Online Knowledge-aided Refined Clutter Suppression Method Research Under Complex Geographical Scene

Ground/ marine moving target indication(GMTI/MMTI)technology is one of the main tasks of the airborne/spaceborne early warning and surveillance system.With the synthetic aperture radar(SAR)system,while ensuring high-resolution imaging,it can detect and locate the ground/sea moving targets,which has great value and significance for many tasks,such as early warning and surveillance,combat perception,traffic planning,ship detection and so on.In the complex geographical scene(sea-land,suburban,complex urban,and et al.),it will be difficult to obtain the independent and identically distributed(IID)clutter training samples because of the spatially inhomogeneous distribution clutter.This will worsen clutter suppression and slow-target detection performance.With the development of technology,radar systems are increasingly equipped with multi-dimensional observation capabilities.Therefore,this dissertation synthesizes the multi-dimensional domain information(space,time,frequency band,and polarization)to obtain the online knowledge about the clutter type and spatial distribution in complex geographical scenes.Then,we select the IID training sample to estimate the background clutter covariance matrix(CCM)of the corresponding clutter and perform clutter suppression,thereby improving the clutter suppression ability and target detection performance for multichannel system.The main works can be summarized as follows:1.To deal with the positioning and suppression of azimuth ambiguity in inshore region,an azimuth ambiguity clutter suppression method is proposed based on the feature extraction with tensor discriminant alternating optimization(TDAO)and clutter classification with the affine invariant Riemannian(AIR)distance.Firstly,the correlation coefficients between channels are obtained to characterize the normalized SAR image amplitude and compress the amplitude dynamic range.The coherence coefficient gradients and the multi-look interference phase gradients are used to describe the local texture characteristics of the image.In order to maintain the structural characteristics of the three-dimensional data and take the neighborhood information of pixels into consideration,we exploit tensors to describe the extracted multiple features.Then,under the constraints of similarity and difference of samples,the core feature tensors are extracted by the proposed TDAO method,which are conducive to the further classification.In this processing,the data reduction can be reduced while retaining the core characteristics to determine the clutter categories,which can not only reduce the computational complexity of subsequent classification but also remove unfavorable factors affecting the classification.Next,the AIR distance between the feature covariance matrices(which are expanded along the feature dimension of the core feature tensor)is used to classify the clutter in the inshore region to position the azimuth ambiguity.Finally,the IID clutter training samples of the ambiguity clutter area are selected to accurately estimate its CCM and the fine ambiguity clutter suppression is completed.The simulation data and the measured data of Terra SAR-X satellite show that the proposed method does not require prior knowledge(ambiguity source location,system parameters,et al.)and can achieve the accurate positioning and suppression of azimuth ambiguity clutter in inshore region.2.To address the problem that adaptive clutter suppression and target detection performance are degraded because of the inhomogeneous clutter in suburban,with the range-Doppler-band SAR data from the high-resolution SAR imaging and GMTI-compatible system,a clutter suppression method assisted by multi-band adaptive weight penalty MRF(AWP-MRF)area division is proposed.This method firstly uses the spatial distribution consistency of clutter between multi-band SAR images,and comprehensively considers the spatial distance between pixels,fisher distance,local roughness distance and gradient distance in the range-Doppler-band three-dimensional neighborhood system,which can characterize the force between neighborhood pixels from different aspects.Further,the adaptive weighted penalty function is constructed.The priori probability of pixels is calculated by the K-Wishart distribution model of the multi-look covariance matrix in the frequency band of the center pixel.And,an adaptive weighted penalty terms is further constructed,which can smooth the homogeneous regions while maintaining the edge of image.Meanwhile,the priori probability of the pixel is calculated using the K-Wishart distribution model of the multi-look covariance matrix in the band where the center pixel is located,which can make full use of the amplitude and phase information of the multichannel SAR images.Then,In Bayesian framework,AWP-MRF classification of multi-band SAR images is performed and multiple independent homogeneous or inhomogeneous clutter regions are extracted by morphological operations.Finally,with the help of the obtained region location,different sample selection methods are used to estimate the background CCM for different clutter regions,and the clutter suppression are completed.The experimental results based on the measured data show that the proposed method does not require prior knowledge and can avoid the problem of insufficient universality of the traditional sample selection method,reduce the false alarm of the inhomogeneous region and improve the detection performance of the slow-moving target in the homogeneous region.3.To avoid the degradation of clutter suppression and GMTI performance caused by the difficulty to select the IID training samples in complex urban,with the multichannel and full-polarimetric system,we propose a clutter suppression method assisted by the hybrid weighted local K-means(HWLKM)polarization classification.Firstly,a multi-channel weighted estimation method is proposed to avoid the over-smoothing problem and improve the estimation accuracy of the polarization coherence matrix.Then,the region covariance matrix(RCM)is used to extract the polarimetric features from the scattering mechanisms and human vision,including the target decomposition features(based on eigenvalue decomposition and based on the scattering model)and the spatial structure features(spatial texture,color).In order to maintain the metric invariance when RCMs have the variances of scale and rotation,the geometric distance(GD)based on Riemann space is used to measure the distance between RCMs.In the K-means framework,considering the similarity between similar samples and the difference between heterogeneous samples at different scales,the local hybrid weights are constructed and the cluster centers are updated.Further,the GD-HWLKM polarization classification is performed.Then,with the help of the acquired online knowledge,the background CCM of each land cover is estimated by using clutter samples with the same scattering and distribution characteristics,and various types of clutter are suppressed in a refined way.Further,the multichannel clutter suppression and moving target detection are performed.In addition,the non-coherence integration detection(NCID)technology is used to further eliminate the false alarms.Finally,the experimental results of the measured data show that the proposed method can significantly improve the clutter suppression ability of strong clutter,significantly reduce the false alarms,reduce the output signal noise loss and improve the detection performance of slow moving targets.