The goal of this project is to develop the sawmill industry significantly via digitalisation. One of the main objectives is to build an information system connecting all the production steps of the mill so that it is possible to track the life cycle of raw material from the beginning to the end product. Raw material and its characteristics become individually traceable (e.g., logs and their parts) in the real-time industrial environment. This makes the use of raw material more efficient. At the same time, resource-efficient processes become more sustainable which benefits conservation of nature, especially in controlling the climate change.
Modern sensor systems, new computational methods, and more robust mathematical models enable developing novel machine vision based measurements for quality control systems. The capabilities of the systems can be extended by using measured information from the sawmill processes and raw material. The obtained quality information can be connected via feedback loops to improve log sorting and to optimize the sawing processes. The developed innovations will be tested in the real industrial environment.
The project includes international collaboration with the Computer Graphics Research Group at Brno University of Technology (BUT) and Uncertainty Quantification Group at Massachusetts Institute of Technology.
People
Fedor Zolotarev
 |
Office: 2419 | Doctoral student | |
| Tuomas Eerola
|
Office: 2425 | Tel: +358 50 404 2054 | Associate Professor, Project Manager |
| Heikki Kälviäinen
|
Office: 2415 | Tel: +358 40 586 7552 | Professor, Project Leader |
| Lasse Lensu
|
Office: 2413 | Tel: +358 40 759 1720 | Professor |
| Heikki Haario
|
Office: 2409 | Tel: +358 400 814 092 | Professor |
Publications
Virtual sawing using generative adversarial networks
The paper studies the utilization of generative adversarial networks (GANs) in virtual sawing (predicting the outcome of sawing process) and generating photorealistic board images based on log measurements.
Modelling internal knot distribution using external log features
The paper proposes a method to model internal knot locations based on laser scans of logs allowing virtual sawing (estimating knot locations in the resulting boards) for sawmill process control and optimization purposes.
Timber Tracing with Multimodal Encoder-Decoder Networks
The paper proposes a method to match RGB images of boards to surface scans of logs using multimodal encoder-decoder networks. The method makes it possible to identify from which log the board was sawn from and enables tracing of material flows in sawmill environment.
Fine-Grained Wood Species Identification Using Convolutional Neural Networks
The paper proposes a framework to efficiently identify the wood species from board images using convolutional neural networks (CNN) and compares various CNN architectures for the task.
Detection of Mechanical Damages in Sawn Timber Using Convolutional Neural Networks
The paper proposes a novel and efficient method for detection defects in sawn timber using convolutional neural networks.
Detection of mechanical damages in sawn timber using convolutional neural networks
The thesis presents and evaluates a convolutional neural networks based method to detect mechanical damages in images of wooden boards.
Download: PDF
Species identification of wooden material using convolutional neural networks
The thesis compares various concolutional neural network architectures in the task of species identification from images of wooden boards.
Download: PDF
Knot detection from laser point clouds for multimodal matching of wooden material
The thesis presents a method for knot detection from laser point clouds of log surfaces.
Download: PDF