Drone propeller recognition through machine learning with millimeter wave radar

Mini-sized Unmanned Aerial Vehicles (UAVs) are commonly called "drones", have become ubiquitous in civilian and military applications, readily available off-the-shelf and widely utilized. Detecting and classifying these devices are essential requirements that involve the application of various technologies. The distinctive feature of these drones is their propellers. This thesis concentrates on employing millimeter-wave radar to detect the presence, rotation speed, and dimensions of these propellers. The research explores the integration of a consumer-grade millimeter-wave Frequency Modulated Continuous Wave (FMCW) radar module, specifically the Texas Instruments IWR1843, with machine learning techniques for UAV propeller identification. Utilizing the unique capabilities of millimeter-wave radar technology, this study aims to capture and analyze intricate frequency patterns produced by UAV propellers through machine learning models. Experimental validation in controlled environments demonstrates the potential of the millimeter-wave FMCW radar module, combined with machine learning algorithms, to accurately detect and identify UAVs based on propeller characteristics, achieving up to 92% accuracy in presence, propeller size, and throttle level. Additionally, this thesis highlights the superiority of the proposed ensemble neural network method over a multi-output single Multi Layer Perceptron (MLP) model, as demonstrated through experiments.