Traditional digital image correlation (DIC) methods that rely subset-based correlation calculations are prone to decorrelation and strong parameter sensitivity under rotational motions. To overcome these limitations, this study proposes a Gaussian process regression-guided self-supervised learning DIC method (GPR-SSL-DIC) for accurate rotational displacement field measurement. The method develops a self-supervised learning framework based on the Kolmogorov-Arnold Network (KAN) network, in which a loss function is formulated using the grayscale differences between the reconstructed and reference images together with a displacement-field smoothness constraint, driving adaptive optimization of the displacement field and thereby overcoming the limitations of the conventional subset-matching paradigm in traditional DIC. To improve convergence under large-angle rotations, rotation-invariant SURF feature points are detected in the structural target region, and their displacement information is used to construct sparse observation samples. Furthermore, Gaussian process regression is employed to predict a global displacement field as an initial solution, thereby guiding the network to converge toward the true solution space. Numerical simulations show that the proposed method achieves average end-point errors below 0.001 7 pixels under rigid-body rotation and coupled large-deformation conditions, and below 0.007 4 pixels with sinusoidal displacements are superimposed, corresponding to a 93.5% improvement over traditional DIC. Rotating-blade dispacement experiments further demonstrate that at a 9° inter-frame rotation, the standard deviation of fixed-point distance measurements decreases by 54.3% compared with traditional DIC. At 30°, where traditional DIC fails due to severe decorrelation, the proposed method is still able to obtain the displacement distribution of the blade. These results confirm that the proposed framework is robust under large-angle rotational conditions and offers an effective solution for displacement field measurement in rotating structures.