Research On Ground Moving Target Detection And Imaging In Multichannel Sar System

Synthetic aperture radar(SAR)has the advantage of all-weather,all-day,longdistance,and high-resolution imaging for the observation scene.It is developing in the direction of multi-platform,multi-channel,multi-function,and multi-polarization.The multichannel SAR system provides spatial information of the moving target signal and realizes the ground moving target indication(GMTI)for the observation scene,which greatly improves the moving target observation performance.Therefore,the SAR system has indisputable benefits in both military and civilian applications.However,airborne and spaceborne multichannel SAR-GMTI systems face many common problems in practical complex observation scenes.First,the practical observation scene usually contains various types of background clutter,of which scattering coefficients greatly vary.This leads to the heterogeneous clutter distribution.The multichannel space-time adaptive clutter suppression performance deteriorates and the false alarm probability increases owing to residual strong clutter.Moreover,there are usually multiple moving targets in the observation scene.The moving target will be both defocused and displaced in the SAR image owing to its motion.Nearby targets are defocused with overlapped images and sidelobe interference,and even false targets appear,which constitutes additional challenges for simultaneous imaging multiple moving targets.Given the small interference phase difference between the slow-moving target and the background clutter,the multichannel adaptive filter response of the slow-moving target is close to the clutter suppression notch.The output signal clutter noise ratio(SCNR)decreases.It is difficult to realize the slow-moving target detection and even imaging.Aiming at moving target detection and imaging problems in the practical complex observation scene,therefore,multichannel SAR ground moving target detection and imaging methods are studied by utilizing the spatial and temporal information of multichannel SAR complex-valued data as well as sparse reconstruction and deep learning theory.The main contribution of this dissertation includes the following four parts:1.A new clutter suppression method based on DPCA-BCS is proposed by utilizing the moving target sparsity prior in the dual-channel SAR system.First,the displaced phase center antenna(DPCA)technique is employed by utilizing a small amount of azimuth observation data to eliminate the most of background clutter.Then,the sparse observation model is established by utilizing the signal after DPCA.By introducing the Laplace prior distribution to the moving target,the Bayesian compressive sensing(BCS)method is exploited to realize the moving target reconstruction and the clutter suppression.A clutter suppression method based on STAP-BCS,combining the space time adaptive processing(STAP)and sparse Bayesian learning,is proposed for the multichannel SAR system.Finally,simulations and experiments show that the clutter suppression performance is improved with a small amount of observation data for proposed algorithms.2.To solve the problem of moving target detection in the heterogeneous clutter environment,a new moving target detection method is proposed based on the improved Gaussian mixture probability hypothesis density(GMPHD)filter for the multichannel SAR system.First,the SAR multi-aspect image sequence is generated based on the sub-aperture mode.And the moving target observation information is extracted by utilizing the clutter suppression and the preliminary constant false alarm rate detector.Then,the multiple-target dynamical and observation random finite set models are established by analyzing effects of the target radial velocity on the target position.Aiming at the problem of the traditional GMPHD filter in the SAR-GMTI system,an improved GMPHD filter is proposed for the application of the SAR image sequence.Finally,simulations and experiments show that the false alarm probability is decreased while a high detection probability is achieved by the proposed algorithm in the heterogeneous clutter environment.Moreover,the target relocation is realized by the proposed method.3.To solve the problem of SAR multiple moving target imaging,a new imaging method of multiple moving targets based on Chirplet-BCS is proposed by utilizing the multi-component linear frequency modulation signal form and the sparsity prior of the moving target.First,the multiple target sparse observation model is established.The adaptive decomposition based on Chirplet basis is employed to estimate target Doppler parameters for the observation matrix construction,which effectively avoids the interference of cross terms.Then BCS sparse reconstruction algorithm is applied to reconstruct the moving target image.Finally,simulations and experiments show that the proposed algorithm achieves excellent imaging quality and residual clutter suppression performance.4.This dissertation introduces the deep learning theory into the field of SAR moving target imaging and studies the multiple slow-moving target fast imaging method based on deep convolutional neural network for the multichannel SAR system.To solve the problem of SAR multiple moving target fast imaging,a convolutional neural network-based method is proposed for SAR multiple moving target imaging.The proposed imaging network Deep Imaging utilizes residual learning strategies to achieve effective feature and gradient transfer.The network parameters are updated by supervised learning with a large number of moving target training samples.Finally,the trained network is able to establish an imaging model suitable for the observation scene.The proposed method realizes the fast imaging of moving targets.The network Deep Imaging relies on the multichannel clutter suppression result,which makes it difficult for slow-moving target detection and imaging.Aiming this problem,the clutter suppression task is further integrated into the deep network.A complex-valued convolutional neural network-based method is proposed for the multichannel SAR multiple slow-moving target imaging.The proposed complex-valued imaging network CV-GMTINet extends both feature maps and network parameters into the complex domain.The proposed network not only takes the complex signals as the network input but also propagates the phase information throughout the network.Combining advantages of the dense network and residual network,the main module adaptively learns effective features of single channel and between channels.The propagation efficiency of features and gradients is improved,while the gradient vanishing problem is alleviated.The complex backpropagation algorithm is utilized to solve the gradient of network parameters,which are updated by the gradient-based optimization algorithm.The effectiveness of the proposed method is verified through experiments.Compared with the traditional method and the real-valued network,this proposed method achieves significant improvements in both moving target imaging accuracy and clutter suppression performance.