Advances in Timber Identification Using Deep Learning: A Review of Convolutional Neural Network Models

Abstract

Machine Vision (MV) software has emerged as a powerful tool for timber species identification, offering significant advantages including species-level accuracy, cost-effectiveness, and the elimination of human errors. The development of MV software relies on three major supporting technologies: computer vision, machine learning (ML), and deep learning (DL). This paper provides an in-depth exploration of the role of DL, with a particular focus on convolutional neural networks (CNNs), in enhancing MV software for timber identification. The potential of CNN architectures is examined in detail, including a review of commonly used CNN models and their effectiveness in identifying different timber species. This paper also discusses current practices in developing CNN models for integration into MV software for timber identification tasks, highlighting the standards and procedures researchers should follow to ensure optimal performance and reliability. Additionally, the challenges associated with implementing CNN models for timber identification are addressed. They include limited model usability due to the diversity of timber species and geographical variation. Variability in methodologies, imaging devices, and data processing approaches across studies further complicates the comparison and integration of results. This paper emphasises the need for standardised practices and further empirical research to address these inconsistencies and improve CNN-based MV systems for timber identification.