Identification of Phoebe and Machilus Nees wood combining digital imaging wood combining digital imaging with near-infrared spectroscopy for biomass utilization

To explore the feasibility of accurately identifying wood species by fusing near-infrared spectroscopy with digital image features, and to address challenges in fusion models such as class imbalance, hard-sample learning, and label noise, this study proposes a novel multimodal framework termed VWC-Net. The framework integrates digital image features and near-infrared spectroscopy through a dedicated fusion architecture to enhance both discriminative accuracy and robustness. The proposed approach was applied to identify four wood species, Phoebe hui , Phoebe bournei , Machilus pauhoi , and Machilus leptophylla , which exhibit similar macro- and micro-scopic characteristics, making them challenging to classify using conventional methods. The results demonstrate that VWC-Net achieves state-of-the-art performance, with an accuracy of 93.93 %, precision of 93.12 %, recall of 93.66 %, and an F1-score of 93.38 %, significantly outperforming existing methods. Overall, this framework provides a reliable and practical solution for intelligent classification and quality assessment of wood resources, addressing critical challenges in sustainable forestry, industrial manufacturing, and international trade. • Near-infrared spectroscopy with macroscopic imaging enables non-destructive, species-level Lauraceae wood identification. • Wavelet-guided attention enhances anatomical and texture feature extraction under natural wood surface variability. • Multimodal fusion improves recognition of visually similar Phoebe and Machilus species versus single-feature methods. • The framework addresses class imbalance and labeling uncertainty common in large wood datasets. • Accurate discrimination (93.9 %) of high-value Nanmu species aids quality control and sustainable timber utilization.