An Advanced Joint Extraction Method of ATC Intention and Information

To record and analyze control instructions through natural language processing technology, extract control intentions and key information, reduce potential flight conflicts, and ensure aircraft flight safety. This article proposes an improved joint extraction model for regulatory intent and information, CII-BERT. The pre trained language model BERT is used to semantically represent regulatory instructions, and then DNN is used for intent recognition and information extraction. Experimental verification was conducted on the collected control instruction dataset, and the results showed that CII-BERT can significantly improve the accuracy of control intent recognition and control information extraction. The experimental results further reveal that after continuous pre training of BERT, the performance of the model in downstream tasks is further improved, with an accuracy rate of no less than 99%. The CII-BERT model performs better in the task of identifying regulatory intent and extracting regulatory information.