Civil aviation radiotelephony communication is the main way of voice communication between controllers and pilots. Verifying radiotelephony communication’s read-back, extracting intent information, and making control decisions through artificial intelligence is significant. Training the deep models necessitates a lot of text data. The radiotelephony communication data is not easy to obtain and often requires extensive manual correction, the use of machines to automatically generate standardized text data has become a pressing issue that must be addressed. Aiming at the problems of error accumulation and exposure bias in traditional supervised learning text generation methods, this paper proposes an intelligent text processing method for civil aviation radiotelephony communication based on Generative Adversarial Networks (GAN), which can achieve high-quality command generation and read-back generation. This paper's key contributions are in two areas. Firstly, this paper proposes a text generation method for civil aviation radiotelephone communications by introducing a Long Short-Term Memory (LSTM) network-based sequence-to-sequence (Seq2Seq) model into a GAN model as its generator. It enhances the ability to capture text features and can generate samples that are as close to the actual text as possible to confuse the discriminator's judgment. Secondly, this paper presents a text categorization approach for civil aviation radiotelephony communication based on Convolutional Neural Network (CNN) and cross-entropy function. The discriminator of GAN built with the CNN model strives to distinguish between real text and generated text without being fooled by the generator. At the same time, the concept of policy gradient in reinforcement learning is introduced during adversarial training. The generator and the discriminator achieve a dynamic balance in the alternating adversarial training. According to experimental findings, the designed model outperforms other comparable models and can automatically generate texts that conform to civil aviation radiotelephone communication specifications without supervision. It can provide new ideas for the intelligent feedback system of civil aviation radiotelephony communication, speed up the digitization process of air traffic control, and lay a solid foundation for reducing crew operations.
An Intelligent Text Processing Method for Civil Aviation Radiotelephony Communication Based on Generative Adversarial Network
18.09.2022
1617053 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Engineering Index Backfile | 1921
|British Library Online Contents | 2018
|Engineering Index Backfile | 1932
|TIBKAT | 1984