Research On Plot-to-Track And Track-to-Track Association Method Based On JVC And CNN For Compact HFSWR

Vertically polarized high-frequency electromagnetic waves(3 ～ 30MHz)are characterized by low attenuation in their diffraction propagation over the sea surface,which enables HighFrequency Surface Wave Radar(HFSWR)to detect targets beyond the line of sight.The maximum detection range of HFSWR can reach 200 nautical miles.To achieve accurate target azimuth measurement,most HFSWR used for target detection adopts a large array receiving antenna,the high cost of whose siting,deployment,and maintenance give rise to practical limitations for its promotion and application.On the circumstances,developing a Compact HFSWR system and corresponding target detection algorithms has become a promising direction for further research.However,the reduced receiving antenna array and transmitting power impact negatively on the estimation accuracy of the target azimuth.As a result,the subsequent target tracking tends to suffer from problems of track fragmenting and target deviating.To improve the continuity and accuracy of target tracking,this thesis studies the plotto-track association and tracklet association for monostatic Compact HFSWR in multi-target tracking scene,as well as the track-to-track association of T/R-R bistatic Compact HFSWR.The main work of the thesis include:Firstly,the sequential plot-to-track association is prone to lead to the false association and false tracking in the algorithm of multi-target tracking for monostatic compact surface wave radar.As a result,this thesis proposes a multi-target plot-to-track association method based on the JVC(Jonker-Volgenant-Castanon)algorithm.For multiple tracks of common candidate plots of their overlapped association gate,the minimum cost function is used to determine the similarity between common candidate plots and all tracks,and the association cost matrix is obtained.Taking the minimization of the total association cost as the optimization criterion,the JVC algorithm is used to obtain the optimal plot-to-track association result.The experimental results show that the method can effectively improve the continuity of target tracking.Secondly,monostatic Compact HFSWR tends to suffer from problems of track fragmentation and tracking discontinuity due to low detection probability and serious clutter interference.As a result,this thesis proposes a tracklet multi-level association method based on bidirectional prediction.The coarse association for tracklets is completed by using target track features,and the association range is further reduced by using space-time constraints.Utilizing bidirectional prediction and the JVC algorithm,the optimal matching between tracklets is achieved.The experimental results show that compared with the existing methods,the association accuracy of the proposed method is significantly improved,and the continuity of target tracking is improved.Finally,aiming at the problem that the track features of monostatic Compact HFSWR are few and difficult to extract,a track-to-track association method based on CNN(Convolutional Neural Network)for T/R-R bistatic Compact HFSWR is proposed.The CNN model is used to extract the track features automatically,and the track-to-track association for multiple targets is completed based on the track features.The experimental results show that compared with traditional methods,the proposed method can save the process of manual feature extraction and effectively improve the accuracy of track-to-track association,which lays a foundation for track-to-track fusion to improve the target tracking accuracy.