Micro-Doppler Analysis under Various Aspect Angles for Small UAV Classification

When a drone is identified using micro-Doppler (μ-D) signatures, the aspect angle changes continuously as the drone approaches. Thus, analyzing the relationship between the aspect angle and μ-D signatures is necessary. In this study, we conduct experimental learning of μ-D signatures at multiple aspect angles and perform identification evaluation on four types of drones using the proposed method of projection metric learning (PML). PML is shown to have strong identification performance because it uses the eigenvector space itself as input. This study found that when the aspect angle of unknown data is different from that of training data, PML shows 85.0% accuracy in the identification rate, which is 6.8% higher than that achieved using the conventional method.