Research On Airborne Forward-looking Super-Resolution Imaging Method With Bayesian Framework

Airborne real-beam scanning radar forward-looking high-resolution imaging has all-day and all-weather rapid target detection capabilities and is widely used in autonomous navigation,terrain tracking,and terrain following.High-resolution imaging mainly relies on utilizing Doppler information or improving the antenna aperture size of real-beam radar.When the radar works on the forward-looking model,the Doppler bandwidth is very small or even zero,and the imaging methods which rely on Doppler information are invalid.Due to the limitation of the airborne platform,the antenna aperture size cannot be increased without limitation,and the imaging methods which rely on the antenna aperture is difficult to improve the azimuth resolution.Super-resolution method can obtain resolution improvement beyond the Rayleigh limit without changing hardware conditions which can be widely used in real-beam scanning radar imaging.The statistical optimization method based on Bayesian theory is an effective way to achieve super-resolution imaging.The core of the statistical optimization method is to find appropriate prior information according to different forward-looking scenes.Therefore,according to different forward-looking imaging requirements,this thesis deeply studies the forward-looking super-resolution imaging of airborne real-beam scanning radar based on the Bayesian framework.The main innovations can be summarized as follows:1.This thesis studies the low azimuth resolution of the forward-looking target in complex scene,and proposes a Bayesian super-resolution method based on Gaussian mixture modelLaplace hierarchical prior(GMM-LP).First,considering the clutter is a combination with various simple distributions,the Gaussian mixture model(GMM)is introduced as the statistical model of the clutter and noise.Second,the Laplace hierarchical prior model is presented to express the sparsity property of targets.Then,the expectation maximization-maximum a posterior(EM-MAP)method is used to estimate the parameters.Simulation results and semi-real data show that the proposed method can not only suppress imaging clutter but also improve the azimuth resolution of forward-looking targets.2.This thesis studies the scene where target edge information is required in applications such as autonomous landing,and proposes a Bayesian super-resolution method based on Gaussian mixture model-multiple prior(GMM-MP)for the target in complex scene.First,GMM is introduced to model the complex clutter and noise.Secondly,a multi-prior distribution is proposed to express the target via fusing the Laplace prior and the total variation(TV)prior which not only describes the sparsity of the target but also preserves the contour information.Finally,several experiments prove that the proposed method obviously suppresses the clutter,restores the contour information of the target,and enhances the azimuth resolution.3.This thesis studies forward-looking imaging in the presence of outliers in practical applications,and proposes a Variational Bayesian(VB)super-resolution method based on Student-t distribution.First,since the radar may be affected by unintentional electromagnetic interference or equipment performance anomaly,the echo signal consists of outliers.To suppress the outliers,the Student-t distribution is introduced to model noise.Second,to reconstruct the sparse target,the Laplace hierarchical distribution is introduced,which is more flexible to model the target.Then,the VB technique is utilized to estimate the imaging parameters based on the Bayesian framework.Finally,simulation results and semi-real data show that the proposed method has a better super-resolution performance in the presence of outliers.4.This thesis studies the high-speed airborne forward-looking real-time imaging,and proposes a low-dimensional sparse Bayesian learning with Doppler compensation(LDSBL-DC)method for airborne forward-looking imaging,in which the forward-looking image can be quickly obtained with a high azimuth resolution.First,the Doppler frequency information is introduced to construct the Doppler convolution model in high-speed airborne.Second,to avoid calculating the Doppler convolution matrix in each range cell,the Doppler convolution matrix is compensated and the construction time of the matrix is reduced.Third,a low-dimensional projection model based on the singular value decomposition is proposed.In the low-dimensional projection model,the high-dimensional echo data is compressed to lowdimensional data.Then,sparse Bayesian learning(SBL)is utilized to estimate the imaging parameters.In the estimation of the targets’ scattering coefficient,the matrix transformation further reduces the computational complexity.Finally,simulation results show that the LDSBL-DC method can improve the azimuth resolution with low computational complexity.