Instance segmentation of 3D woven fabric from tomography images by Deep Learning and morphological pseudo-labeling

Abstract

In the field of composite materials, mesoscale modelings based on X-ray computed tomography are becoming ever more widespread. This tool, aiming to increase the fidelity of the descriptive modeling of textile geometry for Finite Elements Analysis (FEA), requires image processing to identify the different objects within the material. In the present study, we propose a novel Deep Learning based segmentation of yarns from tomographic images aiming to provide a complete descriptive modeling of fabrics. The instance segmentation is achieved through an original two-step approach: (i) the determination of labeled yarn paths, by a tracking algorithm on detected 2D points, based on custom neighbor rules (distance and slope), and regression of parametric curves onto selected points, and (ii) a semantic segmentation of the yarn sections. For the second step, in absence of manual labeling of the yarn envelopes, we propose the use of morphological pseudo-labeling for training a Deep Convolutional Neural Network (DCNN), in which the yarn sections are represented by their distance transform. This approach is applied on two samples of a dry 3D woven ply-to-ply angle-interlock (at low and high compaction levels).