Fiber Identification in Cross Section of Blended Yarn on Back Propagation Neural Network. (December 2021)

• Record Type: Journal Article
• Authors: Rui, Zhang; Ding, Na; Lu, Xin-pin; Xu, Ying-qi; Xin, Bin-jie
• Is Part Of: AATCC Journal of Research. Volume 8(2021)Supplement 2
• Page Start: 95
• Page End: 99
• Publication Date: 2021-12
• Subjects: Cross Section -- Fiber Identification -- Flame Resistant -- Neural Network
• DOI: 10.14504/ajr.8.S2.19
• Languages: English
• ISSNs: 2472-3444

Abstract

An intelligent recognition algorithm was developed to identify fibers in the cross sections of blended yarn containing meta-aramid 1313 (Nomex), poly(phenylene-1, 3, 4-oxadiazole) (POD), flame resistant viscose, and flame-resistant vinylon. The yarn cross section image was obtained at x400 magnification. Drawing software was used to manually isolate single fiber images for training the back propagation (BP) neural network model in Matlab language image processing software. The GrabCut algorithm was used to de-noise the image and separate the target from the background. Finally, single fiber images and fiber distributions were obtained through the program. The result showed that the BP neural network model with the GrabCut algorithm can identify fiber type in a complex background more easily and more accurately than traditional algorithms.