Matching Fiber Skeleton Junctions in Pulp Micrographs via Multi-Feature Aggregation

Ji Lin; Toqeer Ahmed; Li Feng; Pengfei Yin; Yiqi Liu

doi:10.5566/ias.4074

Authors

Ji Lin South China University of Technology
Toqeer Ahmed South China University of Technology
Li Feng Xi'an North Huian Chemical Industries Co., Ltd.
Pengfei Yin Xi'an North Huian Chemical Industries Co., Ltd.
Yiqi Liu South China University of Technology

DOI:

https://doi.org/10.5566/ias.4074

Keywords:

Feature engineering, Fiber analysis, Fiber intersection matching, Microscopic image processing

Abstract

In pulp and paper manufacturing, fiber morphology determined from micrographs is critical for process control. Automation is hindered by difficulties in separating fiber trajectories at skeleton intersections. Existing heuristic rules, typically based on directional continuity or width similarity, often fail on collapsed and entangled fiber networks in industrial pulps. This introduces systematic errors. These errors degrade downstream control actions, such as refining adjustments and furnish blending. This paper proposes a data-driven method that extracts four local geometric features at each junction breakpoint: direction angle, local width, cross-region relation, and curvature. These features are aggregated using a Kolmogorov–Arnold Network (KAN). On a synthetic dataset with ground truth, the proposed method achieves 0.978 in terms of F1. It substantially improves upon heuristic baselines. Ablation confirms that every feature is important. Applied to industrial pulp micrographs, our method produces fiber length and width distributions that match those of a commercial analyzer (FS5, Valmet). Without retraining, it generalizes to six independent pulp samples. These include both bamboo and softwood across multiple refining intensities. High measurement stability and full statistical discrimination of the six conditions (η² > 0.99) confirm its suitability. This method serves as a reliable, image-based sensing front-end for data-driven refining control and process decision-making.

References

Alejo AM, Beltrán CA (2024). A deep learning approach to distance map generation applied to automatic fiber diameter computation from digital micrographs. Sensors 24(17):5497.

Arafat Y, Cuesta-Apausa C, Castellano E, Reyes-Aldasoro CC (2025). Fibre tracing in biomedical images: an objective comparison between seven algorithms. PLoS ONE 20(4):e0320006.

Baez C, Ringania U, Bhamla S, Moon RJ (2026). Semi-automated image analysis of cellulose nanofibrils using machine learning segmentation and morphological thinning. Cellulose 33(6):3259–82.

Blum H (1967). A transformation for extracting new descriptors of shape. In: Wathen-Dunn W (Ed.), Models for the perception of speech and visual form. MIT Press, Cambridge, pp. 362–80.

Bradley D, Roth G (2007). Adaptive thresholding using the integral image. J Graph Tools 12(2):13–21.

Chen W, Liu M, Zhan Q, Tan Y, Meijering E, Radojevic M, Wang Y (2021). Spherical-patches extraction for deep-learning-based critical points detection in 3D neuron microscopy images. IEEE Trans Med Imaging 40(2):527–38.

Chirakitsakul T, Wattuya P, Somboon P, Jansakr P, Watcharopas C (2025). Integration of convolutional neural network and image processing for pulp fibril detection and measurement. IEEE Access 13: 74630–74642.

D'Amore A, Stella JA, Wagner WR, Sacks MS (2010). Characterization of the complete fiber network topology of planar fibrous tissues and scaffolds. Biomaterials 31(20):5345–54.

Deng Z, Yu L, Wang L, Ke W (2024). An algorithm for cross-fiber separation in yarn hairiness image processing. Vis Comput 40:3591–9.

Dias PAN, Rodrigues R, Reis MS (2024). Experimental assessment of paper formation conditions and structural two-sidedness and their impacts on curl phenomena. Processes 12(7):1536.

Gaiselmann G, Manke I, Lehnert W, Schmidt V (2013). Extraction of curved fibers from 3D data. Image Anal Stereol 32(1):57–63.

Guo C, Liu M, Guan T, Chen W, Wen H, Zeng T, Wang Y (2021). Crossover structure separation with application to neuron tracing in volumetric images. IEEE Trans Instrum Meas 70:1–13.

Howard CV, Reed MG (2005). Unbiased Stereology, 2nd ed. Abingdon: Garland Science/BIOS Scientific.

Huang X, Wen D, Zhao Y, Wang Q, Zhou W, Deng D (2016). Skeleton-based tracing of curved fibers from 3D X-ray microtomographic imaging. Results Phys 6:170–7.

Kayali Y, Gleadall A, Silberschmidt VV (2025). Geometric characterisation of stochastic fibrous networks: a comprehensive review. Fibers 13(3):27.

Lee TC, Kashyap RL, Chu CN (1994). Building skeleton models via 3-D medial surface axis thinning algorithms. CVGIP Graph Models Image Process 56(6):462–78.

Lin J, Feng L, Yin P, Liu Y (2026). Achieving synergistic morphology–topology analysis of microscale fiber networks in pulp and paper via a collaborative computational framework. In: Twelfth International Conference on Mechanical Engineering, Materials and Automation Technology (MMEAT 2026), Proc. SPIE, vol. 14355. SPIE.

Lindström SB, Amjad R, Gåhlin E, Andersson L, Kaarto M, Liubytska K, Persson J, Berg JE, Engberg BA, Nilsson F (2024). Pulp particle classification based on optical fiber analysis and machine learning techniques. Fibers 12(1):2.

Liu Z, Wang Y, Vaidya S, Ruehle F, Halverson J, Soljačić M, Hou TY, Tegmark M (2024). KAN: Kolmogorov-Arnold networks. In: International Conference on Learning Representations (ICLR 2025). OpenReview.

Lux J (2013). Automatic segmentation and structural characterization of low density fibreboards. Image Anal Stereol 32(1):13–25.

Muñoz A, Docaj A, Fernandez J, Carriero A (2025). FiberO for an automated quantitative analysis of fibers orientation and organization in biological fibrous tissues. Front Bioeng Biotechnol 12:1497837.

Otsu N (1979). A threshold selection method from gray-level histograms. IEEE Trans Syst Man Cybern 9(1):62–6.

Paunonen S, Mattsson A, Pettersson G, Ketoja JA (2026). Property variations of binder-free lignin-rich fiber networks driven by forming processes and hot pressing. ACS Omega 11(5):7129–41.

Pennells J, Heuberger B, Chaléat C, Martin DJ (2022). Assessing cellulose micro/nanofibre morphology using a high throughput fibre analysis device to predict nanopaper performance. Cellulose 29(4):2599–2616.

Pourdeyhimi B, Dent R (1999). Measuring fiber diameter distribution in nonwovens. Text Res J 69(4):233–6.

Qamar S, Baba AI, Verger S, Andersson M (2024). Segmentation and characterization of macerated fibers and vessels using deep learning. Plant Methods 20:126.

Ronneberger O, Fischer P, Brox T (2015). U-Net: convolutional networks for biomedical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, Lecture Notes in Computer Science, vol. 9351. Springer, Berlin, pp. 234–41.

Soille P (2003). Morphological image analysis: principles and applications, 2nd ed. Springer, Berlin.

Sten O, Del Dottore E, Pugno N, Mazzolai B (2024). A ridge-based detection algorithm with filament overlap identification for 2D mycelium network analysis. Ecol Inform 82:102670.

Togo R, Cardona S, Nagle I, Koenderink GH, Fereidoonnezhad B, Peirlinck M (2026). ToFiE, a topology-aware fiber extraction workflow for 3D reconstruction of dense and heterogeneous biological fiber networks from microscopy images. arXiv preprint arXiv:2604.18230.

Underwood EE (1970). Quantitative Stereology. Reading, MA: Addison-Wesley.

Verho T, Turpeinen T, Asad F, Immonen K (2025). A skeletonization-based approach for individual fiber separation in tomography images of biocomposites. Comput Mater Sci 246:113372.