| Compared with other types of imaging equipment,the Synthetic Aperture Radar(SAR)imaging technology is not affected by distance,light,weather and other factors.SAR is with all day,all-weather,high-resolution imaging characteristics.At present,SAR imaging technology has been widely used in various fields of military detection and civil mapping.However,SAR images are always with a large of complex speckle noise,due to the coherent imaging principle of SAR image.On the other hand,because of the lack of knowledge about SAR radiation characteristics and imaging principles,collecting effective features to target recognition is very difficult.Therefore,the development of SAR ATR technology is constrained seriously.To against this challenge,this paper is learning a discriminative dictionary to sparse coding the sample,using the coding coefficient for SAR target recognition,instead of collection features to recognition.The results of experiments show the feasibility of this kind of methods.The achievements of this thesis are listed as following several points:(1)To improve the quality of SAR image,this thesis studies the relevant pre-processing works,including the noise suppression,target detection and so on;(2)In this thesis,we modify the object function of Fisher Discrimination Dictionary Learning(FDDL)to improve the train database coefficient's sparseness,which leads to better performance in SAR target recognition.(3)The new method is compared with the classical sparse classification methods,such as SRC,K-SVD,D-KSVD,FDDL,and the superiority of the new method is summarized.(4)By designing experiments on different size training database,the effect of the size of the dictionary on the recognition is analyzed,the recognition rate increases with the growing size of dictionary.Futhermore,the experimental results verify that the recognition rate of the sub-class which have less samples will be greatly reduced,when the quantity of samples in each sub-class have great difference. |
PDF Full Text Request