Machine learning for non-destructive evaluation of wood

The integration of artificial intelligence with non-destructive evaluation techniques has advanced wood science by enabling automated, scalable, and accurate methods for defect detection and property prediction. This review adopts a modality-centred perspective, synthesising developments across sensing technologies such as RGB imaging, infrared thermography, near-infrared spectroscopy, acoustic wave analysis, X-ray computed tomography, and three-dimensional point clouds, and mapping them to corresponding data types, from scalar and time series signals to two-dimensional images and three-dimensional volumes. For each modality, we examine diagnostic capabilities, data characteristics, and applications, along with artificial intelligence methodologies ranging from conventional algorithms to deep learning architectures that leverage these inputs for classification, regression, and segmentation tasks. The review further organises methods by preprocessing workflows, classical machine learning approaches, and neural network architectures and discusses their reported performance and practical trade-offs in terms of accuracy, interpretability, computational cost, and deployment feasibility. Key challenges, including modality-specific data scarcity, generalisation across species and acquisition conditions, and integration of artificial intelligence into real-time industrial workflows, are discussed alongside emerging trends, such as multimodal learning, lightweight model design, and explainable artificial intelligence.