Comparative Analysis of Deep Learning Methods for Unmanned Aerial Vehicles (UAVs) Recognition and Identification Using Micro-Doppler Signatures

In this work, we introduce a novel methodology for small Unmanned Aerial Vehicles (UAVs) classification by leveraging their specific micro-Doppler signatures (mDs). The proposed approach involves the direct application of Recurrent Neural Network (RNN) techniques to temporal radar signals. Hence, we have constructed neural network architectures incorporating Gated Recurrent Unit (GRU) layers, achieving classification accuracies of 100%. To comprehensively evaluate our methodology, we compare our findings with two alternative approaches for UAVs classification: one utilizing Convolutional Neural Networks (CNN)-based networks where mDs are represented as spectrograms transformed into images, and another employing RNN-based methods applied to spectrograms of the mDs.