Detection of knots in logs using x-ray imaging

Dissertation

Markku Pietikäinen

Research output: ThesisDissertationMonograph

Abstract

The economy of the sawing process would be greatly improved, if the internal properties of logs were known beforehand. The output quality would be more predictable, resulting in a higher yield and better utilisation of timber. Our fundamental idea was to apply the principles of computed tomography (CT) to knot detection in logs. CT is a standard method in medical applications for internal diagnosis of the human body. Unfortunately, the high-speed sawing process leaves a very limited time for log imaging. Rotation or multiple passes cannot be used to obtain hundreds of projections of a log; thus a detailed reconstruction in the sense of CT is not possible. However, we found that even from three fixed projections valuable information can be acquired. This was demonstrated by analysing images of both simulated and real logs. An x-ray imaging system was constructed to measure full-sized logs moving at normal sawing speeds. At the first stage, only one source-detector pair was available; thus three passes per log were needed in the tests. A new method was developed for computing 3-D properties of knot clusters. We call it the sector oriented reconstruction technique, or SORT. The name refers to the principle of applying a cylindrical co-ordinate system with discrete sectors, rings, and slices. The object space is composed of volume elements with dimensions far larger than the imaging pixel size. The densities of the volume elements are estimated to recognise potential knot locations and sizes. The method uses a priori knowledge of typical shapes and densities of knots and stems, along with evidential reasoning when looking for candidate knot directions. The method produces estimates of knot characteristics at two levels: (1) volumes and co-ordinates of knot clusters, and (2) thicknesses, lengths, volumes, and co-ordinates of individual knots. In some cases, the information from three projections is not enough to separate out individual knots. A confidence index is therefore calculated to indicate the reliability of the results. The performance of the detection algorithms was tested with data from simulated and real logs. For real logs the relative volumes of detected, undetected, and ghost knots were 0.88 : 0.12 : 0.15, and for simulated logs 0.96 : 0.04 : 0.02.
Original languageEnglish
QualificationDoctor Degree
Awarding Institution
  • University of Oulu
Supervisors/Advisors
  • Pietikäinen, Matti, Supervisor, External person
Award date8 Mar 1996
Place of PublicationEspoo
Publisher
Print ISBNs951-38-4924-4
Publication statusPublished - 1996
MoE publication typeG4 Doctoral dissertation (monograph)

Fingerprint

X-Rays
Tomography
Human Body
Reproducibility of Results
Names

Keywords

  • knots
  • detection
  • logs
  • structural timber
  • x-rays
  • x-ray inspection
  • quality
  • quality control
  • properties
  • computers
  • tomography

Cite this

Pietikäinen, M. (1996). Detection of knots in logs using x-ray imaging: Dissertation. Espoo: VTT Technical Research Centre of Finland.
Pietikäinen, Markku. / Detection of knots in logs using x-ray imaging : Dissertation. Espoo : VTT Technical Research Centre of Finland, 1996. 90 p.
@phdthesis{09e2a3d9b670400fae35ee00e67c11eb,
title = "Detection of knots in logs using x-ray imaging: Dissertation",
abstract = "The economy of the sawing process would be greatly improved, if the internal properties of logs were known beforehand. The output quality would be more predictable, resulting in a higher yield and better utilisation of timber. Our fundamental idea was to apply the principles of computed tomography (CT) to knot detection in logs. CT is a standard method in medical applications for internal diagnosis of the human body. Unfortunately, the high-speed sawing process leaves a very limited time for log imaging. Rotation or multiple passes cannot be used to obtain hundreds of projections of a log; thus a detailed reconstruction in the sense of CT is not possible. However, we found that even from three fixed projections valuable information can be acquired. This was demonstrated by analysing images of both simulated and real logs. An x-ray imaging system was constructed to measure full-sized logs moving at normal sawing speeds. At the first stage, only one source-detector pair was available; thus three passes per log were needed in the tests. A new method was developed for computing 3-D properties of knot clusters. We call it the sector oriented reconstruction technique, or SORT. The name refers to the principle of applying a cylindrical co-ordinate system with discrete sectors, rings, and slices. The object space is composed of volume elements with dimensions far larger than the imaging pixel size. The densities of the volume elements are estimated to recognise potential knot locations and sizes. The method uses a priori knowledge of typical shapes and densities of knots and stems, along with evidential reasoning when looking for candidate knot directions. The method produces estimates of knot characteristics at two levels: (1) volumes and co-ordinates of knot clusters, and (2) thicknesses, lengths, volumes, and co-ordinates of individual knots. In some cases, the information from three projections is not enough to separate out individual knots. A confidence index is therefore calculated to indicate the reliability of the results. The performance of the detection algorithms was tested with data from simulated and real logs. For real logs the relative volumes of detected, undetected, and ghost knots were 0.88 : 0.12 : 0.15, and for simulated logs 0.96 : 0.04 : 0.02.",
keywords = "knots, detection, logs, structural timber, x-rays, x-ray inspection, quality, quality control, properties, computers, tomography",
author = "Markku Pietik{\"a}inen",
note = "Project code: ELET9411",
year = "1996",
language = "English",
isbn = "951-38-4924-4",
series = "VTT Publications",
publisher = "VTT Technical Research Centre of Finland",
number = "266",
address = "Finland",
school = "University of Oulu",

}

Pietikäinen, M 1996, 'Detection of knots in logs using x-ray imaging: Dissertation', Doctor Degree, University of Oulu, Espoo.

Detection of knots in logs using x-ray imaging : Dissertation. / Pietikäinen, Markku.

Espoo : VTT Technical Research Centre of Finland, 1996. 90 p.

Research output: ThesisDissertationMonograph

TY - THES

T1 - Detection of knots in logs using x-ray imaging

T2 - Dissertation

AU - Pietikäinen, Markku

N1 - Project code: ELET9411

PY - 1996

Y1 - 1996

N2 - The economy of the sawing process would be greatly improved, if the internal properties of logs were known beforehand. The output quality would be more predictable, resulting in a higher yield and better utilisation of timber. Our fundamental idea was to apply the principles of computed tomography (CT) to knot detection in logs. CT is a standard method in medical applications for internal diagnosis of the human body. Unfortunately, the high-speed sawing process leaves a very limited time for log imaging. Rotation or multiple passes cannot be used to obtain hundreds of projections of a log; thus a detailed reconstruction in the sense of CT is not possible. However, we found that even from three fixed projections valuable information can be acquired. This was demonstrated by analysing images of both simulated and real logs. An x-ray imaging system was constructed to measure full-sized logs moving at normal sawing speeds. At the first stage, only one source-detector pair was available; thus three passes per log were needed in the tests. A new method was developed for computing 3-D properties of knot clusters. We call it the sector oriented reconstruction technique, or SORT. The name refers to the principle of applying a cylindrical co-ordinate system with discrete sectors, rings, and slices. The object space is composed of volume elements with dimensions far larger than the imaging pixel size. The densities of the volume elements are estimated to recognise potential knot locations and sizes. The method uses a priori knowledge of typical shapes and densities of knots and stems, along with evidential reasoning when looking for candidate knot directions. The method produces estimates of knot characteristics at two levels: (1) volumes and co-ordinates of knot clusters, and (2) thicknesses, lengths, volumes, and co-ordinates of individual knots. In some cases, the information from three projections is not enough to separate out individual knots. A confidence index is therefore calculated to indicate the reliability of the results. The performance of the detection algorithms was tested with data from simulated and real logs. For real logs the relative volumes of detected, undetected, and ghost knots were 0.88 : 0.12 : 0.15, and for simulated logs 0.96 : 0.04 : 0.02.

AB - The economy of the sawing process would be greatly improved, if the internal properties of logs were known beforehand. The output quality would be more predictable, resulting in a higher yield and better utilisation of timber. Our fundamental idea was to apply the principles of computed tomography (CT) to knot detection in logs. CT is a standard method in medical applications for internal diagnosis of the human body. Unfortunately, the high-speed sawing process leaves a very limited time for log imaging. Rotation or multiple passes cannot be used to obtain hundreds of projections of a log; thus a detailed reconstruction in the sense of CT is not possible. However, we found that even from three fixed projections valuable information can be acquired. This was demonstrated by analysing images of both simulated and real logs. An x-ray imaging system was constructed to measure full-sized logs moving at normal sawing speeds. At the first stage, only one source-detector pair was available; thus three passes per log were needed in the tests. A new method was developed for computing 3-D properties of knot clusters. We call it the sector oriented reconstruction technique, or SORT. The name refers to the principle of applying a cylindrical co-ordinate system with discrete sectors, rings, and slices. The object space is composed of volume elements with dimensions far larger than the imaging pixel size. The densities of the volume elements are estimated to recognise potential knot locations and sizes. The method uses a priori knowledge of typical shapes and densities of knots and stems, along with evidential reasoning when looking for candidate knot directions. The method produces estimates of knot characteristics at two levels: (1) volumes and co-ordinates of knot clusters, and (2) thicknesses, lengths, volumes, and co-ordinates of individual knots. In some cases, the information from three projections is not enough to separate out individual knots. A confidence index is therefore calculated to indicate the reliability of the results. The performance of the detection algorithms was tested with data from simulated and real logs. For real logs the relative volumes of detected, undetected, and ghost knots were 0.88 : 0.12 : 0.15, and for simulated logs 0.96 : 0.04 : 0.02.

KW - knots

KW - detection

KW - logs

KW - structural timber

KW - x-rays

KW - x-ray inspection

KW - quality

KW - quality control

KW - properties

KW - computers

KW - tomography

M3 - Dissertation

SN - 951-38-4924-4

T3 - VTT Publications

PB - VTT Technical Research Centre of Finland

CY - Espoo

ER -

Pietikäinen M. Detection of knots in logs using x-ray imaging: Dissertation. Espoo: VTT Technical Research Centre of Finland, 1996. 90 p.