SAR-LAM:A Lightweight Adaptation Method Being Geared to Few-Shot SAR Target Recognition
SHI Songhao
WANG Xiaodan
Abstract：In view of the issue of model performance degradation caused by cross-domain transfer in few-shot learning,a lightweight adaptation strategy for few-shot SAR target recognition named SAR-LAM is proposed.This method is to utilize knowledge distillation for pre-training a generalized encoder and em-bedding an adaptation module trained only with very few target domain samples.The extracted features are then mapped into a more discriminative space,and finally,the query set samples are classified by tak-ing a prototypical network as the baseline.This adaptation strategy is to increase at a few cost in learning parameters,and by so doing,the limitations of model transfer caused by data distribution differences is overcome,improving the model's ability to extract features in the target domain,and simultaneously im-proving the accuracy of SAR target recognition by at least 1.93 percentage points under few-shot condi-tions.And this adaptation strategy is superior in performance to the other methods.
Keywords：SAR target recognitioncross-domain few-shot learninglightweight
Publication Date：2024-06-25
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：9( 103-111 )