A universal transfer-learning-based detection model for characterizing vascular bundles in Phyllostachys

A comprehensive understanding of vascular bundles is the key to elucidate the excellent intrinsic mechanical properties of bamboo. This research aims to investigate the gradient distribution of fiber volume fraction and the gradient changes in the size of vascular bundles along the radial axis in Phyllostachys. The inter-nodes of twenty-nine kinds of Phyllostachys were collected, which the cross section was sanded by sanding pads with 320 mesh and scanned with a resolution of 9600 ppi. A universal transfer-learning-based vascular bundle detection model with high precision of up to 96.97% were built, which can help to obtain the characteristics of vascular bundles quickly and accurately. The total number of vascular bundles, total fiber sheath area, the length, width and area of fiber sheath of individual vascular bundles within the entire cross-section were counted and analyzed. The results showed that these parameters had a strongly positive linear correlation with the outer circumference and wall thickness of bamboo culms, but the fiber volume fraction (25.50 ± 3.51%) and the length-to-width ratio of the vascular bundles (1.226 ± 0.091) were relatively constant. Furthermore, the cross section of bamboo were divided into multi-layer sheet along the wall thickness direction and the characteristics of vascular bundle were counted in each layer. The results showed that the fiber volume fraction decreased exponentially along the radial direction from skin to core, the length-to-width ratio of vascular bundle decreased quadratically along the radial direction from skin to core, the width of vascular bundle increased linearly along the radial direction from skin to core. The trends of the gradient change in vascular bundle’s characteristics were found highly consistent among bamboo species in Phyllostachys.