A Multimodal Deep Learning Approach for the Detection of Unmanned Aerial Vehicles Using Radio Frequency and Acoustic Signals

Unmanned aerial vehicles (UAVs) or drones are rapidly gaining popularity in almost all fields of scientific research, agriculture, photography, sports racing, rescue operations, and even entertainment.However, as this technology becomes more widely used, serious concerns arise, such as military threats, terrorism, privacy violations, and public safety.This highlights the urgent need for reliable and rapid identification systems to detect potentially hostile drones.In this work, we present a novel method for automatic drone detection by exploiting radio frequency (RF) signals and acoustic signatures generated by their propellers.Our approach utilizes a combination of traditional AI algorithms and deep learning models, improving detection and classification efficiency.This work analyzes the effectiveness of a variety of classifiers, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and support vector machines (SVMs).The results demonstrate the effectiveness of our feature multimodal approach against publicly available drones, especially in low signal-to-noise ratio (SNR) environments.The long shortterm memory (LSTM) based model provides a classification accuracy of 92% at -10 dB noise.