Research On The Algorithm Of Ship Target Detection In SAR Image

Ship target detection in SAR image has been widely applied in the fields of militarytarget tracking, surveillance of the battlefield environment and remote strategy warning,which makes it a critical factor in winning modern hi-tech wars. Thus, the research on shiptarget detection in SAR image is of great significance. It has been the primary task in militaryfields and attracted great attention all over the world. Two aspects are mainly concerned inthis paper: sea clutter modeling and ship target detection algorithm.Firstly, statistical characteristic of sea clutter is analyzed, frequently-used models of seaclutter are discussed and the use of distribution models in sea clutter modeling is given. Andthen, the relationship of threshold and false alarm probability is qualitatively studied. Thispaper has developed the research on selecting model, parameter estimation and evaluation ofmodel’s performance and the experiment results are displayed.Secondly, the theory and process of ship detection based on CFAR are reviewed, and thestructure and performance of several CFAR detectors are summarized. The principle andprocedure of global and local CFAR algorithms used in ship target detection are given.Experiment is established based on these algorithms under different distribution models.The existing CFAR algorithm lacks the ability of taking both global and localinformation into consideration simultaneously. To solve this problem, an improved CFARalgorithm is introduced, in which local threshold is adjusted according to the global thresholdand gray-level of SAR images is reconstructed depending on the detected units’ neighbors.Experiment verifies the introduced algorithm improved the performance of ship targetdetection and could balance the local and global information of the image.At last, this paper integrate VI-CFAR detector into distribution model with a widedynamic range, which makes the algorithm adapt to a more extensive range of sea clutter andcan select detector adaptively. Combined with the work above, improved CFAR algorithmbased on Weibull and K distributions are proposed and the results confirmed its effectiveness.