A Study into Methods for Classifying Cellulose Fibre Distributions in Paperboard
(2026) FHLM01 20261Solid Mechanics
Department of Construction Sciences
- Abstract
- This Master’s Thesis explored how traditional computer vision methods and deep learning methods could be used to estimate hardwood and softwood fibre distributions in paperboard. 11 mixtures ranging from 100% hardwood to 100% softwood with 10% increments were analysed. The study compared the usefulness of tomograms and radiographs for the given task.
It was found that traditional methods did not identify differentiating features or reliable trends in the tomograms. The tomogram dataset only consisted of two volumes per mixture which made it difficult to generalise the results. The radiograph dataset was larger and combined with pretrained models gave the best results. As the pretrained models found differentiating features in the... (More) - This Master’s Thesis explored how traditional computer vision methods and deep learning methods could be used to estimate hardwood and softwood fibre distributions in paperboard. 11 mixtures ranging from 100% hardwood to 100% softwood with 10% increments were analysed. The study compared the usefulness of tomograms and radiographs for the given task.
It was found that traditional methods did not identify differentiating features or reliable trends in the tomograms. The tomogram dataset only consisted of two volumes per mixture which made it difficult to generalise the results. The radiograph dataset was larger and combined with pretrained models gave the best results. As the pretrained models found differentiating features in the radiographs, the tomograms were deemed to be unnecessarily complex for the task.
Pretrained models were used to extract features from the radiographs, which were then classified into 11 classes using a custom trained MLP. Three ResNet architectures were compared and a hyperparameter search was performed on the MLP. The final model used the ResNet18 and achieved 80% accuracy and 0.80 RMSE on test data. Most incorrect classifications occurred between neighbouring classes. This is likely due to label uncertainty. Each radiograph only covers a small region of the original paperboard sheet and therefore does not accurately
represent the distribution of fibres of the larger sheet.
The generalisation of the model could not be verified because a directionality not visible to the human eye was detected in the data. Only 90 degree rotations were applied to the images meaning a directionality bias could still remain. It would be necessary to perform additional imaging to evaluate the model. (Less) - Popular Abstract
- Material science is a cornerstone in the development of products and packages worldwide. Understanding materials at the micro-scale helps create better, more durable and more sustainable materials. During the last decade, Artificial Intelligence (AI) has grown to become an asset in material science, helping discover patterns and material characteristics in images that are difficult to identify using traditional methods. It has also proven to be effective at reducing manual work and increasing productivity, particularly in industrial applications.
X-Ray Computed Tomography (XCT) is a useful tool within material research that generates tomograms (3D images), revealing the internal structures of materials at a small scale. The combination... (More) - Material science is a cornerstone in the development of products and packages worldwide. Understanding materials at the micro-scale helps create better, more durable and more sustainable materials. During the last decade, Artificial Intelligence (AI) has grown to become an asset in material science, helping discover patterns and material characteristics in images that are difficult to identify using traditional methods. It has also proven to be effective at reducing manual work and increasing productivity, particularly in industrial applications.
X-Ray Computed Tomography (XCT) is a useful tool within material research that generates tomograms (3D images), revealing the internal structures of materials at a small scale. The combination of AI and XCT has the potential to improve our understanding of materials and their structures.
One challenge with tomograms is their large size which can easily reach a few gigabytes each. Therefore, radiographs, which are single 2D images obtained from XCT scans, are investigated as a potentially more effective alternative.
In this work, hardwood and softwood fibres, which are several times thinner than a single human hair, are analysed. The study investigates whether the fibres can be differentiated in tomograms and in radiographs and how their distribution in paperboards can be determined using AI. At the moment, industry experts at Billerud manually make sure that the distribution of fibre types is correct in their produced paperboard batches. Traditional image analysis methods such as segmentation and thresholding are applied to the images. In addition, Deep Learning (DL) models are used to extract features from radiographs. Existing DL models called Residual Neural Networks (ResNets) and various extensions to their architectures are evaluated. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9243506
- author
- Edgren, Otto LU
- supervisor
-
- Eoin Walsh LU
- organization
- course
- FHLM01 20261
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- deep learning, pretrained models, ResNet, classification, principal component analysis (PCA), X-ray tomography, radiographs, paperboard, hardwood fibres, softwood fibres
- report number
- TFHF-5275
- language
- English
- id
- 9243506
- date added to LUP
- 2026-06-25 14:33:23
- date last changed
- 2026-06-25 14:33:23
@misc{9243506,
abstract = {{This Master’s Thesis explored how traditional computer vision methods and deep learning methods could be used to estimate hardwood and softwood fibre distributions in paperboard. 11 mixtures ranging from 100% hardwood to 100% softwood with 10% increments were analysed. The study compared the usefulness of tomograms and radiographs for the given task.
It was found that traditional methods did not identify differentiating features or reliable trends in the tomograms. The tomogram dataset only consisted of two volumes per mixture which made it difficult to generalise the results. The radiograph dataset was larger and combined with pretrained models gave the best results. As the pretrained models found differentiating features in the radiographs, the tomograms were deemed to be unnecessarily complex for the task.
Pretrained models were used to extract features from the radiographs, which were then classified into 11 classes using a custom trained MLP. Three ResNet architectures were compared and a hyperparameter search was performed on the MLP. The final model used the ResNet18 and achieved 80% accuracy and 0.80 RMSE on test data. Most incorrect classifications occurred between neighbouring classes. This is likely due to label uncertainty. Each radiograph only covers a small region of the original paperboard sheet and therefore does not accurately
represent the distribution of fibres of the larger sheet.
The generalisation of the model could not be verified because a directionality not visible to the human eye was detected in the data. Only 90 degree rotations were applied to the images meaning a directionality bias could still remain. It would be necessary to perform additional imaging to evaluate the model.}},
author = {{Edgren, Otto}},
language = {{eng}},
note = {{Student Paper}},
title = {{A Study into Methods for Classifying Cellulose Fibre Distributions in Paperboard}},
year = {{2026}},
}