Deep Learning Based SMR And ADS-B Data Fusion For Airport Surface Surveillance

The measurement of aircraft operation trajectories on airport surfaces is a crucial technology for ensuring the safety of airport operations.Currently,the main methods for monitoring surface operations are Automatic Dependent Surveillance-Broadcast(ADS-B)and Surface Movement Radar(SMR),each of which has its own advantages.To ensure the accuracy of aircraft trajectory measurement,data fusion of the two monitoring methods is usually necessary.The most common method used for monitoring system data fusion is based on the Kalman filter principle.However,neural network methods trained based on deep learning have been successfully applied in many fields,and research on information fusion using deep learning techniques can explore the internal features of data and improve prediction accuracy.This thesis studies the precision problem of monitoring the operating trajectory of targets on airport surfaces,based on the mature Kalman filter fusion results.A neural network model adapted to non-uniform data conditions is designed according to the characteristics of ADS-B and SMR monitoring data.Compared with the traditional Kalman filter fusion method,this model improves the accuracy of data fusion.Firstly,the raw monitoring data of SMR and ADS-B were parsed according to the ASTERIX Category 010 and 021 protocols,respectively,and then time alignment and coordinate transformation operations were performed for subsequent data processing to facilitate research.Secondly,in order to validate the fusion performance of deep learning methods,the Kalman filter algorithm was initially employed to fuse the ADS-B and SMR data at the same time instance.The model is validated using actual data from an airport in Sichuan province,and the experimental results are used as a benchmark for subsequent research.Finally,to achieve better fusion performance,a neural network model that is invariant in space and time is constructed based on deep learning according to the characteristics of unbalanced data.The model is based on a Convolutional Autoencoder(CAE)and a spatio-temporal attention module for predicting aircraft trajectory data.The ADS-B and SMR data are normalized using Min Max Scaler,and the unsupervised learning of the CAE model is implemented using Py Torch to extract features.Pre-trained CAE models and spatio-temporal attention modules are used for supervised learning,and the model parameters are optimized through iterative training.Finally,the performance of the model is evaluated using test data.Through experimental analysis,the deep learning model designed in this thesis achieves higher fusion accuracy for ADS-B and SMR monitoring data than the improved Kalman filter fusion method.