Enhanced 3D X-Ray Tomography: Deep Learning–Based Advanced Algorithms for Fiber Instance Segmentation

Identifying specific properties of fibers in paper hand sheets is a challenge being faced for many decades. In this chapter, some of the advanced algorithms for image segmentation to estimate these properties are systematically presented. The process of determining the appropriate operating conditions from the estimated properties is also elaborated. Moreover, the authors introduce a new machine learning algorithm designed for 3D X-ray tomography. This technique is utilized to produce images of intricate fiber structures. The novel four-step hybrid 3D fiber segmentation algorithm presented in the chapter involves deep learning–assisted semantic segmentation, which generates 2D images from 3D ones for fiber extraction. Additionally, the algorithm combines elliptical contour estimation with the marker-controlled watershed technique to separate fibers from the background area. By employing 3D reconstruction, individual fibers are identified. To validate the performance of this approach, the proposed methodology is implemented on a real-time sample of nylon fiber bundle under compression and its 3D X-ray image volume. The results demonstrate the algorithm’s superiority in terms of precision and efficiency when compared to off-the-shelf image processing algorithms.