Study On Multi-Dimensional Collaborative Radar Detection Of Moving Targets

Ground moving target indication(GMTI)is one of important missions for a radar in the battlefield environment perception.Recently,with the improvement of radar resolution,synthetic aperture radar(SAR)has become a major means of high-resolution observation in situations when optical and infrared sensors fail due to a degraded visual environment.Therefore,GMTI combined with SAR imaging has widely been interested by researchers domestically and globally.However,GMTI or SAR-GMTI based on a single-platform measurement in a single coherent processing interval(CPI)is difficult to achieve reliable detection due to limited information.Therefore,more detection degrees of freedom are desired for a robust moving target detection.At present,mining the temporal characteristics of targets is an important research in radar moving target detection.The detection performance can be improved by jointly processing multi-frame measurements formed from multiple CPIs.As for multi-frame measurement,video SAR can present the dynamic information of the scene in the form of optical-like video and provide sequential high-resolution SAR images.On the other hand,pulse Doppler(PD)radar can provide sequential low-resolution range Doppler(RD)spectra in the staring or tracking mode.In addition,the reliable detection can also be realized by jointly processing of multiple features of the same target in different domains.Furthermore,multiple radars can collaboratively observe the same area,which provides electromagnetic scattering information of the same target from multiple angles of view or frequency bands.These information can be utilized to make up for the lack of single-platform detection Focusing on multi-dimensional collaborative radar detection of moving targets,this dissertation studies the moving target detection in high-resolution video SAR and lowresolution PD radar from three prospects in multi-frame,multi-domain,and multi-platform.The main accomplishments and contributions are summarized as follows:1.The acquisitions of sequential low-resolution RD spectra and high-resolution SAR images are given,and the imaging resolution,frame rate,shadow formation mechanism and moving target shadow characteristic are introduced in the video SAR.Subsequently,the basic principles of multi-frame,multi-domain and multi-platform joint detections are analyzed,respectively.2.Three multi-frame detection algorithms are proposed for moving targets based on shadows in sequential video SAR images and Doppler energies in sequential RD spectra,respectively.Firstly,a video SAR moving target shadow detection algorithm based on deep neural network is proposed,which uses single-frame deep detection combined with inter-frame processing of false alarm and missing alarm to overcome the problems caused by traditional methods.This approach verifies the feasibility of deep learning in video SAR shadow detection.Secondly,a faster track-before-detect(TBD)approach for video SAR shadow detection is proposed to solve the problem of large computational load in the traditional implementation.A large number of invalid and redundant states are eliminated through threestep state filtering,and the subsequent dynamic programming is realized only based on sparse states.Therefore,the proposed approach shows good detection performance in shadow detection with low computational load.Finally,a deep multi-frame detection network for moving target energy is proposed based on sequential RD spectra.On the one hand,the differences between moving target and clutter are used to design a multi-scale intra-frame detection network.On the other hand,an inter-frame detection network is designed by using the inter-frame correlation differences between moving target and false alarm.Through two-step detection,the robust detection can be obtained under low signalto-clutter ratio.3.Two dual-domain joint detection algorithms are proposed to form a more comprehensive cognition of moving targets in video SAR based on sequential high-resolution SAR images and low-resolution RD spectra.Firstly,a dual-domain faster region-based convolutional neural network is proposed,which simultaneously detects the shadow and energy of a moving target based on the shared dual-domain region proposals.The dual-domain joint decision enables the network to achieve reliable detection with less false alarms.Secondly,a dual-domain dynamic programming-based TBD algorithm for video SAR moving target detection is proposed.This method combines the multi-frame and multi-domain detections,utilizing the Doppler frequencies of a moving target to provide velocity guidances for its shadow tracking.Through dual-domain state filtering and dual-domain integration of value function,the problems of selecting the number of valid transition states and difficult tracking of maneuvering target caused by the traditional TBD algorithm are alleviated.The proposed two dual-domain joint detection algorithms combine the complementary features of shadow and energy,which provides a new train of thought for moving target detection.4.The study on collaborative detection of moving targets in distributed video SAR is aiming at the single-platform detection problems and a multi-view detection and trajectory fusion algorithm is proposed in this dissertation.The proposed approach exploits the advantages of complementary observations by combining positions and radial velocities of the same moving target obtained from different angles of view simultaneously.It not only compensates for the disadvantages of occlusion or poor contrast of moving target shadows during single-view detection,but also has the ability to solve the two-dimensional full velocities of the target,and thus the detection performance of the distributed video SAR is improved,especially in urban,hilly and other complex terrain areas.The series of moving target detection algorithms proposed in this dissertation have been verified by multiple sets of simulated and measured radar data.The detection results indicated that rich detection degrees of freedom can achieve a robust moving target detection.