| The traditional ship detection methods represented by CFAR(Constant False Alarm Rate)in single-channel synthetic aperture radar(SAR)imagery are mainly designed for detected products,i.e.,amplitude/intensity data.Common practices under the assumption of fully developed speckle consider that phase content in single-channel SAR imagery is meaningless and thus,relevant complex-valued information in current high-resolution SAR imagery is discarded.However,amplitude/intensity information that can be used is usually very simple and sensitive to radar frequency,polarization,incident angle,interference,and other factors.Thus,there are some performance bottlenecks that are difficult to overcome for traditional methods,e.g.,they can cause false alarms and misdetections easily in the complex background.This dissertation considers that the phase information in single-channel high-resolution SAR imagery is helpful to distinguish ship from clutter.Thus,the complex-valued information in single-channel SAR imagery can improve ship detection performance.A complex-valued image includes two real-valued images(amplitude and phase images or real and imaginary parts)and also the algebraic and geometric analysis between them.Thus,this dissertation first studies the amplitude/intensity information,and then the phase information,finally focuses on the complex-valued information describing the relationship between real and imaginary parts.In particular,this dissertation first studies the fundamental problems such as the extraction and characterization of complex-valued information and then its application to ship detection in single-channel SAR imagery.The results are validated by a variety of real and simulation data.Specifically,the main contribution and novelty of this dissertation are as follows:1.It comprehensively reviews the current ship detection methods in SAR imageryShip detection in single-channel,polarimetric,interferometric SAR imagery and azimuth ambiguity removal methods are studied in detail,which illustrates the development of the context in this field.Meanwhile,focusing on the ship detection in single-channel SAR imagery,both the advantages and disadvantages of the existing methods are analyzed and summarized,followed by the proposal of challenges and opportunities.Finally,the key issues and main research line of this dissertation are put forward.2.It studies in detail the fundamental issues such as the extraction and characterization of complex-valued information in single-channel SAR imagery1)Extractration and characterization of complex-valued information based on statistical modeling in single-channel SAR imagery.This dissertation first introduces the significant concepts of complex-valued random variables.In addition to introducing the traditional amplitude statistical modeling method,it focuses on the phase statistical modeling method.Because the phase image is too abstract and cannot distinguish the target and background,a phase statistical modeling method based on the neighborhood phase difference feature is proposed.It can describe the phase difference of ship and clutter,indicating the feasibility of complex-valued information to improve ship detection performance.It also studies the complex-valued SAR imagery modeling method based on complex generalized Gaussian distribution(CGGD).To improve the low efficiency of the classical method for estimating the shape parameter of CGGD,a near real-time fast estimation method based on high-order moments is proposed.The results have good estimation precision and can characterize the non-Gaussianity of complex-valued data.2)Extractration and characterization of complex-valued information based on noncircularity decomposition in single-channel SAR imagery.Noncircularity is a basic concept for distinguishing complex-valued signal processing from the traditional real one.To explore the concept and application of noncircularity,this dissertation proposes a noncircular decomposition,metric,and generation method.It enriches the noncircularity theory: firstly,the noncircularity decomposition equation clarifies the composition of noncircularity;secondly,the noncircular metric can better measure the difference between different types of noncircularity;thirdly,the noncircularity generation method can simulate any level of noncircularity.On this basis,theoretical analysis and experiments verify the noncircularity difference between ship and sea clutter.This demonstrates the extra value of using complex-valued information in further.3.It studies in detail the application of complex-valued information to ship detection in single-channel SAR imagery1)Ship discrimination from radio frequency interference(RFI)based on complex-valued information in single-channel SAR imagery.RFI is one of the SAR image quality problems that affect ship detection severely.This dissertation studies in detail the generation mechanism and its effects on SAR imagery.To reject false alarms caused by RFI,the author found two characteristics of RFI for the first time,i.e.,its noncircularity is very weak(or considered as circular),and it mainly exhibits sub-Gaussianity.Based on these new findings,a discrimination method based on noncircularity and non-Gaussianity to distinguish ship and RFI is proposed.The results verify RFI as one of the two predicted special noncircular cases: although its brightness is similar to that of ship,their noncircularity and non-Gaussianity are quite different.Meanwhile,the proposed discrimination method can reject RFI in single-channel SAR imagery effectively.2)Ship detection based on complex-valued information in single-channel SAR imagery.To take benefits of complex-valued information and overcome the bottlenecks of traditional detection methods,this dissertation introduces high-order complex-valued information such as complex signal kurtosis(CSK),which can take advantage of both non-Gaussianity and noncircularity.This dissertation points out that CSK is a vital indicator of ocean monitoring and SAR imaging quality,which can be less affected by the radar frequency,polarization,incidence angle,and so on.The author proposes a ship detection method based on CSK in single-channel complex-valued SAR imagery.It consists of three steps: firstly,the fast region proposal can detect potential ship locations in a fast way;secondly,the iterative segmentation can segment and identify ship robustly;thirdly,the feature enhancement can improve the image quality of ship.Experimental results show that it can achieve better performance in the complex background. |
