A Standardization Check Model for Radiotelephony Communication Based on Graph Convolutional Networks and Feature Fusion

To fully exploit both global graph information and local textual features, a hierarchical multi-channel feature fusion framework is proposed based on graph neural networks. This framework integrates local features, temporal dependencies, and graph inference capabilities to enhance the model's representational power. A triple feature interaction mechanism is adopted: local and sequential features are efficiently extracted by employing Multi-scale Convolutional Neural Networks (MSCNN) and Gated Recurrent Units (GRU). In addition, an external knowledge graph tailored to radiotelephony communication is constructed, and global semantic information is incorporated using a Graph Convolutional Network (GCN) to improve the understanding of domain-specific semantic units. Furthermore, an attention mechanism is introduced to dynamically assign weights to different features, thereby enhancing the performance of the hierarchical classification task. Experimental results on the instruction normality test dataset demonstrate that the proposed method achieves an accuracy of 98.03%.