Abstract: Aiming at the key problems of low detection probability and insufficient robustness of Drone Identification (Drone-ID) signal in low SNR scenarios, a detection method based on pulse compression and multi-scale dual correlation fusion is proposed. The linear frequency modulation signal carrying Zadoff-Chu (ZC) sequence in Drone-ID signal of DJI OcuSync protocol is taken as the detection target. Firstly, the signal energy focusing is realized by pulse compression, and the time-frequency domain joint feature enhancement is completed by combining multi-scale delay correlation. Finally, the accurate recognition of real signal is completed based on multi-dimensional feature matching. Experiments show that when the signal-to-noise ratio is as low as -5dB, the detection probability of this method reaches 92.1%, which is 19.8 percentage points higher than that of the traditional frequency domain cross-correlation method, and the false alarm rate is controlled below 1%. It provides a new technical path for reliable detection of Drone-ID signals in complex electromagnetic environment, and has both theoretical value and application prospect.