摘要
针对传统雷达信号识别方法无法有效进行识别类型扩展问题,该文提出一种基于知识蒸馏与注意力图的雷达信号识别方法。首先将雷达信号的平滑伪Wigner-Ville分布(SPWVD)作为输入;然后设计了基于残差网络的增量学习网络结构,利用基于知识蒸馏与注意力图的损失函数,缓解类别增量过程中的灾难性遗忘;最后采用基于样本特征均值距离的方法对数据集进行管理,有效降低存储资源占用空间。实验表明,该方法能在存储资源有限的情况下,对扩展分类的信号快速完成训练,且对原有分类和扩展分类信号均有良好的识别准确率。
In order to solve the problem that traditional radar signal recognition methods can not effectively expand the recognition types,a radar signal recognition method based on knowledge distillation and attention map is proposed.Firstly,the Smooth Pseudo Wigner-Ville Distribution(SPWVD)of the radar signal is used as input;Then,the incremental learning network structure based on residual network is designed,and the loss function based on knowledge distillation and attention map is used to alleviate the catastrophic forgetting in the process of category increment;Finally,a method based on the mean distance of sample features is used to manage the data set,which reduces effectively the occupied storage resources.Experiments show that this method can quickly complete the training of the extended classification signal under the condition of limited storage resources,and has good recognition accuracy for the original classification and the extended classification signal.
作者
曲志昱
李根
邓志安
QU Zhiyu;LI Gen;DENG Zhian(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Key Laboratory of Advanced Marine Communication and Information Technology,Ministry of Industry and Information Technology,Harbin 150001,China)
出处
《电子与信息学报》
EI
CSCD
北大核心
2022年第9期3170-3177,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金(61801143,61971155)。
关键词
雷达信号识别
增量学习
知识蒸馏
注意力图
灾难性遗忘
Radar signal recognition
Incremental learning
Knowledge distillation
Attention map
Catastrophic forgetting
