To address the insufficient positioning accuracy of radar emitters under low time-difference precision in traditional TDOA (Time Difference of Arrival) localization algorithms, this paper proposes a localization estimation algorithm based on TDOA and a modified Unscented Kalman Filter (UKF). First, the hyperbolic TDOA algorithm is employed to calculate the initial position estimate of the radar emitter. Subsequently, the modified UKF algorithm is applied to refine the positioning accuracy. The algorithm employs the SAGE-HUSA method with an exponentially decaying memory-weighted factor to adaptively estimate system noise characteristics in real time, while incorporating an innovation covariance matching criterion to prevent filter divergence and thereby enhance filtering stability. Experimental and simulation results demonstrate that, compared to conventional TDOA methods, the proposed algorithm effectively reduces state estimation errors and enhances localization precision.