| 摘要: |
| 针对信号辐射源个体识别小样本难以稳定收敛、识别准确率不足的问题，提出了一种基于优化孪生网络模型进行小样本辐射源个体识别的方法，分析了通过孪生网络实现不同类别样本对特征向量距离增大、相同类别样本对特征向量距离减小的弹簧模型，达到小样本训练损失函数的快速收敛的目的，并结合交叉熵实现损失函数优化，从而提升了小样本个体识别的准确率和稳定性。试验结果表明，针对每类不大于10个训练样本集的通信电台所提方法能够达到88%以上个体识别准确率。 |
| 关键词: 小样本 个体识别 孪生网络 损失函数优化 Resnet网络 |
| DOI： |
|
| 基金项目: |
|
| An emitter individual identification method for small samples based on optimized siamese networks |
| LIANG Xianming |
| (Southwest China Institute of Electronic Technology，Chengdu 610036，China) |
| Abstract: |
| For the problems that the signal emitter individual identification of small samples is difficult to stably converged and the recognition accuracy is low,a network model based on optimization of siamese networks for small sample emitter individual identification is proposed.The siamese networks similar to the Spring Model are analyzed,which can increase the distance of feature vectors of dissimilar paris,decrease the distance of feature vectors of similar paris,and realize the fast convergence with small sample training loss.Then cross entropy is used to optimize the loss function to improve the accuracy and stability of small sample identification.The experimental result shows that the individual recognition accuracy of the proposed method can reach 88% for each communication station with no more than 10 training samples. |
| Key words: small samples individual identification siamese networks optimized loss fuctio resnet network |
