Assessing convolutional neural networks for wood log re-identification

Modern timber supply chains are undergoing a digital transformation to combat illegal logging and enhance environmental sustainability. Wood re-identification is an essential tool for promoting transparent traceability from forest to the final product and, with it, verifying the legal origin of timber, thereby improving accountability throughout the timber supply chain. Recent advances in artificial intelligence (AI), particularly Convolutional Neural Networks (CNN) and computer vision techniques, have shown promising performance in automating wood re-identification and matching tasks, improving efficiency and accuracy. However, previous studies on CNN-based wood log re-identification provide limited understanding of the underlying reasoning behind model predictions, leaving a gap in the interpretability of AI methods that support reliable wood tracing across the timber supply chain. To address this gap, the present study evaluates a CNN-based framework for closed-set wood log re-identification in a laboratory-scale simulation of the forest-to-sawmill supply chain. Additionally, Explainable AI (XAI) techniques are applied to analyse the CNN’s decision-making process and provide insights into the features driving model predictions. The experimental framework is based on a dataset of 49 larch logs, complemented by the open-source dataset for model development and validation. The proposed models achieved 97–98% accuracy on validation and test sets, while XAI interpretation techniques revealed key textural and structural image features that contribute to wood log re-identification. By integrating high predictive performance with model interpretability, this study contributes to the development of more transparent and reliable AI-based wood log re-identification supporting practical deployment in wood traceability systems.