UAV Detection System with Multiple Acoustic Nodes Using Machine Learning Models

Class 1 unmanned aerial vehicles (UAVs), known as drones, have become popular and accessible, which makes them tools for malicious purposes. As a result, there is an increasing demand for an effective defense system that can detect UAVs. In this paper, a UAV detection system with multiple acoustic nodes using machine learning models is proposed along with an empirically optimized configuration of the nodes for deployment. Features including Mel-frequency cepstral coefficients (MFCC) and short-time Fourier transform (STFT) were used for training. Support vector machines (SVM) and convolutional neural networks (CNN) were trained with the data collected in person. Experiments were done to evaluate models' ability to find the path of the UAV that was flying. Sensing nodes were placed in four different configurations and the best of test set was chosen which maximizes the detection range without blind spots. STFT-SVM model showed the best performance and a semi-circle formation with 75 meters distance between a node and the protected area is found to be the optimized configuration.