Research On The Methods Of Ship Detection In Measured SAR Image

The research of ship target detection technology in SAR image is of great significance to China's marine management and safety.In this paper,the ship target detection algorithm in SAR image is studied and the deep learning convolution neural network Faster RCNN algorithm is introduced to deal with the problem that the current traditional CFAR algorithm can not adapt to the large amount of SAR image data.The main work and innovation of this paper are as follows.The background sea clutter modeling is studied based on SAR images.The common parameter models of sea clutter distribution are discussed and the corresponding parameter estimation methods are introduced.Then,the method of sea clutter modeling is given and the sea clutter statistical modeling experiments are carried out for eight different types of measured SAR images.Normal distribution,rayleigh distribution,lognormal distribution and weibull distribution are used to fit the histogram.In addition,k-s test and chi-square test are used to evaluate the fitting degree of each distribution.CFAR detection algorithm is discussed in depth.Firstly,the essence of CFAR algorithm is explained.And five kinds of detectors are given.In addition,their respective structures and application background are introduced.Then the actual detection experiments of ship target in SAR image are carried out.The CA-CFAR methods based on rayleigh distribution,lognormal distribution and weibull distribution are adopted.Local two-parameter method and OS-CFAR methods based on normal distribution and rayleigh distribution are also used.At the same time,the quality factor is used to evaluate the results of each algorithm.The time of each algorithm is also counted and the results are compared.The Faster RCNN detection algorithm is deeply studied.Firstly,some basic theories of deep learning are introduced.Then the evolution process and corresponding principles of Faster RCNN algorithm are described.Finally,the actual detection experiments are carried out.Before the experiments,a SAR image data set containing a large number of ship targets is created.The samples in this data set are all from the Sentinel-1A satellite.The corresponding creation process is also given.Then Faster RCNN detection experiments are carried out based on the created data set.The detection effect of Faster RCNN in different training batches and dropout is discussed.Meanwhile,the average precision is used to evaluate the detection effect.Then based on three kinds of vehicle targets in MSTAR SAR data set,the effect of Faster RCNN classification detection is studied.Finally based on different sizes of real SAR images,Faster RCNN and CFAR are tested and compared.The quality factor is used to evaluate the performance and the running time of each algorithm is counted.Finally,both Faster RCNN and CFAR have successfully detected ship targets,but Faster RCNN algorithm is much faster than CFAR algorithm in time consumption and it has good real-time performance.The experimental results verify the effectiveness of Faster RCNN algorithm which has a good application prospect in the field of ship target detection in SAR images.