Research And Implementation Of Aerial Surveillance Information Fusion Technology Based On Deep Learning And Kalman Filtering Algorithm

With the development of the world economy,the number of people who choose to travel by plane has increased year by year,leading to an increase in the number of flights.The flight safety issues exposed by the MH370 crash and disappearance in 2014 have also attracted the attention of the government.Therefore,improving the ability to monitor air targets has become a more urgent need.As an important part of aviation surveillance system,aviation surveillance information fusion is the focus of research in this field.It fuses the track data obtained from multiple radars tracking the same flying target to form a more accurate track.Kalman filtering algorithm is a traditional aviation surveillance information fusion algorithm.Affected by its linear model and its fixed parameters,the Kalman filter has the problem of poor adaptive ability to dynamic environments.In recent years,deep neural networks have developed rapidly,and they have specific advantages in learning unknown things.In this thesis,two new single radar surveillance information filtering enhancement algorithms are proposed,by studying the characteristics of Kalman filtering and a variety of commonly used neural networks,combining Kalman filter and neural networks.Combine the back propagation neural network,the generalized regression neural network and the Kalman filter respectively,and use the Kalman filter as the neuron of the recurrent neural network to obtain two algorithms.On the basis of these two algorithms,a real-time multi-radar fusion algorithm with estimated error as the weight is also proposed.Finally,based on these three algorithms,a distributed aviation surveillance information fusion system is designed and implemented.The system implements functions like surveillance data preprocessing,single radar filtering and enhancement,multi-radar fusion,and flight monitoring.The experimental results show that the single radar surveillance data enhancement algorithm proposed in this thesis has better accuracy than the traditional linear Kalman filter algorithm in most cases.Multi radar surveillance information fusion algorithm further reduces the error of single radar enhancement algorithm and reduces data fluctuation.Finally,the average error of radar raw data is reduced from kilometer level to about 100 meters,which proves the effectiveness of the algorithm.