Aiming at the high-precision angle estimation requirement of airborne millimeter-wave radar on unmanned aerial vehicles (UAVs) for "low-slow-small" targets, this paper proposes a three-stage step-by-step angle measurement algorithm of "coarse detection - fine estimation - feedback" integrating digital beamforming (DBF), forward-backward spatial smoothing multiple signal classification (FBSS-MUSIC), and an adaptive feedback mechanism. First, DBF is used to quickly and coarsely scan the large spatial area to lock the target azimuth. Then, FBSS preprocessing is introduced in a local small range to restore the array degrees of freedom under coherent signals, and the Akaike information criterion (AIC)/Bayesian information criterion (BIC) is adaptively applied to estimate the number of signal sources. Finally, based on the eigenvalue gap and the equivalent signal-to-noise ratio (SNR) of the target, a confidence assessment index is constructed. A sliding window is used to dynamically adjust the scanning step size and decision threshold to achieve closed-loop feedback. Simulation experiments show that in the dynamic trajectory tracking of 50 consecutive frames, compared with the DBF + MUSIC combined algorithm, the proposed algorithm reduces the average absolute error and computational load by 38.89% and 47.5%, respectively, achieving a balance between high precision and real-time performance. In the tracking experiments where the target moves at speeds of 1, 2, and 4(°)/s, the angle estimation curve of the proposed algorithm is significantly better than those of DBF, MUSIC, and their combined algorithms, verifying its robustness and engineering applicability in practical operating environments.