| Abstrakt: |
The discrepancies between conventional and fast-growing teak woods are challenging due to the similarity of their characteristics, especially in their appearance and anatomical properties. In general, the wood quality of those teak species is different, so identifying of those wood is urgently required. Because the conventional method of wood identification is labor intensive and long process, this study was conducted to classify those woods using Fourier Transform Near Infrared (FTNIR) in combination with the Random Forest Classifiers (RF). The conventional teak wood from Perhutani (a state forest enterprise) and several fast-growing teak wood varieties (Platinum, JUN, and community) were scanned at the Near Infrared (NIR) spectra of 10,000-4,000 cm−1 and then analyzed by RF. In addition, the feature importance was investigated in order to determine the importance bands for discrimination. The results showed that the best accuracy for distinguishing slow-and fast-grown teak reached 98.2% at the optimal estimator parameters of 11. The RF feature importance correlation showed that the band 6,000 cm−1 assigned as lignin was the main factor affecting the classification. Moreover, the RF analysis revealed that this method could also separate the conventional, Platinum, JUN, and community teak woods with a high accuracy of 98% at n-estimators 31. The third range was the main factor affecting the model, which contains CH vibrations from the aromatic framework. However, some parts of the fourth range, for instance, bands at 4,890-4,620 cm−1, which correspond to the cellulose regions of the NIR wavenumber, have also affected the determination. [ABSTRACT FROM AUTHOR] |
