Man-made Targets Extraction From PolSAR Data Using Time-frequency Analysis

In this thesis, “Man-made target extraction methods from Pol SAR imagery”are investigated. Based on Time-Frequency(TF) analysis technology, the scattering characteristics of targets and background clutters are exploited to enhance targets and suppress clutters. Combined with other effective polarimetric features, some unsupervised target extraction approaches are established. With respect to building and ship extraction within different kinds of complex scenes, experiments prove that the proposed methods are effective, efficient, robust and suitable to engineering applications. The main work includes the following aspects:1. Complex electromagnetic wave may influence SAR imaging result. When the interference power is stronger than that of target’s signal, SAR images will be very noisy and cannot be processed. Concerning this kind of strong interference, an interference suppression flowchart is proposed and realized using a special interfered sea-ice data.Another kind of interference is not so strong, but maybe damage the phase information seriously, and will do harm to TF analysis and polarimetric feature extraction. Therefore,an interference filtering scheme, which is of benefit to TF analysis, is set up for this kind of phase-damaged data.2. Based on TF stationary analysis, a TF stationary indicator is proposed to characterize man-made targets and natural background clutters. Two types of environments are investigated. Firstly, using complex urban single-Pol and full-Pol SAR data, building are extracted due to their non-stationary feature; Then, for the complex polar sea-ice scenery,the unique scattering characteristics sea-ice are analyzed. By using the TF stationary indicator, ship can be discriminated from sea-ice background.3. Based on TF coherence analysis, a coherence measurement indicator is proposed.In particular, a new TF polarimetric characteristic descriptor is derived to describe the extracted targets. Three classes of difficult environments are analyzed. For the complex urban scenario, the single-pol and full-pol coherence indicators can extract the buildings effectively. The new TF polarimetric descriptor may characterize the detected buildings in detail. Besides that, the coherent scatterer extraction results from Pol SAR data in different bands, are compared and analyzed. For the serious SAR artifact environment, the proposed TF coherence measurement method can be used to filter artifacts effectively and detect ships without false alarms that cased by ghost artifacts. The new TF polarimetric descriptor is able to portray the detected ships correctly. For the difficult sea-ice area, our approach can be used to extract, detect, discriminate ships from confused sea-ice. The analysis result with the new TF polarimetric descriptor conforms to an icebreaker ship’s real structure.4. Based on multiple effective features, some feature fusion strategies are constructed for target extraction. Combined the TF stationary indicator with coherent indicator, the relationship of these two indicators is analyzed qualitatively. A scheme with the fusion features is used to extract buildings and is confirmed to be effective. A fusion scheme with joint use of TF features and Pol In SAR coherence feature, is proposed for land-cover classification in urban area. To take advantage of other polarimetric features for urban area detection from low resolution space-borne Pol SAR imagey, TF indicators and polarimetric reflection symmetry characteristic are mixed together. This strategy is proved to be effective. TF stationarity detection approach with polarimetric G0 distribution is proposed to extract ortho-oriented buildings and the circular polarization correlation coefficient is optimal in characterizing slant-oriented buildings. A new fusion scheme is proposed based on TF stationary and circular-pol features, and works well with respect to building extraction.In general, these TF analysis methods and characteristic descriptors are reliable for many complex scenes and different sensors’ data. It is confirmed that our approaches perform efficiently and can be applied in civil or military affairs currently.