Open AccessCashmere and wool identification based on convolutional neural network
Author(s) -
Luo Jun-li,
Kai Lü,
Yueqi Zhong,
BoPing Zhang,
Huizhu Lv
Publication year - 2021
Publication title -
journal of engineered fibers and fabrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.303
H-Index - 29
ISSN - 1558-9250
DOI - 10.1177/15589250211005088
Subject(s) - softmax function , convolutional neural network , wool , identification (biology) , pattern recognition (psychology) , fiber , artificial intelligence , artificial neural network , computer science , materials science , composite material , botany , biology
Wool fiber and cashmere fiber are similar in physical and morphological characteristics. Thus, the identification of these two fibers has always been a challenging proposition. This study identifies five kinds of cashmere and wool fibers using a convolutional neural network model. To this end, image preprocessing was first performed. Then, following the VGGNet model, a convolutional neural network with 13 weight layers was established. A dataset with 50,000 fiber images was prepared for training and testing this newly established model. In the classification layer of the model, softmax regression was used to calculate the probability value of the input fiber image for each category, and the category with the highest probability value was selected as the prediction category of the fiber. In this experiment, the total identification accuracy of samples in the test set is close to 93%. Among these five fibers, Mongolian brown cashmere has the highest identification accuracy, reaching 99.7%. The identification accuracy of Chinese white cashmere is the lowest at 86.4%. Experimental results show that our model is an effective approach to the identification of multi-classification fiber.