UAV Velocity Prediction Using Audio data

Abstract

The Federal Aviation Administration (FAA) set the Unmanned Aerial Vehicles (UAV) speed limit at 100 mph. This research focused on detecting when the UAV exceeds a speed limit for an experiment and using the sound dataset to predict the velocity of a UAV. It is hard to detect a malicious UAV, but we can assume that a UAV over 100 mph is most likely malicious. An indoor environment will be used as a controlled environment and the dataset is divided into two classes: slow (0- 9mph) and fast (over 10mph). Support Vector Machine (SVM), Random Forest, and Light Gradient Boosting Machine (LGBM) were the Machine Learning models used for this research, and Convolutional Neural Network (CNN) was the Deep Learning model used for this research. The result shows that the CNN model has the highest performance (F-1 score: 1.0, Accuracy: 1.0, Recall: 1.0, Precision: 1.0) for classifying the sound of the UAV speed.