Yarn fault classification: a signal processing approach using multiple projections

摘要

A novel mathematical framework is used to classify the faults in different classes, viz. Neps, Thick Place and Thin Place in a spun yarn. Images were converted into one-dimensional signals using discretised version of Radon transformation and divided into training and testing sample sets. Karhunen-Loeve Transformation (KLT) basis is computed for training set for each fault class taking top six highest energy eigen vectors. Euclidean distance of the test sample signal from its projection on the KLT basis was found for each realisation and fault class. The classification of unknown fault class is performed in accordance with the Euclidean distance applied using various techniques. It has been established from the present study that the respective faults have different configurations both within and between the classes, and are differentiated on the basis of their profile.