Fully Convolutional Deep Network Architectures for Automatic Short Glass Fiber Semantic Segmentation from CT scans

TL;DR

This paper introduces deep fully convolutional neural networks for automatic semantic segmentation of short glass fibers in CT scans, demonstrating superior performance over existing methods at various resolutions.

Contribution

It is the first to apply deep learning architectures to segment short glass fibers from CT scans at medium and low resolutions, using both 2D and 3D kernels.

Findings

Neural networks outperform existing segmentation methods.

Effective at both medium and low resolution scans.

Applicable to synthetic and real CT datasets.

Abstract

We present the first attempt to perform short glass fiber semantic segmentation from X-ray computed tomography volumetric datasets at medium (3.9 {\mu}m isotropic) and low (8.3 {\mu}m isotropic) resolution using deep learning architectures. We performed experiments on both synthetic and real CT scans and evaluated deep fully convolutional architectures with both 2D and 3D kernels. Our artificial neural networks outperform existing methods at both medium and low resolution scans.