Research On The Fatigue Detection Methods Of Air Traffic Controller And Application System Design Based On Cognitive Science

With the continuous development of China's civil aviation industry,the security pressure of air traffic control is increasing day by day,and unsafe incidents occur from time to time.Controllers make the important role of air traffic controllers in civil aviation.Control fatigue is a potential hazard of aviation safety.Controller fatigue detection has become one of the hot issues in civil aviation research.Based on the research of related fatigue detection methods at home and abroad,this paper proposes a fatigue detection method for controllers based on cognitive science.The application of cognitive science in fatigue detection is mainly based on behavioral feature detection.Due to the particularity of the air traffic control industry,the fatigue state of the controller must be found in time.Therefore,the fatigue detection strives to be real-time,accurate and non-invasive.Convolutional neural network has a good effect in face detection and image recognition.Therefore,this paper proposes to build a convolutional neural network model to detect the controller's facial features and application system design,so as to realize the detection of control fatigue according to the recognition of controller's facial fatigue features.First of all,on the basis of in-depth study of face detection algorithms,The algorithms are classified and compared with advantages and disadvantages.By comparison,it is found that the data set required by convolutional neural network is small,capable of autonomous learning,and has strong generalization ability,which is suitable for face recognition.Therefore,the algorithm of convolutional neural network: MTCNN,VGG-16 and ResNet-50 selected in this paper is proposed.Secondly,the fatigue feature acquisition experiment was designed,and the control simulator platform was used to build a data acquisition platform to collect the facial features of school controllers.Combined with the MTCNN model,facial features of controllers are identified and extracted.Then,feature images are preprocessed,and classified.The self-built data sets of controllers' eyes and mouths are constructed,and the public data sets of eyes and mouths are collected for model training and testing.After that,the VGG-16 human eye recognition model,ResNet-50 mouth recognition model and the self-built lightweight convolutional neural network model were constructed.The model was trained with the public data set and tested with the self-built data set.The test results show that the VGG-16 network model is superior to the traditional geometric positioning model,and the ResNet-50 model is more accurate than the self-built lightweight convolutional neural network model.Finally,the controller fatigue detection application system is designed and developed on the relevant framework based on Python language.The Pycharm is used as the tool and the Keras is selected as the neural network framework.The Numpy library is used for data acceleration processing.The threshold value was set according to the fatigue parameters to judge the fatigue state and provide the fatigue warning.