DVC-StNet: an efficient digital volume correlation method based on deep learning

Digital Volume Correlation (DVC) is a non-destructive testing technique that analyzes three-dimensional volumetric images before and after deformation to obtain the displacement field, and it has been widely applied in various engineering fields. However, the computational efficiency and accuracy of this method still need improvement. In recent years, deep learning-based methods have introduced new advancements to DVC technology. This study extends the two-dimensional digital image correlation network, StrainNet, to a three-dimensional network structure, proposing DVC-StNet. Unlike traditional correlation-based methods, DVC-StNet eliminates the need for complex numerical iterative calculations, significantly reducing computational complexity. Compared with conventional methods, DVC-StNet achieves a 1000-fold increase in computational efficiency. Additionally, trained on a simulated three-dimensional volumetric dataset, DVC-StNet achieves an average end-point error of less than 0.1 voxels for simple deformation modes such as uniaxial tension, while effectively handling complex displacement fields with an average end-point error calculation accuracy of 0.4 voxels. This innovative approach provides an efficient, accurate, and reliable solution for high-performance digital volumetric displacement field analysis.