Improved knot detection for CT using AI
Industrial CT scanning of logs is becoming a more widespread tool in large-scale sawmill production and has proven to considerable increase the value yield. Still, there is great potential in improving the accuracy even more when measuring important properties such as knot characteristics.
Knots are the decisive and dominant feature describing the quality of wood and about 80% of downgrading in a sawmill process is related to knots. In the sawmill process high precision in knot detection and characterization is needed to be able to optimize the sawing of each log.
The main purpose of this CT WOOD project is to improve the accuracy when measuring knot properties that so far have been difficult to measure in computer tomography (CT) images of green logs. The hypothesis is that this can be achieved by combining detailed laboratory data with new AI-based methods for image analysis of CT-data. The project is the first that systematically compare CT measurements of stem structure/knot structure in green condition with the stem structure in dry condition in order to make better algorithms for knot extraction from CT images. Scots pine trees will be harvested from different regions and growth conditions in Sweden to cover high variability in knot-shape development in the data. The database will contain 24 trees, about 120 logs and 2100 knots that will be CT-scanned in green and dry condition.
Project information
Subject: Knot detection and characterization in green logs
Time span project: August 2021 – December 2023
Contact
Olof Broman
- Senior Lecturer
- 0910-585325
- olof.broman@ltu.se
- Olof Broman
Updated: