Abstract: To accurately predict the flow field of compressor cascades involving adverse pressure gradients and corner separation flows, an improved Spalart-Allmaras (S-A) turbulence model was developed. Based on the actual structural characteristics of the cascade corner region, the wall distance in the turbulence model was redefined. Correction terms for helicity and adverse pressure gradient were introduced into the transport equation of turbulent eddy viscosity, along with adjustments to model coefficients and inlet parameters. The enhanced turbulence model was applied to NACA65009 cascades, and the Ensemble Kalman Filter (EnKF) data assimilation method was employed to calibrate correction coefficients and inlet boundary conditions using experimental static pressure coefficient data at 50% and 5.4% span. Results demonstrate that the modified S-A model significantly improves prediction accuracy for corner separation flows in compressor cascades. Compared with the original S-A model, prediction deviations were reduced by over 80%. The improved model also shows substantially smaller discrepancies when validated against independent experimental data excluded from calibration. Validation under multiple operating conditions confirms the enhanced generalization capability of the modified S-A turbulence model across different working regimes.