Research On Multi-modal Emotion Recognition Algorithm For Air Passengers

At present,the heavy traffic in the terminal building and the heavy working pressure of the staff make it impossible to provide timely services for each passenger and avoid risks in time before passengers' conflicts.Aiming at the above problems,a multi-modal emotion recognition algorithm based on self-attention mechanism is studied,which uses video,audio and text three-modal information to identify passenger emotional states and provide support for caring service.At the same time,an air passenger threat assessment method based on multimodal emotion fusion is proposed,which recognizes the threat degree of air passengers based on the change of emotional state reflected in the dialogue between staff and air passengers or passengers,and provides support for risk management.According to the above,the following three issues are mainly studied:Firstly,the multi-modal emotion recognition algorithm based on self-attention mechanism is studied.The self-attention mechanism is adopted to perform feature layer fusion to obtain multimodal emotional features of video,voice and text,distinguish the importance of features,and reflect innovation.Afterwards,emotion classification is realized by a multi-modal emotion recognition algorithm based on multiple gated loop units.At the same time,the interactive emotion recognition parameter optimization method is used to adjust the hyperparameters and optimize the algorithm structure.The improved algorithm effectively improves the accuracy of emotion recognition.Secondly,a multi-modal emotion database building method based on the form of "reality + art reproduction" is proposed.Considering the difference between Chinese and foreign expression methods,the relevance of civil aviation realistic films and that the situation of disguised emotions cannot be solved,taking Chinese civil aviation realistic films as sample sources,referring to the labeling method of the IEMOCAP database,database marked with multi-modal emotional states and threats in civil aviation environment was built to make the threat assessment of air passengers in a reasonably reproducible scenario,reflecting innovation.Thirdly,an air passenger threat assessment method based on multi-modal emotion fusion is designed.After extracting word vector representation of multi-modal emotional states by the GloVe method,the threat assessment is carried out through the bi-directional Long Short-term Memory Network,which reflects innovation.Finally,the multi-modal emotion recognition network model and the air passenger threat assessment network model are obtained through learning,and the effectiveness of these two models is verified through practical application tests.