Knitted fabric defect classification for uncertain labels based on Dempster-Shafer theory of evidence

► Uncertainty in labels of learning data is considered. ► Different types of features provide complementary information from data. ► The uncertainties in the initial labels of the learning data in each feature space are detected using local information of the data. ► Dempster-Shafer theory is used in a neural network ensemble structure to reduce the accepted uncertainty.