摘要
针对目前复合干扰识别中传统特征识别方法泛化性低和深度学习方法需要大量训练样本的问题,提出了一种融合了高效通道注意力机制的密集卷积网络模型(Efficient Channel Attention-Dense Convolutional Network,ECA-DenseNet),并结合与模型无关元学习(Model-Agnostic Meta-Learning,MAML)算法改进网络模型的初始化参数,以此增强模型在少量样本条件下的识别性能。实验结果表明,文中所提算法在各干噪比下对复合干扰平均识别率达到了98.73%,当样本数变为原来的四分之一时,复合干扰平均识别率依然能够达到95.35%,验证了算法的有效性。
Aiming at the problems of low generalization of traditional feature recognition methods and a large number of training samples required by deep learning in current composite jamming recognition,a composite jamming recognition model ECA-DenseNet,and the model-agnostic meta-learning algorithm(MAML)is used to improve the initialization parameters of the network model,and to enhance the recognition performance of the model under a small sample size.Experimental results show that the average recognition rate of the proposed algorithm reaches 98.73%at different signal-to-interference plus noise ratios.When the number of samples is a quarter of the original,the average recognition rate of the composite jamming can still reach 95.35%,which verifies the effectiveness of the algorithm.
作者
康文威
孙闽红
KANG Wen-wei;SUN Min-hong(School of Communication Engineering,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)
出处
《杭州电子科技大学学报（自然科学版）》
2023年第6期59-65,共7页
Journal of Hangzhou Dianzi University：Natural Sciences
基金
国家自然科学基金(61901149)。
