Quick image analysis of concrete pore structure based on deep learning

Abstract

Abstract A fast image technique was proposed to characterise the pore size (>10\u202fμm) distribution of cement-based materials based on a deep learning model. Furthermore, a dataset of concrete pore structures was established for the analysis, and a number of photographs were required for a single section. The number of photographs required for a single section is calculated, and the calculation results meet the practicalities of image analysis of concrete pore structure. The superiority of the method is demonstrated by comparison with other deep learning models and traditional pore structure imaging methods. Experimental results indicate that the feasibility and efficiency of concrete pore structure analysis via deep leaning, which provides a facile and effective route for linking the microstructure to the macroscale performance of the concrete.