Research On Few-shot SAR Auto Target Recognition Via Data Of Other Sources

Synthetic Aperture Radar（SAR）has the characteristics of all-day and all-weather operation,which can obtain high-resolution microwave images,and plays an important role in civil and military fields.SAR target recognition technology realizes the discrimination of target attributes and categories by extracting target features,which is a key link in SAR image interpretation and intelligence conversion.However,in practical applications,it is often impossible to collect abundant samples for difficult-to-obtain military SAR targets,which restricts the application of SAR target recognition technology.Therefore,few shot SAR auto target recognition,and capturing auxiliary recognition information from its source data are of great significance to improve the ability of SAR target recognition.Focusing on the above problems,this dissertation has developed method research and simulation verification for few-shot SAR target recognition.The main contents are as follows:1.Aiming at the problem of the lack of SAR target samples,the SAR sample augmentation methods based on image processing and sub-band decomposition are studied.Next,by resorting to the simulated SAR target data,a source-assisted SAR sample augmentation method based on generative adversarial network is proposed,which enriches the feature information of few-shot SAR target.The above methods provide data support for the improvement of SAR target recognition capabilities.2.Aiming at the problem of slow convergence speed and limited recognition ability of recognition model while information from other data source with similar distribution is migrated,a few-shot SAR target recognition method based on weight migration and feature embedding is proposed.By establishing the matching relationship between the feature and the classifier,the initial weight of the SAR image domain classifier is directly obtained,and the initial point and optimization direction of the recognition model training are optimized,finally,the recognition rate of the transfer learning few-sample SAR target is improved.3.Aiming at the problem of heterogeneous features and difficulty in using auxiliary information when information from other data source with different distribution is migrated,a few-shot SAR target recognition method based on feature mapping in the difference domain is proposed.By learning the feature transformation between optical samples and SAR samples,the heterologous feature mapping relationship is derived,which realizes the homogeneity of the classification features of different domains in the heterologous data-assisted few-shot recognition task,and effectively improves the performance of the heterologous-assisted few-shot SAR target recognition.The above methods have been verified by the actual measurement data,and the results show that the above methods can make full use of other source information to assist SAR target recognition,and improve the accuracy of SAR target recognition under the condition of limited samples.