Research On Airport Visibility Measurement Based On Machine Vision And Deep Learning

With the rapid development of China's aviation industry,the demand for the safe take-off and landing of aircraft and the orderly operation of airports continues to increase.The objective and accurate measurement of airport visibility can provide important data support for it,so more and more attention has been paid to the measurement of airport visibility.As an automated alternative input to the human eye,machine vision can be an important part of the automated determination of airport visibility.Aiming at the shortcomings in recent years,such as the lack of feature information extraction for airport visibility measurement algorithms based on machine vision and the unutilized temporal information of video,this thesis combines the relevant knowledge of deep learning and proposes a temporal and spatial method that can extract visibility images.Information on the airport visibility determination algorithm.Based on the deep learning algorithm in the field of machine vision,this thesis studies the algorithm model of airport visibility measurement based on VGG network.Then,the preprocessing process of the airport visibility-related data used in this paper is introduced,and conducts experiments.And by introducing transfer learning training,compare with the model effect when it is not introduced.It can be found that the model based on transfer learning has better generalization ability and recall rate than the measurement model without introduction,and the convergence speed has been greatly improved.The principle and advantages and disadvantages of the airport visibility measurement algorithm based on the VGG network are analyzed,and the variant gated recurrent unit GRU of the recurrent neural network is introduced.Combined with the Kalman filter,the GRU-KF model is proposed: it can be used to extract the time dimension information.At the same time,the airport visibility value is corrected.On the basis of the previous algorithm,an improved algorithm model of VGG-GRU-KF is constructed.According to whether the weight of the GRU network itself is trained and updated in the process of time step iteration,two kinds of experiments are designed,and the initial value of the GRU-KF model is preset for comparative experiments.Then compare the experimental results with the effect of the model before improvement.Experiments show that the effect of the airport visibility measurement algorithm based on VGG-GRU-KF proposed in this thesis is significantly improved compared with the airport visibility measurement algorithm based on VGG network,and in the process of time step iteration,after the GRU-KF model The overall evaluation indicators such as the recall rate of the model that update the weight of the GRU network by the estimated correction value are better than the model that freezes the weight of the GRU network.Finally,this thesis proposes an airport visibility regression model based on PV-CNN.Combined with the convolutional neural network,the corresponding visibility value is regressed through the probability vector.Experiments show that the model can effectively regress the airport visibility value.