Aerial Vehicle Detection and Classification Through Fusion of Multi-Domain Features of Acoustic Signals

Detection of aerial vehicles is a challenging task. Over time, the significance of developing such vehicles has increased rapidly. However, an utmost concern arises from the potential misuse of these small-sized vehicles in illicit activities such as unauthorized surveillance, smuggling contraband, or disrupting critical infrastructure significant security threats. Recognizing the gravity of this issue, we have opted to introduce an innovative approach to automate the detection processes for aerial vehicles. This initiative aims to deploy effective drone detection systems to mitigate risks and safeguard against illegal activities. We put use to two datasets from GitHub depot. Cepstral, spectral, and time domain features were extracted from the data, followed by classification. We conducted two experiments, with the first (Drone, No-Drone) yielding 98.6% accuracy, using the Ensemble (Bagged Trees) classifier. The second set of experimentation addressed five classes: Background Noises, Bebop Drone, Drone, Helicopter, and Mambo Drone. Ensemble (Bagged trees) again outperformed all other classifiers and achieved 98.3% accuracy. The results highlight that our proposed framework gives effective results based on audio signals of different aerial vehicles.