Long Short-Term Memory Model for Drone Detection and Classification

In this paper a method for detection and classification of radio signals from civilian drones is proposed. Raw, windowed in-phase and quadrature data is used as input for a simple long short-term memory model without further feature engineering. A description of the used dataset, the data processing and the used machine learning model is provided. The influence of the cell size on the model performance is evaluated and a cell size of 256 is chosen for further testing. With this model an accuracy of more than 99.99% is achieved for drone detection. For classification of seven drone types and two noise classes an overall accuracy of 80.66% is achieved.