A displacement hybrid aware network for real-time high-resolution digital image correlation

Abstract Digital image correlation (DIC) has emerged as a pivotal full-field deformation measurement technique for non-destructive testing systems, playing a critical role in material characterization and structural health monitoring. Nevertheless, achieving real-time, high-resolution, and high-accuracy performance remains a significant challenge. In this paper, we introduce a novel deep learning approach to address these challenges for DIC. The proposed method addresses the limitations of conventional and existing deep learning-based DIC approaches, which either rely on local iterative optimization or incur excessive computational cost when enforcing global displacement consistency. The proposed network integrates convolutional neural networks (CNNs) and the Transformer architecture. Instead of performing global modeling directly on high-resolution image features, a displacement-transformer module is introduced to capture long-range motion dependencies within a semantically compact displacement feature space, thereby significantly reducing computational overhead. Local deformation details are preserved through displacement-aware feature encoding and factorized displacement correlation, ensuring both high spatial resolution (SR) and global physical consistency. To evaluate the effectiveness of the proposed method, systematic experiments were conducted on the standard DIC Challenge benchmark, where it achieves a validation average endpoint error of 0.037 pixels, along with a SR of 27.93 pixels and a measurement resolution of 9 × 10 − 3 pixels. In terms of efficiency, the reported performance corresponds to the model inference speed, achieving up to 1.8 × 10 7 points of interest per second (POI/s). Furthermore, practical validation in composite material elastic modulus estimation results in a relative error of approximately 1.1 % , indicating its accuracy for real-world material characterization. The results demonstrate that the proposed method offers a dependable and effective DIC solution, rendering it highly suitable for implementation in intelligent inspection technologies.