Machine Learning Techniques in Wood Recognition of Six Amazonian Fabaceae Species Based on NIR Spectra

The visual similarity of certain wood species poses significant challenges for accurate identification, which can lead to overexploitation of specific species. To address this issue and enable rapid analysis, nondestructive techniques combined with statistical and artificial intelligence methods have emerged as promising approaches to distinguish wood species and predict their quality. This study tested the hypothesis that wood from the Fabaceae family can be discriminated using near-infrared spectroscopy (NIR) combined with machine learning classifiers. To achieve this, NIR spectra were collected directly from wood surfaces, and the data were analyzed using R software to evaluate support-vector machines (SVM), partial least squares-discriminant analysis (PLS-DA), and k-nearest neighbor (k-NN) classifiers. The dataset was divided into training (70%) and testing (30%) subsets using random stratified sampling, and hyperparameters were optimized with k-fold cross-validation. PLS-DA achieved 100% recognition accuracy for Pseudopiptadenia suaveolens, Ormosia sp. Hymenolobium courbaril, and Enterolobium maximum. Tangential sections demonstrated 98.87% accuracy and an F1-score of 98.61% for PLS-DA, while NIR spectra from transversal sections provided better results for sSVM and k-NN classifications. Overall, the results demonstrate the potential of NIR spectroscopy for discriminating six Fabaceae species, with PLS-DA being the most effective approach.