Quantifying knots by image analysis and modeling their effects on the mechanical properties of loblolly pine lumber

Automated grading machines that quantify knots are increasingly deployed by lumber mills, however their use in mill studies that assess lumber quality have been limited. The objective here was to develop a method to evaluate the knots of loblolly pine lumber using image analysis and to develop models to predict modulus of elasticity (MOE) and modulus of rupture (MOR) from 171 pieces of dimension lumber. Lumber was photographed on the wide faces and individual knots were identified using the k-means clustering algorithm. The percentage of wood made up of knots on the wide faces (Knot%) was calculated by summing the individual knot areas over the total surface area, as well as on a sub-section of the lumber span which was optimized separately for MOE (Knot%(MOE)) and MOR (Knot%(MOR)). Models were built using the knot measurements and compared to models built using specific gravity (SG) and acoustic velocity squared (AV(2)). Knot% explained 30% of the variation in MOE and 39% of the variation in MOR. Incorporating Knot%(MOE) into a model with SG and AV(2) did not appreciably improve model performance (R-2 = 0.75, RMSE = 1.1 GPa) over the base SG and AV(2) model (R-2 = 0.74, RMSE = 1.2 GPa). Incorporating Knot%(MOR) into a model with SG and AV(2) significantly improved the prediction (R-2 = 0.65, RMSE = 7.2 MPa) compared to the base SG and AV(2) model (R-2 = 0.56, RMSE = 8.0 MPa). This study demonstrates the feasibility of using image analysis to assess knot information in lumber to improve predictions of mechanical properties.