The study addresses the critical task of distinguishing UAVs from birds using mmWave MIMO radar under conditions where optical sensing is unreliable . Micro-Doppler signatures generated by wing flapping and propeller rotation provide discriminative phase information in complex I/Q radar signals . A region-of-interest is selected via the mean resultant length to isolate coherent micro-motion components . The radar echoes are decomposed by Complex Empirical Mode Decomposition (CEMD) to obtain intrinsic mode function pairs . Features include instantaneous frequency, bandwidth, total energy, and spectral entropy of IMFs . A classification pipeline using PCA and RBF-SVM is trained on controlled anechoic-chamber data . The method achieves about 95% accuracy across cross-validation and random train–test splits . Permutation analysis shows high-frequency IMFs and spectral entropy are most informative . Robustness is validated under additive complex Gaussian noise using EEMD variants . The framework offers a scalable basis for real-time counter-UAS classification with future work on field deployment .