Motion-Based Bird-UAV Classification Using 3D-CNN for Long-Range Anti-UAV Systems

Abstract

The increasing threat of malicious unmanned aerial vehicles (UAVs) necessitates robust anti-UAV systems. However, their performance is often degraded by bird misclassification caused by low-resolution imagery and unseen UAV types. This study proposes a motion-based 3D convolutional neural network (3D-CNN) trained on image sequences acquired from a radar-camera integrated anti-UAV solution. The proposed method effectively distinguishes UAVs from birds, even under low-resolution conditions and when encountering previously unseen UAV types.