Specific Emitter Identification Based on Knowledge Embedding

Specific Emitter Identification (SEI) technology has become one of the core components within electronic reconnaissance systems due to its crucial role in battlefield situational awareness and decision-making support. Addressing the limitations of traditional data-driven electromagnetic signal identification models, characterized by weak generalization capabilities and poor robustness, which hinder the attainment of reliable identification accuracy, this paper introduces a knowledge-embedding approach for SEI. To begin with, we establish an electromagnetic signal feature knowledge graph encompassing four categories and eleven distinct features. This electromagnetic signal knowledge graph serves to construct an electromagnetic signal knowledge repository and uncover pro-found interconnections among various features. Subsequently, we employ Relation-Graph Convolutional Networks (R-GCNs) to learn from the constructed knowledge graph and successfully carry out the task of specific emitter identification. Results obtained from real-world Automatic Dependent Surveillance-Broadcast (ADS-B) datasets demonstrate that even at a signal-to-noise ratio of 0dB, an accuracy exceeding 85% can be achieved. The overall recognition rate significantly outperforms three comparative models: CNN, CLDNN, and ResNet.