Accuracy enhancement of DVC measurement on volumetric images reconstructed with fewer projections via deep learning

Digital volume correlation (DVC) extracts full-field internal deformations of solid objects or biological tissues by correlating volumetric images reconstructed using dense scans. However, due to the limitations of reconstruction methods for volumetric images, when enhancing the efficiency of DVC experiments by sparse scans, severe image artefacts occur and hence substantially reduce the accuracy of DVC measurements. To address this issue, a recently established implicit Neural Representation learning with Prior embedding (NeRP) network is introduced to reconstruct volumetric images with fewer projections. By using a reference volume and 30 projection images as network inputs, high quality deformed volume can be obtained that enhance the accuracy of DVC measurements. For validation, both numerical simulation and real uniaxial compression experiments were conducted to evaluate the displacement measurement accuracy of the proposed method. The experimental results demonstrate that the proposed method significantly reduces displacement errors compared to conventional reconstruction methods under sparse scan conditions. Specifically, the errors in the u and v directions decrease by at least an order of magnitude, while the w direction error is reduced by a factor of two. This improvement enables more accurate DVC measurements with sparse scans.