Treffer: Identification and reconstruction of concrete mesostructure based on deep learning in artificial intelligence.

• The accuracy of concrete image recognition is improved by improving the deep learning framework, providing the basis for 3D reconstruction as well as deep generative network applications. • The mesostructures of concrete similar to CT scanning result are established. • Meso models conforming to the characteristics of real aggregates are generated. The morphology and distribution of coarse aggregate are important factors that affect the concrete properties. This paper establishes two-dimensional and three-dimensional models of real coarse aggregate in concrete based on the computer vision method of in-depth learning and makes parameterized analysis. To improve the accuracy of identifying coarse aggregates, the data set construction method in the convolution neural network is improved, and the attention mechanism module is added. In the parameterized analysis, the random natural and recycled coarse aggregate models are generated using the deep generation network technology. The modeling method in this paper significantly improves the accuracy and efficiency of concrete meso -modeling, which provides a new idea for concrete meso -modeling. [ABSTRACT FROM AUTHOR]