Semantic Recognition Of Radiotelephony Communication In Civil Aviation Based On Convolutional Neural Network

In radiotelephony communication,the readback between pilots and air traffic controllers(ATC)is an important routine to guarantee the civil aviation safety.Although the International Civil Aviation Organization(ICAO)has improved the communication standard,the incorrect or wrong readbacks are still unavoidable.Therefore,it is significant to detect the semantic inconsistency in readback for the civil aviation safety.Recent progress in natural language processing has shown that CNN(Convolutional Neural Network)based models have outstanding performance on semantic analysis.Thus,it is of great significance to study the semantic analysis and recognition method of radiotelephony communication based on CNN model.The main contributions of this thesis are summarized as follow:(1)According to the real radiotelephony communication recordings,a corpus of radiotelephony communication is established.(2)The semantic recognition method of radiotelephony communication is studied based on traditional CNN model.The samples from the built corpus are segmented into words and represented by one-hot or word2 vec.Then the traditional CNN is applied to map the meanings of instructions and readbakcs to semantic vectors respectively.Similarities between vectors are employed to recognize the semantic consistency.(3)Two improved CNN models are proposed.The first model is constructed by changing the size of the convolution kernel;the second model is constructed by transforming the instruction and readback into one matrix as the input of CNN.The two improved CNN models are utilized to analyze the semantic of radiotelephony communication respectively.The experimental results show that CNN is effective in semantic recognition of radiotelephony communication,and the second improved CNN model can achieve the highest average test accuracy rate at 84.50%.