Radial distribution of vascular bundle morphology in Chinese bamboos: machine learning methodology for rapid sampling and classification

Abstract Variation in anatomical features of the culm wall namely the shape and size distributions of vascular bundles between different genera and species of bamboo is not well understood due to the cumbersome task of manual measurements. Using machine learning methodology, this work presents a universal vascular bundle detection model for rapid, reliable, and automatic characterization of vascular bundles in culm cross sections of 213 species across 23 genera of Chinese bamboos. The number of vascular bundles and the fiber sheath area have positive linear correlations with the outer circumference and the wall thickness, respectively. The distribution density of vascular bundles has a decay exponential correlation with the outer circumference and the wall thickness. The average fiber volume fraction was 35.2 % ± 7 % with relatively small variation between species. Bamboo species could be grouped into three categories based the endodermis to epidermis distribution pattern of radial and tangential length of vascular bundles, two categories of radial-to-tangential ratio and four categories of fiber sheath area distribution pattern. Implications on bamboo classification, structural and pulp/paper applications were discussed. The findings from this study provide groundwork for the establishment of a unified, authoritative and objective bamboo classification system based on the vascular tissue morphology.