Enhancing StrainNet-3D with a hybrid attention mechanism for spatial high-frequency displacement measurement

The conventional digital image correlation (DIC) method often encounters challenges with process complexity and time-consuming. Recently, the convolutional neural network has shown considerable promise in handling DIC measurement tasks. Notably, the StrainNet-3D model has successfully measured common spatial low-frequency three-dimensional displacement fields. In this study, we enhance the StrainNet-3D model by integrating a hybrid attention mechanism (HAM), resulting in the optimized model named HAM-StrainNet, which excels at spatial high-frequency (SHF) displacement measurements. We constructed a speckle dataset featuring a variety of types, multiple categories, and differing interpolation region sizes for model training. The integration of a multi-level feature extraction module and a HAM significantly improves the model's capability to represent SHF displacements. Experimental results confirm that HAM-StrainNet outperforms its predecessor in measuring SHF displacement fields.