Classification of Bird and Drone Targets Based on Radar Data Augmented with Feature Engineering using Trajectory Kinematics Analysis and using Random Forest Algorithm

This paper investigates the application of Artificial Intelligence (AI) and Machine Learning (ML) algorithms for the classification of bird and drone targets in radar systems. It gives stress on using motion characteristics and Random Forest machine learning model. It details about dynamic motion of Bird and Drone tracks obtained from target tracks from X-Band Drone Tracking Radar. Multiple motion descriptors were derived from the said Radar track data and tested and finally three key motion descriptors namely Range Rate, Curvature and Oscillation Factor were identified for pursuing further. These derived features are added to the existing data obtained from the Radar Data Processor and both are merged to form a consolidated dataset for various tracks. Thereafter, Random Forest (RF) machine learning model was applied on the consolidated data. The approach is validated using real time data from Bird and Drone tracks by integrating trained machine learning models with X-Band Radar. The proposed hybrid approach achieves better classification accuracy and lower false-positive rates in real-world scenarios. This paper is centered around low-altitude airspace surveillance aspects for modern evolving scenarios where drones and UAVs are used in abundance.