Study on Small Samples SAR Image Recognition Detection Method Based on Transfer CNN

Synchronous Aperture Radar (SAR) images can be acquired by different microwave imaging modes, systems, etc. For the recognition detection of SAR images of heterogeneous small samples(ie a small number of samples),this paper proposes a transfer-based Convolutional Neural Network (CNN) small samples SAR images ground object recognition detection method.First of all,the source domain sufficient samples were used to pre-train CNN,and then CNN was fine-tuned by using the transfer of small samples of the target domain SAR images to obtain a new CNN.The network finally realizes ground object recognition detection of small samples SAR images.The effectiveness and accuracy of the proposed method are verified by experiments,and good results can be obtained.