<p>Mirror milling trajectory planning based on measure–machining integration establishes the initial locating datum through a global scan of workpiece surface, providing the basis for trajectory planning. However, during mirror milling, thin-walled part undergoes local deformation due to cutting forces, supporting forces, and residual stresses, which can invalidate the initial datum and degrade wall-thickness accuracy. To address this issue, a mirror milling datum reconstruction and trajectory adjustment method based on combined global and local measurement was established in this paper. Specifically, global measurement is used to obtain the clamped workpiece geometry and establish an initial machining datum, while online local measurement is used to update the machining datum during milling, enabling subsequent trajectory adjustment. An online local surface measurement system was developed to acquire local surface data in real time. A machining datum reconstruction model based on Improved Generalized Radial Basis Function Neural Network (IGRBF-NN) was constructed by introducing an MLP-based adaptive weighting function into the conventional RBF-NN to reweight measurement points during reconstruction, enabling fast and accurate online machining datum reconstruction. Based on the reconstructed datum, the supporting side trajectory is updated by projecting the next position point onto the reconstructed surface, and the milling side trajectory is synchronously adjusted according to mirror-milling geometric constraints to maintain wall thickness. Comparative mirror milling experiments on grid panels demonstrate that the proposed method satisfies the machining requirements and improves wall-thickness uniformity.</p>