Research On Airport Visibility Detection Algorithm For Aviation Safety

Low visibility caused by smog is an important hidden danger to aviation safety.Real-time and effective haze visibility detection is one of the important ways to alleviate this problem.According to the regulations of the aviation management department,when the visibility is lower than 550 meters,low-visibility operation management must be carried out.In view of the limitations of existing detection methods,this thesis constructs a multi-modal haze visibility detection algorithm from the perspectives of video images and key meteorological factors,and verifies it with relevant measured data at Nanjing Lukou International Airport.The main work of this thesis is as follows:(1)Construct a multimodal airport visibility detection data set(Vi SOF: Visibility detection data Set of airp Ort Fog and haze).The dataset covers 19,468 airport haze pictures and 97,340 key meteorological factor data,such as air relative humidity,air flow rate,air temperature,and rainfall.The true value of the relevant visibility is obtained by an optical visibility meter.This thesis collects continuous data in 2 scenarios in total,with a duration of 5 days.On this basis,a lot of data preprocessing work is carried out in this thesis to establish a one-to-one correspondence between "key meteorological factors,haze images" and visibility to form a data set.(2)Single-modal visibility detection algorithm based on improved Swin-Transformer.Deep learning networks can extract effective features from haze images that contain visibility information.Aiming at this characteristic,this thesis sharpens the haze image,builds a Double SwinTransformer network framework,and validates it with the measured data of Nanjing Lukou International Airport.The data results show the effectiveness of the algorithm proposed in this thesis.(3)Multimodal visibility detection algorithm based on deep learning.Based on the research of haze visibility in video images,this thesis integrates key meteorological factors and constructs a multi-modal visibility detection algorithm to improve the accuracy of visibility detection.The deep convolutional network is used to extract the visibility features in the haze pictures,and the BP neural network is used to extract the visibility features of the key meteorological factors,and the two features are fused to train the detection model.The verification results show that due to the correlation between meteorological factors and haze,the detection algorithm that integrates haze images and key meteorological factors is superior to the single-modality detection algorithm in terms of detection accuracy.(4)Improve the multi-modal visibility detection algorithm(AST)of the transformer.The research of the above two algorithms shows that the Swin-Transformer network can effectively extract haze features,and the multi-modal detection method can effectively improve the detection accuracy.On this basis,this thesis proposes an Assimilate Swin-Transformer(AST)network structure.Mass data verify the effectiveness and robustness of the algorithm.