Neural-fuzzy classification for fabric defects

Image classification by a neural-fuzzy system is presented for normal fabri cs and eight kinds of fabric defects. This system combines the fuzzificatio n technique with fuzzy logic and a back-propagation learning algorithm with neural networks. Four input features-the ratio of projection lengths in th e horizontal and vertical directions, the gray-level mean and standard devi ation of the image, and the large number emphasis (LNE) based on the neighb oring gray level dependence matrix for the defect area-are selected and the ir usefulness is justified. The neural network is also implemented and comp ared with the neural-fuzzy system. The results demonstrate that the neural- fuzzy system is superior to the neural network in classification ability.