Sitka Spruce Quality Estimation using Neural Networks

This paper describes an automated classifier for the identification of good wood and knotty wood from computer tomography (CT) images of logs. Such a system is intended to allow better assessment of saw logs before being cut into timber. We describe a new empirical model for the growth of Sitka Spruce (Picea Stichensis (Bong, Carr)) whose operation is adapted to Irish conditions. The use of Hopfield networks for 2D cross-section image reconstruction from CT data obtained from the model is investigated. We also used a multi-layer feedforward neural network trained with fast-backpropagation to identify good wood from knotty wood. The Hopfield approach to image reconstruction was seen as being unsuitable for application with the wood industry. However, the use of a feedforward neural network for wood classification produced very promising results when trained on our tree model. It is expected that results from real wood data would be even more accurate.