Lumber defect scanning including multi-dimensional pattern recognition

• US 5,703,960 A
• Filed: 09/16/1996
• Issued: 12/30/1997
• Est. Priority Date: 08/24/1994
• Status: Expired due to Term

• Current Assignee: US Natural Resources Incorporated (U.S. Natural Resources)
• Original Assignee: US Natural Resources Incorporated (U.S. Natural Resources)
• Inventors: Soest, Jon F.
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): SHALWALA, BIPIN H
• Application Number:
  US08/714,632
• Time in Patent Office: 470 Days
• Field of Search: 382/100, 382/108, 382/110, 382/141, 382/143, 382/149, 382/181, 382/209, 382/224, 250/559.18, 250/559.45, 250/559.44, 356/430, 356/429, 356/431, 356/445, 356/446, 348/91, 348/92, 348/125, 348/128, 348/129, 348/130
• US Class Current: 382/141
• CPC Class Codes: G01N 21/8986 Wood
• CPC Class Codes: G01N 33/46 Wood

Abstract

Grain defect scanning takes into account a broad set of data representing both wood grain structure and wood grain image to provide a multi-dimensional scan vector for an inspection point with wide variation therein relative to defect types. A library of similarly structured multi-dimensional training set vectors developed during a preliminary training session with known defect types is referenced by multivariate pattern recognition analysis to classify a collection of scan vectors associated with an article under inspection. By statistically matching scan vectors with training set vectors under pattern recognition analysis, physical locations on a wood article are identified according to known defect types.

Claims

1. A method for identifying a defect relative to a given wood article, said method comprising the steps:

collecting a first set of sensor data values taken from sensors directed toward a portion of a sample wood article surface, said portion of said sample wood article surface corresponding to a known defect, said set of sensor data values including a first specular reflection value and a first diffuse reflection value each representing wood grain structure at said portion of said wood article;

selecting at least one mathematic function applied to said set of sensor data values and providing a given distance therebetween;

collecting a second set of sensor data values taken from sensors directed toward a portion of said given wood article, said second set of sensor data values corresponding to said first set of sensor data values and including a second specular reflection value and a second diffuse reflection value;

applying said mathematic function to said second set of sensor data values; and

if said at least one mathematic function indicates distance between said second set of sensor data values at or below said given distance then identifying said portion of said given wood article as corresponding to said known defect.

2. A method according to claim 1 wherein said first set of sensor data values include a calculated ratio of said first specular reflection value and said first diffuse reflection value and said second set of sensor data values include a calculated ratio of said second specular reflection value and said second diffuse reflection value.

3. A method according to claim 2 wherein said first set of sensor data values include a first measure of height and said second set of sensor data values include a second measure of height.

4. A method according to claim 3 wherein said first set of sensor data values include a first measure of tracheid effect and said second set of specular data values include a second measure of tracheid effect.

5. A method according to claim 1 wherein said method comprises selecting said mathematic function according to multivariate pattern recognition analysis.

6. A method according to claim 4 wherein said first sensor data values include first image value and said second set of sensor data values includes a second image value.

7. A method according to claim 6 wherein said first and second image values correspond to light intensity in a given color band.

8. A method for detecting presence of at least one of a set of wood cell structure conditions at a wood article surface, said method comprising the steps:

executing a training session wherein for each member of said set of wood cell structure conditions at least one training vector is defined, each training vector including a designation of the associated wood cell structure condition and a set of training vector sensor data values taken from sensors directed toward an example of the associated wood structure condition, said training vector sensor data values including a first specular reflection value and a first diffuse reflection value each representing wood cell structure at said example of the associated wood structure condition;

executing a scanning procedure against inspection points of said given wood article wherein a scan vector is collected for each inspection point, each scan vector including location data indicating a physical location for the associated inspection point and including a set of scan vector sensor data values taken from sensors directed toward the associated inspection point and corresponding to said training vector sensor data values and including a second specular reflection value and a second diffuse reflection value;

matching by pattern recognition analysis each scan vector with each training vector; and

detecting presence of a given wood structure condition at a given location when a training set vector corresponding to said given wood structure condition matches under said pattern recognition analysis a scan vector for said given location.

9. A method according to claim 8 wherein each training vector includes a calculated ratio of said first specular reflection value and said first diffuse reflection value and each scan vector includes a calculated ratio of said second specular reflection value and said second diffuse reflection value.

10. A method according to claim 9 wherein each training vector includes a first measure of height and each scan vector includes a second measure of height.

11. A method according to claim 10 wherein each training vector includes a first measure of tracheid effect and each scan vector includes a second measure of tracheid effect.

12. A method according to claim 11 wherein each training vector includes a first image value and each scan vector includes a second image value.

13. A method according to claim 12 wherein said first and second image values represent a magnitude of light intensity in a first color band.

14. A method according to claim 13 wherein said first and second image values represent a magnitude of light intensity in a second color band.

15. A method according to claim 8 wherein said pattern recognition is by multivariate pattern recognition.

16. A method of identifying the condition of inspection of inspection points of a wood article surface, the method comprising the steps:

scanning a plurality of inspection points of said wood article surface, said scanning step collecting for each inspection point both reflective grain defect data and tracheid effect data;

identifying a first class of surface conditions by reference to said grain defect scanning data; and

identifying a second class of surface conditions by reference to said tracheid effect data while excluding from consideration said first class of surface conditions said first class of surface conditions being more reliably detected relative to said second class of surface condition.

17. A method according to claim 16 wherein said first class of surface conditions corresponds to knot defects.

18. A method according to claim 16 wherein said second class of surface conditions to broad-area defects.

19. A method according to claim 18 wherein said broad area defects includes at least one of wane, wood decay, stains, and compression wood.