Radar Target Recognition Based On Transfer Learning

RATR is an interdisciplinary field of radar technology and pattern recognition technology.It plays an important role in improving national defense capability.With the improvement of radar resolution and wideband radar technology,radar echo can provide more reliable target characteristics,HRRP,as a large category of wideband high-resolution radar imaging technology,has become a hot spot in recent years.The deep learning performs well in target recognition.In order to explore the performance of deep learning in HRRP target recognition,this paper uses the traditional two-dimensional CNN and one-dimensional CNN model for the specific data structure of HRRP.Batch normalization and Dropout methods are used in model.The result shows that the one-dimensional CNN has faster convergence and higher recognition rate,which can reach 99.8%.However,when the sea clutter affects the data,the original data distribution is affected,which greatly reduces the performance of the model.In order to improve the generalization ability of the target recognition model based on CNN and enhance the recognition ability under clutter,this paper introduces the deep adaptation networks(DAN)method into the recognition of HRRP by adding adaptive layers.In order to further improve the performance of DAN method,this paper innovatively proposes the hybrid kernel MMD instead of MK-MMD in original DAN,and designs hybrid kernel MMD loss.In the training,HRRP data with Rayleigh sea clutter is used as the target domain to help the model to learn the inherent laws of the clutter data.The experimental results show that,this method can improve the recognition rate of target domain data by about 15%under the influence of sea clutter compared with the conventional transfer learning method and DAN method,and significantly improve the generalization performance and robustness of the model.In order to explore the performance of other transfer learning methods,this paper also introduces residual migration network(RTN)into HRRP target recognition for the first time.RTN is proposed based on the residual network,and entropy loss is added to the model according to the specific structure of the target domain data.In this paper,joint distribution adaptive is added to the RTN model innovatively.The experimental results show that the RTN and the hybrid nuclear DAN method has similar effects.