Research On Change Detection In Polarimetric SAR Images

With the appearance of Polarimetric Synthetic Aperture Radar(PolSAR),complete polarization measurement of terrain targets has been realized by utilizing the superiority of Synthetic Aperture Radar(SAR),which has provided considerable data for change detection techniques.In this thesis,we focus on the issue of PolSAR-based change detection techniques,quantificationally assess the performance of classical SAR change detectors and make researches on how to effectively utilize the scattering mechanism of ground objects in PolSAR image change detection.The major work can be summarized as follows:(1)Traditional evaluation methods of SAR change detectors mainly assess the performance of detectors by applying some specific scenes and areas,conclusions drew by these methods are too depend on the data and lack of effective demonstrations to summarize the merits of detectors to a generalized extent.Aiming at above issues,this thesis uses the Moving and Stationary Target Acquisition and Recognition(MSTAR)dataset acquired at HH polarization mode as experimental subject to objectively evaluate classical change detectors based on high-capacity data.The residual dataset are generated between the test sample and the training sample set by utilizing the classical change detectors in the first place.Then,the residual data with the minimal difference is taken as the optimum type of the test sample.After that,the comparison is implemented between the true type of the test sample and the optimum type.Finally,the performance of each change detector can be quantificationally assessed through the target recognition.Experimental results indicate that the proposed method can avoid the procedure of threshold selection,hence it can objectively and effectively evaluate the detectors.(2)Utilizing the scattering mechanism of different ground objects,a novel PolSAR change detection method based on three-component decomposition is proposed.Firstly,the mechanisim of surface scattering,double-bounce scattering,and volume scattering are used to modeling the terrain targets so that the scattering centers can be divided into different scattering mechanism.Then,the elements in the covariance matrices are filtered by the refined lee filter,which achieves the coarse detection of land features.Next,the complex wishart distance detector is applied to measure the “changed distance” between the similar ground objects.At last,the changed mask can be obtained through the Manual Trial-and-Error Procedure(MTEP).The proposed method embodies the principle of pre-classification and later detection,which can commendably remove the false alarm and omitted alarm.In addition,the complex wishart distance detector can get rid of the restraint of the statistical characteristic of SAR image.Therefore,it can preserve the detailed information and make the changed results convincible.(3)The decomposition based on three-component is usually conducted under the hypothesis of reflection symmetry.However,the cross terms of covariance matrices of some scattering units may be zero under such hypothesis and this situation is inconsistent with the reality.In order to solve the problem,a novel change detection method based on four-component decomposition is proposed.Helix scattering power is added as the fourth component to the three-component scattering model which describes surface,double-bounce,and volume scattering.The final changed result is acquired according to the steps discussed in section(2).Experimental results on real PolSAR data demonstrate that the introduction of helix scattering mechanism can suppress the disturbance of man-made targets effectively and discard the circumstance of negative power of some scattering units,making the change detection result much subtle.