Filamentous target segmentation of weft micro-CT image based on U-Net.

Abstract

Textile fabrics inspection is an important part of textile manufacturing industry, more and more new inspections technologies are adopted for the application. Micro CT imaging technology is recently explored for textile material inspection. This paper proposed a method of weft micro-CT image segmentation based on U-Net, by using deep learning theory and X-ray micro-CT nondestructive detection technology to realize automatic segmentation of filamentous objects. Firstly, the weft micro-CT image was obtained by X-ray micro-CT scan, and then the segmentation target was manually divided. A high segmentation accuracy CT image segmentation dataset of textile materials was built for training the network model. Based on the original U-Net, through experimental exploration, the attention mechanism was introduced, and the encoder module, decoder module and loss function module were adjusted, so as to get a good segmentation effect. The experimental results show that the segmentation performance superiority of this proposed algorithm and the Dice similarity coefficient reaches 0.843. The method proposed in this paper provides a direction for the combination of deep learning technology and micro-CT technology in industrial detection.