Characterisation of pore structure of bulk wheat mixed with dockage using X-ray micro-computed tomography and deep learning

Characterisation of pore structure inside grain bulks is essential for predicting the airflow resistance during grain drying and aeration. Herein, the 3D pore network of wheat mixed with different percentages of canola and dockage was characterised. The characterised parameters include pore size, throat length, coordination number, airpath length, and tortuosity. To simulate the grain storage condition, the loaded wheat mixtures were cured using wax at 110 degrees C. The images of the waxed wheat mixtures were produced by using a high-resolution X-ray micro-computed tomography (mu CT) system at a resolution of 50 mu m per pixel. The computed space distribution was based on the 3D medial axis analysis of each image stack using Dragonfly 4.1 software. The pore space was segmented using a deep-learning model with 90.1-98.0% accuracy. A watershed-based image algorithm was used to generate the pore network from the segmented pore space. The pore structure parameters were extracted from the pore network of each wheat mixture. Most pores were classified as mesopores (>99%) in a range from 62.5 to 4000 mu m, with less than 1% classified as micropores (1-62.5 mu m). The mean pore size of clean bulk wheat was 648 +/- 403 mu m with a mean throat length of 1115 +/- 611 mu m. In the mixtures with 10% canola, 5 and 10% wheat dockage, the mean pore size was reduced by 10%, 29%, and 17%, respectively, from clean wheat. Adding canola and dockage generally influenced the connectivity in the pore microstructure with changes in bulk porosity, tortuosity, airpath number, and length.