Comparative Analysis of Artificial Intelligence Methods for Unmanned Aerial Vehicle (UAV) Recognition and Identification Using Micro-Doppler Signatures

The significance of small Unmanned Aerial Vehicles (UAV) in modern warfare, highlighted by recent conflicts such as those between Russia and Ukraine, necessitates urgent measures to address their diverse and persistent threats [1]. Efforts must prioritize the enhancement of UAV recognition and identification capabilities to effectively counter their impact on the battlefield. In this work, we introduce a novel methodology for UAV classification leveraging their unique micro-Doppler signatures (mDs). Our approach involves the direct application of Recurrent Neural Network (RNN) techniques to temporal micro-Doppler signals. Specifically, we have constructed neural network architectures incorporating Gated Recurrent Unit (GRU) layers, resulting in classification accuracies surpassing 98%. To comprehensively evaluate our methodology, we compare our findings with two alternative approaches for UAV classification: one employing RNN-based methods applied using mDs representations like spectrograms, and another utilizing Convolutional Neural Networks (CNN)-based networks where mDs are represented as spectrograms transformed into images.