THz-based Density Prediction in Timber Wood using Stacked Ensemble Regression Strategy

Density is a crucial parameter that greatly influences the mechanical properties, structural grading, and thereby, overall utilization of timber wood across diverse industrial domains. Hence, the accurate prediction of wood density is very important, which in turn enables the evaluation of defects and the estimation of moisture content in timber wood species. Although traditional density estimation techniques, such as gravimetry, X-ray densitometry, and manual inspection, are effective to some extent, they are often limited by their destructive nature, high cost, subjectivity, sensitivity, or safety concerns. To circumvent the limitations of existing techniques, terahertz (THz) technology has been explored here as a nondestructive, non-ionizing alternative for assessing the quality of timber wood. THz radiation can capture the difference in material properties arising from variations in internal structure, which in turn are correlated with wood density. In this work, a publicly available THz open dataset has been utilized, which contains the THz temporal signals of 78 samples from six diverse wood species. Here, a uniquely tailored combination of pre-processing steps, feature extraction techniques, and a stacked regression ensemble has been adopted to predict the wood density precisely. The stacked ensemble regressor sequentially integrates different ensemble learners strategically, with random forest, XGBoost, CatBoost, and the ridge regressor. This proposed approach yields a very high R2score of 0.9996, highlighting the advantages of integrating THz technology with machine learning techniques for precise, non-destructive estimation of wood density in the timber industry.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.