Advanced

Statistical Methods for Classification of Wooden Boards

Lindell, Johan (2014) FMS820 20142
Mathematical Statistics
Abstract (Swedish)
The quality inspection of wooden boards is experiencing a large change. By the use of
camera and laser technology board characteristics and defects can be instantly
identified and measured. This thesis investigates how the information from a quality
inspection system can be used to classify boards into different quality classes, by the
use of statistical classification models. Two types of classification models have been
tested, Logistic Regression and Support Vector Machines. To deal with potential
overfitting a regularized version of Logistic Regression is implemented, and to deal
with ordinal dependent variables a logistic regression model for ordinal variables has
been implemented. The classification models have been tested... (More)
The quality inspection of wooden boards is experiencing a large change. By the use of
camera and laser technology board characteristics and defects can be instantly
identified and measured. This thesis investigates how the information from a quality
inspection system can be used to classify boards into different quality classes, by the
use of statistical classification models. Two types of classification models have been
tested, Logistic Regression and Support Vector Machines. To deal with potential
overfitting a regularized version of Logistic Regression is implemented, and to deal
with ordinal dependent variables a logistic regression model for ordinal variables has
been implemented. The classification models have been tested against board strength
classes, and similar results have been obtained by most models. It is concluded that
the regularized logistic regression is the model that manages to classify most boards
correctly, but the Support Vector Machine produces a better result on classes where
training data is scarce.
The thesis was done on behalf of RemaSawco AB, a company that manufactures
measurement systems and inspection systems for the sawmill industry. (Less)
Please use this url to cite or link to this publication:
author
Lindell, Johan
supervisor
organization
course
FMS820 20142
year
type
H2 - Master's Degree (Two Years)
subject
language
English
id
4858231
date added to LUP
2014-12-09 08:03:05
date last changed
2014-12-09 08:03:05
@misc{4858231,
  abstract     = {The quality inspection of wooden boards is experiencing a large change. By the use of
camera and laser technology board characteristics and defects can be instantly
identified and measured. This thesis investigates how the information from a quality
inspection system can be used to classify boards into different quality classes, by the
use of statistical classification models. Two types of classification models have been
tested, Logistic Regression and Support Vector Machines. To deal with potential
overfitting a regularized version of Logistic Regression is implemented, and to deal
with ordinal dependent variables a logistic regression model for ordinal variables has
been implemented. The classification models have been tested against board strength
classes, and similar results have been obtained by most models. It is concluded that
the regularized logistic regression is the model that manages to classify most boards
correctly, but the Support Vector Machine produces a better result on classes where
training data is scarce.
The thesis was done on behalf of RemaSawco AB, a company that manufactures
measurement systems and inspection systems for the sawmill industry.},
  author       = {Lindell, Johan},
  language     = {eng},
  note         = {Student Paper},
  title        = {Statistical Methods for Classification of Wooden Boards},
  year         = {2014},
}