| SAR images are synthetic aperture radar images,which play an important role in military target reconnaissance and weather detection.Therefore,SAR image recognition has always been a research hotspot in the field of SAR image research.However,SAR image recognition requires a large number of SAR images as data support,and the existing SAR image data has been repeatedly used for research.For the generality of the new SAR image recognition method,more new SAR image data is needed.Therefore,this article first studies the SAR image generation method.The traditional SAR image generation method is based on feature extraction.First,edge detection is performed on the SAR image,then the SAR image is segmented based on the two-dimensional entropy threshold segmentation algorithm,and feature extraction is performed on the segmented image.Finally,the RD algorithm is used.Perform SAR image reconstruction.However,compared with the real SAR image data,the SAR images generated based on feature extraction have poor visual effects,not realistic enough,and the features are relatively single,which cannot meet the diversity.In view of this,this article discusses the generation of SAR images based on GAN derived models.First introduced the framework and network structure of several derivative models of the GAN model,and then used the MSTAR data set to train two of them to obtain the generated SAR image.According to the generated images,it shows that GAN-based SAR images have better visual effects,more lifelike,and diversity.Then,in the aspect of SAR image recognition,three methods of ship target detection in SAR image are proposed.The first method is a detection method based on GAN and HMRF.First,the GAN model is trained with the image to be detected to obtain the corresponding generated image,and then the target is generated by GAN and the target is fused with the generated SAR image to obtain a new image to be detected The purpose of fusion of false targets is to reduce the false alarm rate,and then preprocess the image with CFAR,and finally use the HMRF model for final detection and segmentation of the generated target,and then obtain the final detection image.The other two methods are both ship target detection methods based on information geometry.The first one uses CFAR to obtain the initial distribution parameters of the image to be detected,then constructs Gauss Fisher metric parameters,and then constructs the energy function of the MRF model,which can transform the detection problem into a maximum posterior probability decision criterion problem.Another method first uses the CFAR detector based on the Weber distribution for preprocessing,and then constructs the metric tensor of the Weber distribution manifold for iterative optimization.The experimental results show that the three methods can achieve good detection results and are not sensitive to sea clutter. |
