream to the production line. Nowadays, sponged-like fabric grading is performed by human experts by means of manual inspection and classification; though this manual classification proves to be effective in fabric grading, the process is subjective and its results may vary depending on the operator skills. Accordingly, the definition of a method for the automatic and objective grading of sponged-like fabrics is necessary. The present work aims to provide a computer-based tool capable of classifying sponged-like fabrics, as closely as possible to classifications performed by skilled operators. Such a tool, composed by an appositely devised machine vision system, is capable of extracting a number of numerical parameters characterizing the fabric veins and discoloured areas. Such parameters are, then, used for training an Artificial Neural Network (ANN) with the aim of classifying the fabrics in terms of quality. Finally, a comparison between the ANN-based classification and the one provided by fabric inspectors is performed. The proposed method, tested on a validation set composed by 65 sponged-like fabrics, proves to be able to classify the fabrics into the correct quality class in 93.8% of the cases, with respect to the selection provided by human operators.