Research On Semantic Analysis And Verification Of Radiotelephony Communication

In the process of civil aviation transportation,safety is the top priority.As a means of information exchange between pilots and controllers,the accuracy of radiotelephony communication is an important factor affecting the flight safety of aircraft.The problem of inconsistent information between the two parties of radiotelephony communication will threaten flight safety,so how to ensure the accuracy of radiotelephony communication has always been the focus of the civil aviation industry.At present,there is less analysis of the actual content of radiotelephony communications,and the research content focuses on repetition errors and lack of research on question and answer.Therefore,it is of great significance to build an automatic semantic analysis model suitable for radiotelephony communication.The content of this article is as follows:Firstly,aiming at the question-and-answer and recitation dialogues of radiotelephony communication,a radiotelephony communication database is established.According to the relevant standards of radiotelephony communication,the language characteristics of radiotelephony communication are analyzed.On the basis of the existing radiotelephony communication recitation corpus,question-and-answer corpus is added and the data is rearranged.Secondly,based on the characteristics of radiotelephony communication language,a method of semantic analysis and verification of radiotelephony communication based on attention mechanism is proposed.In order to obtain rich semantic information input,this method combines word vectors,character level vectors,and feature flag information in the input module to represent instructions and responses.Then,Bi-directional Long Short-Term Memory Recurrent Neural Network(BiLSTM)is used for semantic encoding,and a soft-aligned representation of instructions and responses is constructed by constructing an attention interaction matrix.A feedforward neural network and a method of enhancing differentiation is used for semantic contrast analysis,and finally the matching features are identified through the fully connected network.Thirdly,a semantic analysis and verification model of radiotelephony communicationbased on dense network and multi-layer attention mechanism is proposed.The model uses a densely connected BiLSTM structure,and joins multiple layers of attention mechanisms in the network to obtain semantic vectors containing matching features,and then uses average pooling and maximum pooling to extract matching features.Finally,the matching feature analysis is performed by associating the two semantic vectors with the difference strategy,and the matching verification task is completed.Semantic analysis of radiotelephony communication recitation and question and answer content using the two proposed models proves that the automatic analysis and verification model of radiotelephony communication is feasible.By comparing the experimental results of the two models,it is shown that the attention mechanism can improve the effect of semantic analysis,and the semantic analysis and verification model based on multi-layer densely connected networks and multi-layer attention is the best.