Controlled feasibility study for moisture-based grouping of dried wood: use of near infrared spectroscopy and support vector machine classification
摘要
Grouping kiln-dried lumber according to moisture content can help with quality control by preventing the mixing of stock that is under-dried or over-dried with stock that has the right range of moisture content. It has been established that near-infrared spectroscopy (NIRS) is a useful, non-destructive method of determining the moisture content of wood. Support Vector Machines (SVM) is a supervised learning technique for regression and classification problems. This study explores the use of NIR spectroscopy and SVM classification to group Populus deltoides (Indian hardwood) specimens based on moisture content after drying, with the aim of contributing to more informed decision-making in wooden product processing. Absorbance spectra (9000 cm−1 to 4500 cm−1) of the training set specimens were pre-processed using standard normal variate (SNV) transformation. The data set (517 numbers) was divided into two parts: the training set and the test set, in a ratio of 3:1. The training data set was used to create four classes of moisture content: acceptable for production, full-scale drying needed, extended level of drying needed, and moderate level of drying needed. These classes had moisture content ranges of 8.0% to 13.31%, 19.5% to 30.0%, 13.32% to 19.31%, and 30.1% and higher, respectively. The SVM classification model was developed using classification type NuSVC, kernel type radial basis function, gamma value 0.1, and Nu value 0.255. The cross-validation accuracy of the SVM classification training model was 82.77%, while its training accuracy was 91.64%. The test specimens with unknown moisture content were classified with an average success rate of 92.5% using the SVM classification model. The study, based on a limited controlled dataset, warrants future validation on larger, more diverse industrial samples to assess its robustness and scalability.