Research On Track Recognition And Clustering Based On Radar Data

In the field of aerospace,the radar finds target objects in the air by radio and determines their spatial position.The radar plays an irreplaceable role in battlefield,civil aviation and other environments.With the increasing number of equipment using frequency resources in the air and the increasingly complex air environment,the research on the historical track data of aerial target objects is conducive to the deep understanding of the air traffic flow organization mode and to grasp the changes of air traffic characteristics and the spatial and temporal distribution characteristics.In this thesis,the target track trajectory recognition and clustering studies are conducted bases on the historical track data.The main research contents are as follows:(1)Radar target track recognition based on convolutional neural network.Currently,in the field of radar target recognition,the target characteristic information such as amplitude,phase,frequency spectrum and polarization in radar echo is usually used to estimate the size,shape,weight and physical property parameters of the surface layer through multi-dimensional space transformation,etc.,and to identify target types and models.In this thesis,using the spatio-temporal attribute data of the target object provided by the radar equipment,such as time,azimuth,distance,etc.,we extract the slope value(k value)between the track points and analyze the effectiveness of k value with Shapely.We then construct a convolutional neural network model to achieve the recognition and detection of radar target track types.Finally,based on the target track data generated by the simulation,we test the constructed model.The experimental results show that the model can improve the performance of detecting and identifying the target track.(2)Deep track clustering based on spatiotemporal distance and denoising autoencoder.Existing track clustering studies ignore the track space-time characteristics when designing track similarity measurement,and the track clustering algorithms does not consider the timing features of the tracks.In this thesis,we firstly propose the SIM-T(Similarity of Track)metric for trajectory similarity measurement based on the spacetime characteristics of the trajectory,and then propose the deep temporal clustering based on denoising autoencoder(DAE-DTC)for trajectory clustering.SIM-T comprehensively uses the space-time characteristics between track points to measure the similarity of tracks.DAE-DTC model firstly uses the deep denoising autoencoder network(DTDA)to effectively extract the potential representation of the track sequence,and then uses time a clustering layer(TCL)to cluster the tracks based on the idea of k-means.ADS-B track data are used to verify the performance of the proposed track similarity measurement method SIM-T and track clustering algorithm DAE-DTC.Experimental results show that the proposed SIM-T metric and DAE-DTC clustering algorithm can effectively improve the performance of track clustering.