Classification of Drones vs Non-Drone Targets using Deep Learning on Radar tracks

The last decade has seen rapid advancements in drone technology. The threats posed by them on strategic assets like warships, airports in present time is unmatched. Hence, it is paramount to device counter measures against non cooperating drones. Radars are an efficient solution in long range detection of drones due to their all weather operational capability. Even though micro-doppler signature of drones is a promising feature to classify them from other non-cooperating targets like birds, its efficacy at long range is yet to be proved. We propose two new methods to classify radar tracks into two classes-drone or non-drone using deep learning. In the first method, a process to capture the target trajectory pattern and its kinematics into an image is proposed. Then, these images are classified using CNN based classifier. We propose a new CNN based architecture to classify radar tracks in the second method. The proposed classifiers are compared against a Random Forest classifier using hand crafted features.