In order to solve the problem of determining whether the pilot's recitation of instructions is consistent with the in-structions given by the controller,a semantic verification model for land-air calls is constructed,it is Ernie-Gram-BiGRU-Atten-tion.First,Ernie-Gram module is used to encode the text-to-message information.BiGRU module is used to further extract the fea-tures from the encoded text-to-message information to obtain the compressed sentence vector.Since the keyword information is cru-cial in land-air calls,the attention mechanism is used to realize the information extraction of keyword features,and the fully con-nected layer is used to fuse the attention vector with the sentence vector.Finally,the fused vectors are spliced with the categorized word vectors output from the Ernie-Gram module and sent to the fully-connected layer for semantic verification of text pairs.The constructed model achieves 97.1%accuracy on the land-air call text pair dataset,which is a 2%improvement in accuracy compared to the Ernie-Gram benchmark model,and an improvement in precision,recall,F1 value,and accuracy compared to other Bert-based improved models.
关键词
空中交通安全/迁移学习/陆空通话/Ernie-Gram/注意力机制
Key words
air traffic safety/transfer learning/land-air calls/Ernie-Gram/attention mechanisms
维普
万方数据