In complex low-altitude electromagnetic environment, the frequency hopping (FH) signal of unmanned aerial vehicle (UAV) is affected by OFDM aliasing interference and strong noise, which leads to the lack of anti-noise robustness and feature fidelity of the existing noise reduction algorithms. To solve this problem, a set of frequency hopping signal noise reduction method is proposed, which integrates blind separation preprocessing, adaptive time-frequency mask noise reduction and parameter estimation verification. Uses the double sliding window soft and hard threshold adaptive separation algorithm, strips the OFDM broadband interference, provides high purity input for the noise reduction link, through the time-frequency mask noise reduction algorithm based on robust peak tracking, combined with two-dimensional morphological post-processing, realizes the high fidelity purification of frequency hopping characteristics under strong noise, and then verifies the performance of the algorithm through the blind estimation of frequency hopping parameters. The simulation results show that the proposed algorithm can completely recover the time-frequency trajectory of frequency hopping in the extreme low SNR of-16 dB, effectively suppress the noise and false peak interference, and the performance is significantly better than traditional algorithm, which can provide reliable technical support for UAV RF fingerprint identification.