Abstract
Conventional wood quality assessments often involve invasive methods that increase costs and weaken material integrity. Recent studies using machine learning, particularly convolutional neural networks (CNNs) and ensemble learning, have shown high accuracy in predicting wood density and mechanical properties from images. However, real-time evaluation of moisture content (MC) across different species remains underexplored. This research introduces a non-destructive approach that integrates 2D image analysis with an advanced ensemble learning model to predict both wood density and MC, using high-resolution images from Line Scanner and Camera Systems to extract key color features and wood attributes. A novel real-time MC determination method employing a gravimetric formula based on actual sample dimensions and weights further enhances immediate wood evaluation. The model achieves reliable predictions with a Coefficient of Determination (R2) of 0.94 for oven-dry density and 0.85 for MC, and Root Mean Square Errors (RMSE) of 0.01 g/cm3 and 8.33 %, respectively, offering a precise, viable tool for real-time wood quality and moisture assessment, and advancing sustainable industry practices.