Convolutional Neural Networks for Analyzing Unmanned Aerial Vehicles Sound

The emergence of Unmanned Aerial Vehicles (UAV) is pervasive throughout society. A growing segment of usage is of a dubious nature for harassment, illegal activity and terrorism. Detection of unknown UAV's has become a requirement for many organizations and agencies to thwart the emergence of UAV's that are in some way threatening. To detect UAV, the use of acoustic signals has become an useful area of research. Convolutional Neural Networks (CNNs) are one of several models of deep learning, applied in various fields such as image recognition and natural language processing. In this project, we design a system to detect the presence of possible detection and payload detection using CNNs on the basis of sound data generated from UAV flights. The sound of recorded drones is pre-processed into spectral data by Fast Fourier Transform (FFT) and Mel-Frequency Cepstrum (MFCC) and given as the input value to the CNN model. The results show that it is possible to detect and differentiate UAVs which have standard weight and also with additional payload. In short, the project has two detection goals. One is the acoustic detection of a UAV, and the second is the determination if that UAV has a payload.