Abstract: Individual identification of drones is an important aspect of drone supervision. This article proposed a method for extracting drone RF fingerprint features based on frame division short-time Fourier transform and combined it with ShuffleNetV2 for drone individual recognition. Compared to the short-time Fourier transform, the segmented short-time Fourier transform used information entropy to preprocess the data of unmanned aerial vehicle RF signals into frames, and then extracted the time-frequency characteristics of the signal through the short-time Fourier transform. Frame based preprocessing effectively improved the problem of local feature instability caused by time-varying or sudden changes in drone signals. This article used publicly available datasets for simulation verification and built an experimental platform for experimental verification. The proposed algorithm improved recognition rate by more than 4% compared to the algorithm without preprocessing.