Research On Ship Detection Techniques Of Compact Polarimetric Synthetic Aperture Radar

Ship is the main carrier of marine activities,its detection has been widely concerned by coastal countries around the world for a long time,and it is also an inevitable demand for China to protect the energy lifeline of the marine economy and monitor marine pollution.Under the current global all-day and all-weather satellite observation systems,compact polarization synthetic aperture radar(SAR)has shown great application potential in ocean target observation,and is one of the main trends in the development of spaceborne SAR systems in the future due to its system design complexity,system maintenance,and data acquisition costs,which are more advantageous than traditional full polarization.Moreover,the observation coverage is doubled in theory and almost does not lose target information.However,due to the complex modulation and scattering mechanisms of the marine environment and targets,the accurate recognition of sea clutter,the mining of physical representations of targets under compact polarization systems,and effective detection are extremely challenging,and have received widespread attention from the academic and industrial communities in the SAR field.Focusing on the practical engineering application requirements of high target detection rate and low false alarm rate,this thesis has conducted comprehensive and in-depth research on the above issues.The main research results include:(1)In order to solve the problem of high-precision statistical modeling of sea clutter,a new model is proposed.The logarithmic cumulant estimator of the model is derived in theory,and the uniqueness of the parameter estimation solution is mathematically proved.The flexible and high-precision description of the statistical characteristics of uniform and non-uniform/rough sea clutter in SAR images under different wind speeds is achieved.The results of measured data show that the new model takes into account both fitting accuracy and efficiency,and has more advantages in practical applications compared to current advanced K+R and GK models.(2)Aiming at the problem of sea clutter universal statistical modeling,the applicable conditions of the new model parameter estimator are theoretically derived,and the reason why the new model cannot be used for statistical modeling of all local areas in SAR scenes is explained.After pointing out the existing deficiency of the generalized Gamma distribution(GГD),we propose a joint statistical modeling method for sea clutter based on the new model and GГD.Both theoretical and experimental results reveal that this method can reliably and robustly describe the statistical behavior of any local SAR region data,and is a universal statistical modeling method for SAR data.(3)Aiming at the problem of mining target polarization scattering layer representation,a new method named the power entropy decomposition for the compact polarization SAR eigenvalue decomposition is proposed.The resulting high entropy scattering amplitude(HESA)can be seen as a combination of wave polarization and scattering power characteristics,which can reveal the differences between ships and the background from the perspective of electromagnetic scattering,such as scattering power and electromagnetic wave turbulence.We mathematically prove that HESA is an effective detection quantity for improving signal-to-clutter ratio or ship-to-sea contrast,and pointed out that the former joint-statistical-modeling is also suitable for describing the statistical behavior of the detected quantity,thus developing a constant false alarm rate(CFAR)adaptive detection method based on the detected quantity.(4)Aiming at the problem of target visual layer representation mining,a new version of notch filter suitable for compact polarization SAR is proposed by defining appropriate partial vectors from the covariance matrix of compact polarization SAR data,forming a new representation of compact polarization SAR notch distance.Based on the aforementioned joint-statistical-modeling,the statistical models of compact polarization SAR notch distance is further theoretically derived.At the same time,the CFAR threshold is mathematically derived based on statistical models,and the implementation of CFAR of a notch filter for ship detection in compact polarization SAR is presented.The performance of target detection between compact polarization SAR and conventional dual polarization SAR is compared,providing a new perspective on the advantages of compact polarization SAR over traditional dual polarization SAR in ship detection.(5)Aiming at the integration problem of target compact polarization SAR multi-dimensional representations,using the feature vector formed by the representation quantity HESA and the notch distance as the input of the integration,based on the Retina Net basic network and the channel-spatial attention mechanism,a method of target compact polarization representation integration based on deep learning is proposed.The method uses the compact polarization representation quantity to drive the deep learning network,which can effectively capture the portion between HESA and notch distance that contributes the most to ship targets,and achieve effective integration between multi-dimensional representations,and greatly improve detection performance.