再入飞行器在大气层穿越过程中，面临宽速域与强可压缩效应下复杂的边界层转捩问题，准确预测转捩位置是气动热设计和性能分析的关键，而面向宽速域工况，单一全局修正系数的泛化能力有限。本文提出一种基于本征正交分解（POD）与高斯过程回归(GPR)相结合的两阶段特征增强代理建模方法，通过联合来流方向速度主模态系数、马赫数与飞行高度，实现对修正系数的高保真预测，并建立基于马赫数的分层-融合框架，显著提升了高超声速下修正系数预测的物理一致性与泛化能力。基于上述高保真修正系数预测方法，在间歇率输运方程的产生项中引入预测结果作为可压缩性修正系数，实现了对高马赫数边界层转捩预测的物理一致性。典型钝体再入飞行器工况下的数值验证结果表明，所集成的高保真代理与修正转捩模型的方法体系在复杂参数空间内展现出良好的稳定性与适应性，能够准确预测转捩起始位置等关键转捩特征，对壁面摩阻等流动变量亦具备合理的分布趋势捕捉能力，具有显著的工程应用潜力。

During atmospheric re-entry, vehicles experience complex boundary-layer transition phenomena characterized by a wide range of flight speeds and strong compressibility effects. Accurate prediction of transition location is crucial for aerodynamic heating design and performance analysis; however, employing a single global correction coefficient limits model generalization across broad speed regimes. To address this issue, this paper introduces a two-stage feature-enhanced surrogate modeling approach based on Proper Orthogonal Decomposition (POD) combined with Gaussian Process Regression (GPR). By integrating dominant modal coefficients of streamwise velocity with Mach number and flight altitude, the method achieves high-fidelity prediction of the correction coeffi-cient. Moreover, a hierarchical fusion framework based on Mach number is established, significantly enhancing the physical con-sistency and generalization capability of the correction coefficient predictions under hypersonic conditions. Utilizing these high-fidelity correction coefficients within the production term of the intermittency transport equation effectively ensures the physical con-sistency of transition prediction at high Mach numbers. Numerical validation conducted for typical blunt-body re-entry vehicle sce-narios demonstrates that the proposed integrated methodology—comprising high-fidelity surrogate modeling and the corrected tran-sition model—exhibits robust stability and adaptability within complex parameter spaces. The approach accurately captures key tran-sition characteristics, including the transition onset position, and provides reasonable distribution trends of flow variables such as wall friction, highlighting its substantial potential for engineering applications.