Abstract: With the widespread use of low-altitude， slow-speed， and small unmanned aerial vehicles （UAVs） across various applications， the associated airspace security risks have become increasingly significant. Consequently， the development of high-precision and reliable UAV identification technologies has become essential for effective low-altitude risk management and airspace safety assurance. Physical parameters such as rotor speed and blade length play an important role in UAV identification. Therefore， this study focuses on accurately estimating these parameters through time-frequency analysis of UAV echo signals. However， conventional parameter estimation methods based on traditional time-frequency analysis often suffer from spectral leakage and cross-term interference when processing the complex echoes of multi-rotor UAVs. These limitations lead to blurred micro-Doppler signatures and prevent the estimation accuracy from meeting practical application requirements. To address these issues， a feature extraction method based on the threshold energy local maximum synchronous extraction transform （TELSET） is proposed. The proposed method determines an adaptive energy threshold according to the background energy distribution in the time-frequency domain， effectively suppressing energy diffusion caused by fixed windowing and reducing cross-term interference in multi-rotor echo signals. In addition， the time-frequency ridge extraction strategy of the local maximum synchronous extraction transform （LSET） is further improved by applying Gaussian sliding averaging along the time axis， which enhances the energy concentration of rotor micro-motion features. Validation experiments using measured millimeter-wave radar data and a public L-band dataset demonstrate that the TELSET method achieves both high time-frequency resolution and accurate rotor parameter estimation. Compared with the short time Fourier transform （STFT）， the proposed method reduces entropy by 2.86 dB， decreases the rotor speed estimation error to 0.86%， and improves blade length estimation accuracy by an average of 6.17%. These results demonstrate that the proposed method provides a reliable technical approach for UAV identification in low-altitude environments.