Radar Phrase Extraction Based on Deep Convolutional Conditional Random Fields

Extracting radar phrases from multifunctional radar pulse sequences and identifying the internal operation law and working mechanism among the working modes is the foundation for subsequent working mode prediction. However, most existing studies at the syntactic level fail to address the full range of issues, focusing only on radar word extraction and working pattern recognition. In this paper, we propose an improved CNN-BiLSTM-CRF model for radar phrase extraction. It is based on the BiLSTM network model, which is able to extract richer local features using convolutional neural networks. The model is simulated on generated radar pulse word sequences, and the results show that the CNN-BiLSTM-CRF model improves the F-value of radar phrase extraction by 2.688% on the test set compared with the BiLSTM-CRF model. The simulation results prove that the improved method is effective.