Deep Learning-based Image Segmentation On General Aviation Emergency Rescue

As an important part of emergency rescue,general aviation emergency rescue has the characteristics of fast response speed and strong on-site interaction ability,which can provide an effective way to reduce the loss of disaster events.Before arriving at the disaster site,obtaining and analyzing relevant information on the path(such as bird strike problem and dangerous weather impact)can provide an important guarantee for the efficient and orderly progress of the whole emergency rescue work.At present,general aviation still relies on pilots’ own observation of the environment to understand the activities of birds,and the way of scare the birds methods at home and abroad are not suitable for widespread use;and its understanding of dangerous weather is mainly based on real-time weather conditions.Although the current research on weather has turned to dynamic prediction,the accuracy of prediction methods also needs to be improved.Focusing on the above problems,this thesis uses image segmentation method based on deep learning to improve the safety of general aviation emergency rescue.The main work is carried out from the following aspects:(1)An optimized U-Net network model is proposed.Adaptive selection structure is added to the connection structure of the network to enhance the attention of the network,so as to improve the accuracy of image segmentation of the network.(2)According to the application needs,we establish our own bird movement target sample set and dangerous weather target sample set.Both of the sample size are 512×512,and marked them respectively.(3)The optimization of U-Net applied in the flock of image segmentation,provide warning for the pilot,alert the pilot to the flock.Reduce the pressure of the pilot work,improve the safety of the general aviation emergency rescue by the economic convenient way.(4)For the dangerous weather affecting flight safety,this thesis uses the image segmentation method based on optimized U-Net to segment,combined with the residual Grey Markov prediction model to predict the change of its influence range,which can provide reference for the aircraft diversion strategy and ensure the safety of navigation rescue path.In this thesis,the optimized network model is applied to the image segmentation of birds.Through experiments,the commonly used image segmentation evaluation indexes m Io U and F1 score are 0.8462 and 0.8172 respectively,which are 3.6% and 4.49% higher than that of UNet;The dangerous weather range is segmented through the optimization network,and the corresponding m Io U and F1 score are 0.8861 and 0.8725.Finally,combining the image and data,use the residual Grey Markov prediction model,we can make a short-term accurate prediction of the change of the influence range of dangerous weather.