Tracking-Aided Open-Set Recognition of Radar Formation Data Based on Fourier Descriptor Features

Accurate recognition of group target formations in radar data is crucial for comprehending the targets' intentions. This paper proposes a tracking-aided open-set formation recognition method using Fourier descriptors (FDs). By leveraging temporal correlations in consecutive radar detection frames, we first apply tracking algorithm to enhance recognition stability. Then a contour coordinate extraction algorithm combines measurements source model and 2D convex hull to derive formation shapes, later parameterized as FDs to serve as the recognition input. Additionally, we design a 1D-CNN open-set classifier for formation classification, which introduces probability vector entropy to detect unknown class. Simulations demonstrate that the proposed method achieves temporally accurate and stable open-set recognition performance in multi-temporal radar data.