Radar Emitter Sorting Based on Multi‐Head ResGAT

ABSTRACT Conventional graph neural networks (GNNs) fail to effectively capture high‐order relationships among radar pulses, thereby compromising discrimination accuracy in precise signal sorting. Therefore, this paper proposes a radar emitter signal sorting method based on an enhanced graph attention network (GAT). The model combines a multi‐head attention mechanism with a residual network structure, enabling dynamic weight allocation to graph nodes. This effectively captures the complex correlation patterns of radar signals across a multi‐dimensional parameter space and thus enhances classification performance. In scenarios with scarcely available labels and complex signal features, the proposed method demonstrates stronger average accuracy and robustness when handling radar signal sorting tasks .