Study On Target Feature Extraction Based On Radar Image

Radar imaging technique has the advantages of all-weather and day/night for long range observation,which provides more structure information of targets leading to its broad application in military and civilian fields.With the rapid development of technologies such as semiconductors,computers,and signal processing,the performance of radar imaging has been improved a lot,with better the image quality and resolution,which is able to provide more detailed information about targets.Faced with a large amount of radar image data,how to extract target information quickly,accurately and effectively,and classify the target has become a bottleneck problem of radar in the real application.The traditional feature extraction methods based on model matching depend on the prior information of the target in the image,such as contour features,texture features,edge features.However,those features will change with the actual environment,and also requires appropriate training data,which limits the application of those methods.The target attributed information is usually the reflection of some physical characteristics of the target,such as the micro-motion,structural and polarimetric characteristics.The feature extraction is less affected by environmental factors,so it is of great value both in theoretical research and engineering.For exploiting the physical mechanism of target electromagnetic scattering from radar images,the dissertation discusses three aspects: micro-motion feature extraction,structure feature extraction and attributed scattering center feature extraction.The main work can be summarized as follows:1.Aiming at the problem of the classification for aircraft targets under low SNR,a Jet Engine Modulation(JEM)feature extraction algorithm based on time-domain flashing period for aircraft targets is proposed.The differences of the time domain modulation period and the time-frequency domain waveform entropy are used to classify aircraft.Firstly,the window function is used to intercept the signal and the characteristics of time domain and Doppler energy entropy with window sliding are are explored and captured.Then,the length of the window function is optimized for the better classification performance.Finally,the simulation results and the measured results verify the validity of the proposed algorithm for effectively classifying the aircraft targets under low SNR.2.Aiming at the structural feature extraction for spatial orbit target,a three-dimensional imaging algorithm based on multistatic ISAR images is proposed.The algorithm introduces the three-dimensional ISAR turntable model.Then the three-dimensional turntable model is used to obtain the projection equation between the three-dimensional coordinates of spatial orbit target and the two-dimensional ISAR image.The projection equation consists of a projection matrix and a measurement matrix of the scattering center.It is noted that the measurement matrix is composed of the longitudinal distance and the micro-Doppler frequency of the scattering center in different images.The advantage is that there is no need to perform the cross-range scaling of ISAR image.For constructing the projection matrix and the measurement matrix of scattering center,the algorithm is divided into two parts: the first part establishes the relative motion model between the spatial orbit target and the multistatic radar.Then,the projection matrix can be constructed.The second part first proposes a scattering center extraction method based on watershed-spectrum estimation to extract the longitudinal distance and the micro-Doppler frequency of scattering center.Secondly,for combining the scattering information of scattering centers in different ISAR images,a scattering center association method based on the epipolar geometry theory is proposed to construct the measurement matrix of the scattering center.Finally,the simulation results verify the validity of the proposed algorithm for scattering center extraction,scattering center association and three-dimensional imaging.3.Aiming at feature extraction for cone target under short-dwell observation time,a parameter estimation algorithm based on sequential SAR images for cone target is proposed.The algorithm uses the image rotation correlation method to estimate the rotation angle velocity of cone target under the turntable model.Then,by using on the micro-motion model of the three scattering centers on the conic point and bottom edge of cone target,the angle between the target symmetry axis and the radar line of sight is estimated.On this basis,the longitudinal distance and micro-Doppler frequency of scattering center in the half precession period can be used to establish the micro-motion characteristic equations.Then,the structural parameters and the micro-motion parameters of the cone target can be obtained by solving the equations.Finally,the simulation results verify the validity of the proposed algorithm for micro-motion feature and structural feature extraction of cone target with short-dwell observation time.4.Aiming at three-dimensional fully polarimetric attributed scattering center extraction,a three-dimensional fully polarimetric attributed scattering center extraction algorithm based on sequential SAR images is proposed.The algorithm makes use of the sparsity of radar data in the frequency-azimuth-elevation parametric model domain.Firstly,for combining the electromagnetic scattering information of the attributed scattering center in the sequential SAR image,a fully polarimetric attributed scattering center association method is proposed.Secondly,due to the large number of model parameters,the dimension of parameterized dictionary is very large.To address that issue,the positional information of the scattering center in the SAR image sequence is used to initialize the parameters,and range-azimuth characteristic decoupling is performed for reducing the dictionary dimension.Thirdly,the parameters of three-dimensional fully polarimetric attributed scattering center model can be estimated by solving the sparse signal recovery problem.According to the extracted threedimensional fully polarimetric attribute scattering center,the three-dimensional spatial position,geometric size,attitude angle and polarimetric information of the target or important components can be effectively estimated.Finally,the effectiveness of the above method based on electromagnetic simulation data is verified.